Look at this $ARM . Still going and up another 4.5%. https://t.co/RZx8vAQswi
What $20bn of $ARM CPU revenue out of thin air looks like. https://t.co/qOu6zMecPg
And now $ARM, a $300B+ company, hit triple digit returns in a short timeframe since $135… Cool name CPUs go brrrr? https://t.co/67Zz6EWfnL
$ARM — The Silent Monopolist
$ARM all-time high +8%. Jensen is getting to work selling $NVDA Vera CPUs. LFG. https://t.co/DLjKV5BToo
$ARM — The Silent Monopolist This is the most underappreciated play in the entire theme. ARM’s market share of CPU compute now represents ~50% among top hyperscalers. Data centers are expected to require more than 4x current CPU capacity per gigawatt as agentic AI scales — creating a market opportunity of more than $100 billion by 2030. ARM launched the Arm AGI CPU — its first production silicon purpose-built for agentic AI. It delivers more than 2x performance per rack vs. x86-based platforms, reducing AI data center CapEx by up to $10 billion per gigawatt. Meta is the lead partner and co-developer. Customer demand for ARM AGI CPUs already exceeded $2 billion across FY2027–FY2028 shortly after launch. ARM wins whether Intel wins, AMD wins, or custom silicon wins — because they license the architecture underneath all of it. It’s the IP royalty on the entire CPU renaissance.
$NVDA $ARM Incredible forecasted Vera CPU launch. NVDA believes they will become the world’s leading CPU supplier. https://t.co/oOCCa2W3On
$ARM Going off this morning. All time high. I’m long 1/21/28 calls. https://t.co/jOKCjp1peh
10 AI Infrastructure Stocks I’m Watching for the Long Game 1. $MU — Memory Is Now a Strategic Asset Micron posted record Q2 FY2026 revenue of $23.86B — nearly 3x the same quarter last year — with record gross margins, EPS, and free cash flow.  HBM capacity is completely sold out through calendar 2026, with pricing locked in on the vast majority of that volume.  Memory is no longer cyclical commodity — it’s the bottleneck of the AI era. 2. $MRVL — The Custom Silicon Kingmaker Marvell delivered record FY2026 revenue of $8.195B, up 42% YoY, with custom AI chip wins at Amazon, Microsoft, and Google anchoring its growth.  Its 800G and 1.6T optical DSPs have become the essential “plumbing” for hyperscale AI clusters.  The most underrated AI infrastructure name hiding in plain sight. 3. $CRDO — The High-Speed Connectivity Pure Play Management is guiding over 50% YoY revenue growth for fiscal 2027 as AI infrastructure scales rapidly.  Rothschild & Co Redburn just launched coverage with a Buy and a $206 price target tied to surging generative AI infrastructure spending.  Ultra-low power, ultra-high speed — exactly what AI data centers need. 4. $AAOI — America’s 800G Transceiver Champion AOI completed its first volume shipment of 800G products to a large hyperscale customer in Q1 2026 and is guiding for sequential revenue growth throughout the year, with significantly larger growth expected in Q3 as additional capacity comes online.  They’re building a 210,000 sq ft manufacturing facility in Sugar Land, TX with up to $300M in planned investment to become one of the largest domestic suppliers of AI datacenter transceivers.  5. $AXTI — The Indium Phosphide Supercycle AXT hit Q1 2026 revenue of $26.9M, up 39% YoY, with InP backlog exceeding $100M, and is doubling indium phosphide capacity in both 2026 and 2027.  Order demand for scale-out optics could grow 2x in 2026 and another 2x in 2027  — and every optical transceiver in every AI data center needs InP. This is the upstream pick-and-shovel few are talking about. 6. $INTC — The Turnaround Nobody Wants to Believe In Intel is partnering with Nokia and Dell to advance next-gen 5G edge infrastructure, with its Xeon 6 Granite Rapids-D SoC delivering enhanced AI capabilities for far-edge deployments.  Foundry ambitions, US manufacturing tailwinds, and a deeply reset valuation. High risk — but if the turnaround lands, the upside is asymmetric. 7. $AMD — The Challenger That Keeps Gaining Ground The iShares Semiconductor ETF is up 77% this year, with AMD among the key beneficiaries of AI infrastructure investment.  MI300X traction in inference, custom silicon partnerships, and a data center GPU roadmap that has Nvidia watching closely. The market keeps underestimating this one. 8. $QCOM — Beyond Phones, Into Physical AI Qualcomm posted $10.6B in Q2 FY2026 revenue with record quarterly QCT automotive revenues, and combined automotive + IoT revenues up 20% YoY.  Momentum across personal, industrial and physical AI is growing  — this is no longer just a smartphone chipmaker. Edge AI, autonomous systems, and robotics are the next chapters. 9. $NOK — The AI-Native Network Play Nokia is delivering advanced optical and IP data center connectivity to power AI computing across continents, while partnering with NVIDIA to define the next generation of global connectivity in the AI-native wireless era.  Deeply undervalued relative to its AI infrastructure footprint. Patient capital play. 10. $ARM — The Architecture Running AI Everywhere In March 2026, Arm made history by launching its first-ever production silicon — the Arm AGI CPU — a data center processor for agentic AI workloads, developed with Meta, delivering more than 2x performance per rack vs. x86 platforms.  Arm’s compute platform now supports AI workloads ranging from milliwatts to gigawatts — from edge to cloud.  Every AI chip runs on Arm’s architecture. The royalty engine of the AI age.
THE CPU RENAISSANCE THESIS The Structural Shift Nobody Priced In This isn’t cyclical. It’s architectural. The AI compute stack is undergoing a fundamental transition. Training workloads use 7–8 GPUs per CPU. Inference tightens that to 3–4 GPUs per CPU. And in agentic/multi-agent AI, the ratio could compress to 1:1 — or even flip in favor of CPUs.  Tool processing on CPUs can account for up to 90.6% of total latency in agentic workflows  — meaning GPU throughput improvements alone don’t solve the bottleneck. You need more CPUs. $AMD — The Real Winner Here AMD’s EPYC CPUs now command 46.2% of total server CPU spending in Q1 2026 — an all-time high — driven by rising AI and data center demand.  AMD CEO Lisa Su doubled the server CPU TAM forecast: “We now expect the server CPU TAM to grow at greater than 35% annually, reaching over $120 billion by 2030.” AMD expects server CPU revenue to grow more than 70% YoY in Q2 2026.  Why $AMD over $INTC long-term? Intel has supply constraints and manufacturing yield challenges. AMD is fabbed at TSMC — cleaner execution, faster ramp. Intel’s planned 2026 launches face in-house process yield issues that may accelerate AMD’s market share gains further.  $ARM — The Silent Monopolist This is the most underappreciated play in the entire theme. ARM’s market share of CPU compute now represents ~50% among top hyperscalers. Data centers are expected to require more than 4x current CPU capacity per gigawatt as agentic AI scales — creating a market opportunity of more than $100 billion by 2030.  ARM launched the Arm AGI CPU — its first production silicon purpose-built for agentic AI. It delivers more than 2x performance per rack vs. x86-based platforms, reducing AI data center CapEx by up to $10 billion per gigawatt. Meta is the lead partner and co-developer.  Customer demand for ARM AGI CPUs already exceeded $2 billion across FY2027–FY2028 shortly after launch.  ARM wins whether Intel wins, AMD wins, or custom silicon wins — because they license the architecture underneath all of it. It’s the IP royalty on the entire CPU renaissance. $INTC — Turnaround Optionality Intel CEO Lip-Bu Tan: “For the last few years, the story around high-performance computing was almost exclusively about GPU and other accelerators. In recent months, we have seen clear signs that the CPU is reinserting itself as the indispensable foundation of the AI era.”  $INTC is higher risk — manufacturing execution is the key variable. But if 18A yields hit mid-year targets and Xeon 6+ ramps, this is a multi-year re-rating story from deeply depressed levels. $NVDA — Still Wins, But Differently Beyond traditional server CPU vendors Intel and AMD, NVIDIA is now entering the server CPU market  via Arm-based Grace architecture. NVDA is vertically integrating — building CPU+GPU stacks together. Grace Blackwell is the proof of concept. Long-term this is bullish for NVDA but compresses their GPU-only moat. Physical AI / Edge Sleepers $ARM (again) + $QCOM for Physical AI Edge and physical AI systems are dominated by control logic, safety monitoring, and system coordination — all CPU-led roles. As AI moves from data centers into vehicles, robots, and machines operating under strict power and safety constraints, CPUs play the leading role.  Qualcomm’s Snapdragon platform is the dominant edge AI compute stack for automotive and mobile inference. As physical AI scales — humanoid robots, autonomous vehicles, smart infrastructure — $QCOM’s edge CPU/NPU stack becomes mission-critical. $AMBA (Ambarella) — The Dark Horse Ambarella posted a revenue record of $390.7M in FY2026, up 37% YoY, with ~80% of revenue attributed to edge AI applications. They’ve shipped 45 million edge AI SoCs.  Small cap, high-growth, pure-play edge AI inference silicon. Under the radar for most investors.
The $NVDA CEO, Jensen Huang just revealed the full 5 layer stack the AI super cycle is built upon… These 5 layers include: 1. Energy ~ $CEG, $VST, $OKLO, $EOSE, $GEV 2. Chips & Computing ~ $NVDA, $AMD, $TSM, $MU, $ARM 3. Cloud & Data Centers ~ $NBIS, $IREN, $CRWV, $APLD, $CIFR 4. AI Models ~ $MSFT, $GOOGL, $META, $AMZN, $ORCL 5. Applications ~ $PLTR, $TSLA, $NOW, $SNOW, $CRM Without these companies there is no AI; which is exactly why these names will continue to see massive long term growth. Save this for later…
$NVDA EARNINGS INCOMING Next week is not just a print — it’s a macro event for the entire AI trade. → Blackwell demand visibility is the #1 watch item → Data center revenue guide will set the tone for hyperscaler capex narratives → Any commentary on CoWoS/HBM supply constraints moves $MU, $SK Hynix proxies → Gross margin trajectory = the bull/bear fulcrum heading into H2 → China export exposure still an overhang — watch management tone carefully → Beat + raise keeps the AI supercycle thesis intact. Miss or soft guide = sector-wide reset Every name in AI infra — $AMD, $MRVL, $CRDO, $ALAB, $AVGO, $ARM — trades off this print. This isn’t just $NVDA earnings. It’s a referendum on the entire AI capital cycle. I am bullish for this. Not financial advice.
Leopold Aschenbrenner is a legend, but I'm not quite sure he can beat 3152.77% YTD in the Serenity Awareness fund. That being said, I've hit 23 different longs this year with 100-1000%+ YTD. 1. $AXTI 2. $AAOI 3. $SIVE 4. $LITE 5. $IQE 6. $AEHR 7. $CRCL 8. $EWY 9. Unimicron 10. Nitto Boseki 11. $OSS 12. $GDRZF 13. $RPI 14. $SOI 15. $ALRIB 16. $SNDK 17. $SIMO 18. $VPG 19. $TSEM 20. $ARM 21. $MRVL 22. $INTC 23. $LPK Do you remember all of these anon?
The Complete Semiconductor Playbook — AI Supercycle 2026 🤖 EDGE AI $QUIK · $CEVA · $SYNA · $QCOM · $INDI · $VLN · $SMTC → AI leaves the data center. Devices. Cars. Factories. Multi-year secular trend just getting started. 🏗️ AI INFRA $SGH · $SKYT · $AVGO · $NVDA · $ARM · $CRDO · $MRVL → The “Nvidia-only” era is over. The full stack is getting priced in. ⚡ POWER $POWI · $TXN · $MPWR · $VICR · $GANX · $AEHR → AI data centers are power monsters. Unsexy. Essential. Increasingly scarce. 🔬 FUTURE-TECH $LWLG · $AXTI · $POET · $TSEM · $GFS · $IPGP → Silicon photonics foundry capacity = national strategic asset. 🛠️ EQUIPMENT $AMAT · $LRCX · $KLAC · $ASML · $ONTO · $ACMR → Without these, nothing above gets built. Map the full stack. That’s where the alpha is. Not financial advice. DYOR
@xuefeng_huang If $SIVE goes the ADR route you can do an ADR conversion from Sweden to US and it will just start trading on US markets like $ARM.
$ARM KEY READ-THROUGHS FROM ARM HOLDINGS PLC Q4 FYE26 EARNINGS CALL Arm’s Q4 FYE26 materials carry unusually broad cross-market implications because the call was less about a one-quarter earnings beat and more about a potential re-architecture of AI infrastructure. The most important message is that Arm is framing the CPU as a renewed bottleneck in agentic AI, not a commoditized support chip behind accelerators. Management stated that data center royalties more than doubled YoY, that Arm has more than $2bn of customer demand for the Arm AGI CPU across FYE27/FYE28, and that cloud AI could become Arm’s largest business. The market implications are most acute for x86 data center CPU incumbents, hyperscaler custom silicon programs, NVIDIA’s rack-scale AI platform, server OEMs/ODMs, advanced foundry and memory suppliers, data center networking vendors, enterprise software/platform companies, EDA vendors, and selected mobile/auto semiconductor suppliers. The highest-conviction conclusion is that AI infrastructure value capture is broadening from GPUs alone toward CPU orchestration, rack-level integration, high-density power/thermal design, network offload, and software ecosystem migration. DATA CENTER X86 CPU SHARE RISK INTENSIFIES (READ-THROUGH 1) Affected companies and impact: Intel Corporation (INTC: US) negative/high; Advanced Micro Devices, Inc. (AMD: US) negative/high for sentiment and longer-duration data center CPU share, though near-term revenue impact is likely moderate given Arm AGI CPU production revenue begins only in Q4 FYE27. Supporting commentary/data: Management said “data center royalty has more than doubled year-over-year,” “data center royalty revenue continues to more than double year-on-year, and we see no break in this momentum,” and “I am actually confident that by the end of the decade, I believe, the largest market share by CPU type will be Arm.” The call also cited Google TPU 8t and TPU 8i “replacing x86 host processors with custom Arm Axion CPUs,” while Arm’s first production data center silicon was described as delivering “more than 2x the performance per rack compared with x86 platforms” and potentially reducing AI data center capex “by up to $10bn per gigawatt.” The investor presentation showed Arm cloud compute share by chip value rising from 9% in FYE22 to 20% in FYE25, with the FYE31 cloud compute market value opportunity shown at approximately $100bn. Transmission mechanism: The core risk is not merely socket substitution in traditional server CPUs. The more damaging mechanism is that accelerator-centric AI platforms increasingly use Arm for head nodes, host processors, orchestration nodes, and potentially dedicated CPU racks adjacent to GPU/accelerator racks. If TPUs, Trainium, NVIDIA Vera/Rubin systems, and hyperscaler custom platforms standardize on Arm-based CPUs, x86 attach to the highest-growth AI infrastructure deployments declines structurally. Enterprise software migration to Arm further erodes x86’s historical compatibility moat. Near-term trading catalyst: The read-through is immediately negative for x86 CPU sentiment because the call directly challenged AMD’s and Intel’s longer-term data center CPU share assumptions. The analyst Q&A explicitly referenced AMD’s stated 2030 CPU TAM/share commentary, and Arm’s response was assertive rather than cautious. Near-term estimate risk is limited because Arm maintained only approximately $90m of AGI CPU revenue expectation for Q4 FYE27, but multiple risk for x86 CPU profit pools is immediate. Longer-duration fundamental shift: The fundamental risk is high. Arm is targeting both royalty capture from hyperscaler-designed Arm CPUs and full chip revenue through the Arm AGI CPU. If the company approaches its FYE31 targets of $10bn IP/CSS revenue and $15bn AGI CPU chip revenue, the implied displaced or avoided x86 CPU market opportunity becomes material. NVIDIA’S RACK-SCALE AI PLATFORM IS FURTHER VALIDATED BY ARM CPU ADOPTION (READ-THROUGH 2) Affected companies and impact: NVIDIA Corporation (NVDA: US) positive/high. Intel Corporation (INTC: US) and Advanced Micro Devices, Inc. (AMD: US) negative/medium-to-high by competitive read-across because NVIDIA’s AI platform roadmap is increasingly paired with Arm rather than x86. Supporting commentary/data: Arm cited NVIDIA Vera as “the next-generation Arm-based CPU built for Agentic AI” and highlighted NVIDIA “building a standalone rack integrating 256 Vera CPUs.” In Q&A, management described “a Vera rack in between or two Vera racks” adjacent to Vera Rubin systems and referenced “a 200-kilowatt liquid-cooled rack” designed to sit in the data center next to GPU systems. Management also stated that “the three largest GPU providers across NVIDIA with Vera or Grace… Google pairing TPUs with Axion… and Trainium… all of those partners are now on Arm.” Transmission mechanism: NVIDIA benefits because Arm-based CPU orchestration strengthens NVIDIA’s full-stack rack architecture. Dedicated Arm CPU racks can improve utilization, scheduling, memory movement, and orchestration around GPU clusters. This increases the strategic value of NVIDIA’s platform beyond accelerator silicon and reinforces the system-level moat around Vera/Rubin. The more agentic workloads require CPU orchestration, the more NVIDIA can sell or influence complete rack-scale systems rather than discrete GPUs. Near-term trading catalyst: Positive for NVIDIA sentiment around the Vera/Rubin roadmap and rack-scale AI systems. The call reinforces the view that CPUs are not a drag on GPU deployment but a system-level enabler of higher GPU utilization and denser AI infrastructure. Longer-duration fundamental shift: Positive and potentially high magnitude. If agentic AI drives dedicated CPU racks adjacent to GPU racks, NVIDIA’s addressable content per AI data center expands while x86 CPU relevance declines. This also supports continued investor focus on NVIDIA’s ability to capture platform-level economics rather than only accelerator ASPs. HYPERSCALER CUSTOM SILICON ECONOMICS IMPROVE AS ARM BECOMES THE DEFAULT AI CPU FABRIC (READ-THROUGH 3) Affected companies and impact: Alphabet Inc. (GOOGL: US) positive/high; https://t.co/SpqvHNUxpK, Inc. (AMZN: US) positive/high; Microsoft Corporation (MSFT: US) positive/medium-high; Meta Platforms, Inc. (META: US) positive/high. Supporting commentary/data: Arm highlighted Google’s TPU 8t and TPU 8i replacing x86 host processors with custom Arm Axion CPUs, AWS scaling Graviton alongside Trainium and Nitro, Microsoft advancing Cobalt for Azure workloads, and Meta as Arm’s lead partner and co-developer for the AGI CPU. The shareholder letter stated that Arm’s market share of CPU compute is now approximately 50% among top hyperscalers. It also cited AWS custom silicon, including Graviton, Trainium, and Nitro, as running at more than $20bn annually and growing triple digits YoY. Management stated that Meta is working with Arm on a multi-generation roadmap “to support personal superintelligence for more than 3bn users.” Transmission mechanism: Custom Arm CPUs allow hyperscalers to reduce power, improve performance-per-dollar, reduce dependence on merchant x86 CPU suppliers, and integrate CPUs more tightly with proprietary accelerators. Alphabet benefits through TPU plus Axion integration. Amazon benefits through Graviton/Trainium/Nitro economics and broader Arm software compatibility. Microsoft benefits as Cobalt adoption expands across Azure workloads. Meta benefits directly from co-developing a CPU roadmap aligned with internal AI infrastructure and potentially lower capex per unit of compute. Near-term trading catalyst: Positive for hyperscaler capex-efficiency narratives. The call gives investors more evidence that AI capex is not only increasing in absolute dollars but also being optimized through custom silicon and architecture-level control. This can partially offset concerns around AI infrastructure returns on invested capital. Longer-duration fundamental shift: Positive and high magnitude. Arm’s architecture reduces the strategic dependence of hyperscalers on x86 merchant CPUs and supports vertical integration. If enterprise software increasingly validates on Arm, the custom silicon advantage becomes more durable because hyperscalers can shift broader workloads, not only internal AI workloads, onto Arm-based infrastructure. META HAS THE CLEAREST DIRECT CUSTOMER READ-THROUGH FROM THE ARM AGI CPU (READ-THROUGH 4) Affected companies and impact: Meta Platforms, Inc. (META: US) positive/high for longer-duration AI infrastructure economics; positive/medium for near-term narrative because financial benefits are unlikely to be visible immediately. Supporting commentary/data: Arm said Meta is the “lead partner and co-developer” for the first production silicon product and is working with Arm on a “multi-generation roadmap to support personal superintelligence for more than 3bn users.” Arm also claimed the first production silicon could deliver more than 2x performance per rack versus x86 platforms and reduce AI data center capex by up to $10bn per gigawatt. Transmission mechanism: Meta is exposed to one of the largest AI capex debates in the market. A co-developed Arm AGI CPU gives Meta a route to optimize CPU orchestration around internal AI workloads, reduce power/capex intensity, and potentially gain architectural differentiation without building every CPU component internally. If the capex-per-compute claims prove even partially valid, the impact on Meta’s long-term AI return profile could be meaningful. Near-term trading catalyst: Positive for sentiment but unlikely to drive immediate earnings revisions. Arm maintained its supply-backed AGI CPU revenue outlook, and meaningful revenue is expected mainly in FYE28. The near-term implication is that Meta’s AI infrastructure strategy includes a credible path to efficiency gains beyond GPU procurement. Longer-duration fundamental shift: Positive and potentially high magnitude. The relationship suggests Meta is not merely buying off-the-shelf AI infrastructure; it is shaping CPU architecture around its own agentic and personal-superintelligence workloads. That creates potential long-term differentiation in cost per inference, workload orchestration, and data center power density. SERVER OEMS AND ODMS GET A DIRECT RACK-LEVEL DEMAND SIGNAL FROM ARM AGI CPU (READ-THROUGH 5) Affected companies and impact: Super Micro Computer, Inc. (SMCI: US) positive/medium-high; Lenovo Group Limited (0992: Hong Kong) positive/medium; Quanta Computer Inc. (2382: Taiwan) positive/medium; ASRock Inc. (3515: Taiwan) positive/medium. Supporting commentary/data: In Q&A, management said customers can buy “finished racks from our partners such as Supermicro, Lenovo, ASRock,” enabling them to “order and deploy quite quickly.” The shareholder letter also stated that commercial systems based on Arm AGI CPU are available to order from Supermicro, Lenovo, Quanta, and ASRock. Management tied the demand increase to the availability of rack designs that customers can put into data halls with limited friction. Transmission mechanism: Arm AGI CPU demand converts into physical rack integration revenue for OEMs/ODMs. These companies benefit from system design, rack assembly, validation, deployment, liquid cooling integration, and customer delivery. The key point is that Arm is not positioning the AGI CPU as a component-only product; it is enabling customers to purchase ready-to-deploy rack systems. Near-term trading catalyst: Positive but supply-gated. Arm disclosed more than $2bn of demand across FYE27/FYE28 but maintained a $1bn supply-backed revenue outlook. Any evidence that Arm secures more wafers, memory, packaging, or test capacity would be a direct catalyst for rack partners because additional supply can translate into incremental system orders. Longer-duration fundamental shift: Positive and medium-to-high magnitude. If Arm-based AI CPU racks become standard complements to GPU/accelerator racks, server OEMs/ODMs gain another high-growth architecture cycle beyond NVIDIA GPU systems and traditional x86 servers. Margins may remain structurally lower than component suppliers, but revenue opportunity and platform relevance improve. DATA CENTER POWER AND LIQUID COOLING BENEFIT FROM HIGHER RACK DENSITY, DESPITE ARM EFFICIENCY CLAIMS (READ-THROUGH 6) Affected companies and impact: Vertiv Holdings Co. (VRT: US) positive/medium; Eaton Corporation plc (ETN: US) positive/medium; Schneider Electric SE (SU: France) positive/medium. Supporting commentary/data: Arm framed agentic AI as requiring approximately 4x more CPU cores per gigawatt, moving from approximately 30m CPU cores/GW to approximately 120m CPU cores/GW. Management also referenced NVIDIA’s dedicated Vera rack as “a 200-kilowatt liquid-cooled rack” designed to sit adjacent to Vera Rubin systems. Transmission mechanism: Even if Arm improves performance per watt, agentic AI increases compute density and orchestration requirements inside the same power envelope. Dedicated CPU racks adjacent to accelerator racks add complexity in power distribution, liquid cooling, thermal management, monitoring, and data center electrical infrastructure. The move from server-level compute to rack-scale, liquid-cooled AI systems increases content opportunity for power and thermal infrastructure suppliers. Near-term trading catalyst: Positive for sentiment around liquid cooling and high-density power distribution. The Arm call provides an additional architectural rationale for 200kW-class racks beyond GPU density alone. Longer-duration fundamental shift: Positive but with an efficiency caveat. Arm’s claim of up to $10bn/GW capex reduction indicates better compute efficiency could reduce infrastructure dollars per unit of output. However, the larger structural driver is that agentic workloads may require materially more CPU orchestration and higher rack density. Net impact is positive for high-density power/thermal suppliers, but not necessarily linear with total AI capex. FOUNDRY, MEMORY, PACKAGING, AND TEST CAPACITY BECOME THE GATING FACTORS FOR ARM’S AI CPU UPSIDE (READ-THROUGH 7) Affected companies and impact: Taiwan Semiconductor Manufacturing Company Limited (2330: Taiwan) positive/high; SK hynix Inc. (000660: Korea) positive/medium; Micron Technology, Inc. (MU: US) positive/medium; Samsung Electronics Co., Ltd. (005930: Korea) positive/medium. Supporting commentary/data: Arm disclosed more than $2bn of AGI CPU demand across FYE27/FYE28 but maintained a $1bn outlook because it must secure supply-chain capacity. Management said the $1bn already supported includes “memory,” “wafers,” “packaging,” and “access to test equipment,” and that for the $2bn demand level the company is “now in the process of securing supply.” The shareholder letter listed Micron, SK Hynix, TSMC, and Samsung among companies supporting expansion of the Arm compute platform into silicon. Transmission mechanism: The supply bottleneck shifts incremental economics toward capacity owners. TSMC benefits from wafer demand if Arm AGI CPU ramps on leading-edge or advanced-node capacity. SK hynix, Micron, and Samsung benefit from incremental memory demand tied to AI CPU racks and data center systems, although exact memory type and content per rack were not disclosed. Samsung also has potential dual exposure through memory, foundry, and Arm ecosystem participation. Near-term trading catalyst: Positive for supplier sentiment because demand exceeds supply-backed revenue. The key trading catalyst is any future indication that Arm has secured incremental capacity above the current $1bn revenue outlook. Longer-duration fundamental shift: Positive and potentially high magnitude if Arm’s FYE31 $15bn AGI CPU chip revenue target is credible. The magnitude is highest for foundry capacity and medium for memory suppliers because the call did not quantify memory content per CPU/rack. DATA CENTER NETWORKING, DPU, SMARTNIC, AND EDGE TRAFFIC MANAGEMENT DEMAND GETS A STRONG POSITIVE SIGNAL (READ-THROUGH 8) Affected companies and impact: Marvell Technology, Inc. (MRVL: US) positive/medium; Broadcom Inc. (AVGO: US) positive/medium; NVIDIA Corporation (NVDA: US) positive/medium through DPU/networking adjacency; F5, Inc. (FFIV: US) positive/medium; Cloudflare, Inc. (NET: US) positive/medium; SK Telecom Co., Ltd. (017670: Korea) positive/low-to-medium. Supporting commentary/data: Arm said the biggest contribution to royalty growth came from Cloud AI, driven partly by “increased deployments of data center networking chips, particularly DPUs and SmartNICs, where Arm has close to 100% market share.” Management also disclosed a next-generation CSS license for data center networking chips. It cited Cloudflare deploying Arm across its global network for traffic management, security, and AI inference closer to users, and design wins with F5 and SK Telecom. Transmission mechanism: AI infrastructure stresses the network and control plane, not only compute. DPUs, SmartNICs, network processors, and traffic-management appliances increasingly need embedded CPUs to manage packet processing, security, telemetry, inference routing, and east-west data movement. Arm’s near-100% share in DPUs/SmartNICs implies that higher AI networking content flows through Arm-based designs and benefits networking silicon vendors and infrastructure companies aligned to that ecosystem. Near-term trading catalyst: Positive for networking and DPU-related sentiment because Arm’s royalty growth already reflects deployment momentum, not only future design wins. Longer-duration fundamental shift: Positive and medium-to-high magnitude. AI inference closer to users and agentic workloads should increase demand for distributed traffic management, security offload, and programmable networking. This supports Marvell/Broadcom networking silicon, NVIDIA DPU/networking assets, F5 application delivery/security infrastructure, Cloudflare’s edge network economics, and SK Telecom’s AI infrastructure positioning. CUSTOM ASIC SUPPLIERS RECEIVE A MIXED SIGNAL: ARM EXPANDS THE MARKET BUT CAPTURES MORE CPU VALUE DIRECTLY (READ-THROUGH 9) Affected companies and impact: Broadcom Inc. (AVGO: US) mixed/medium; Marvell Technology, Inc. (MRVL: US) mixed/medium. Supporting commentary/data: The shareholder letter listed Broadcom and Marvell among companies supporting Arm’s expansion into silicon. However, the investor presentation explicitly showed that the Arm AGI CPU allows Arm to “capture full chip value,” expanding Arm’s FYE26 cloud AI opportunity from $2.4bn under IP/CSS-only monetization to $24bn when Arm captures full chip value. Management also emphasized that Arm now sells a “complete chip solution” directly to cloud and AI data center companies. Transmission mechanism: The positive side is that Arm-based custom silicon growth expands the overall design ecosystem, increases demand for networking, switching, DPU, and accelerator-adjacent silicon, and reinforces Arm software compatibility. The negative side is that Arm is moving up the stack from IP licensor to chip supplier, which could disintermediate some semi-custom CPU or integration opportunities that might otherwise have gone to ASIC vendors. The risk is most relevant where customers want a ready-to-deploy Arm CPU rather than a bespoke design project. Near-term trading catalyst: Mixed but probably not immediately negative. Management said “every single partner we asked said yes,” suggesting no near-term ecosystem backlash. Broadcom and Marvell remain beneficiaries of AI networking and custom silicon demand. Longer-duration fundamental shift: Mixed and important. If Arm reaches $15bn of AGI CPU chip revenue by FYE31, it will have captured a meaningful portion of the cloud AI CPU silicon value chain directly. That could limit some custom CPU ASIC opportunities while still expanding the broader AI silicon ecosystem in which Broadcom and Marvell participate.
We are currently in a “once in a lifetime” AI super cycle… Phase 1 was: (already gone) Semiconductors ~ $NVDA, $AMD, $INTC, $ARM Phase 2 is: (passing by now) Memory ~ $MU, $SNDK, $WDC Photonics ~ $AAOI, $AEHR, $LITE, $MRVL The current phase is Neo Cloud/AI infrastructure: $IREN, $NBIS, $CRWV, $CIFR, $APLD Next wave (many will miss) Rare Earths ~ $USAR, $MP, $UUUU, $FCX Power & Cooling~ $VRT, $CEG, $OKLO, $OSS Finally it all concludes with these 3 sectors: Robotics ~ $TSLA, $PATH, $SERV Space ~ $RKLB, $ASTS, $PL, $LUNR Drones ~ $ONDS, $AVAV, $LMT Many will make generational wealth from this AI super cycle over the next 7 months. Save this to look back on later…
I guess, post earnings when $ARM touched $268... $ARM is now #18 on the individual stock list that I went long on that hit 100%-1000%+ YTD? I've lost count TBH. Some others like $LPK and $SIMO and $HPS.A are getting really close now. But feels like I'm one of the few ones out there on X with actual receipts of all the returns + original thesis post.
BREAKING: $ARM earnings: - EPS: $0.6, est: $0.58 - Revenue: $1.49 billion, est: $1.47 billion $IONQ earnings: - EPS: $2.07, est: -$0.52 - Revenue: $64 million, est: $49 million $DASH: - EPS: $0.42, est: $0.36 - Revenue: $4.03 billion, est: $4.15 billion
$ARM (Bloomberg) -- ARM Holdings reported adjusted earnings per share for the fourth quarter that beat the average analyst estimate. FOURTH QUARTER RESULTS Adjusted EPS 60c, estimate 58c (Bloomberg Consensus) EPS 29c Total revenue $1.49 billion, estimate $1.47 billion Adjusted net income $641 million, estimate $624.3 million Adjusted gross profit $1.47 billionAdjusted gross margin 98.3%, estimate 98.1% Adjusted operating income $731 million, estimate $696.4 millionAdjusted operating margin 49.1%
$ARM seems to be having a fun time. Almost triple digit return in 1 1/2 months since I went long on their CPU projections. Markets really like AI CPUs huh? https://t.co/hgWPOB6tIN
Capital rotation is getting louder in semis and AI infrastructure. Money is flowing into names like $MU, $SNDK, $AMD, $ARM, $CRDO, $MRVL, $INTC, $AAOI, $ALAB, and $TSEM. The market is rewarding memory, networking, packaging, and AI compute plays right now
$ARM doesn’t need to beat — it needs to guide. Watch FY2027 royalty trajectory and licensing ACV growth. That’s the real print. Earnings AH tomorrow
Earnings watchlist for the week: 𝗠𝗼𝗻𝗱𝗮𝘆 After Hours: $PLTR $FN $BWXT $AEIS $FLY $ADEA $ON 𝗧𝘂𝗲𝘀𝗱𝗮𝘆 Pre-Market: $SHOP $PYPL $ENLT $ETN After Hours: $AMD $ANET $LITE $ALAB $NVTS $NRGV $SU $WOLF 𝗪𝗲𝗱𝗻𝗲𝘀𝗱𝗮𝘆 Pre-Market: $UBER $FLEX $SOLS $NRG $HUT After Hours: $ARM $COHR $UUUU $FSLY $SEZL $AOSL $APP $IONQ $SNAP 𝗧𝗵𝘂𝗿𝘀𝗱𝗮𝘆 Pre-Market: $DAVE $HWM $VST $DDOG $BSKY After Hours: $NET $CRWV $RKLB $IREN $LASR $PDFS $OPEN 𝗙𝗿𝗶𝗱𝗮𝘆 Pre-Market: $WULF $UI
Still can’t believe Spirit Airlines is shutting down. Very harmful, now there’s less incentives to cut costs. There’s a difference between blocking $ARM and $NVDA acquisition (actual antitrust). And an Airline rescuing Spirit from bankruptcy. Thanks Senator Warren. https://t.co/BTn9D61eek
I had a decent month. Everything from: $AAOI went up 100% to $SOI went up 153.9%. $SNDK +70% or $INTC + 97.7% or $MRVL +54% or $ARM +41.5% were underperformers for me personally. Curious how you all did?
Further confirmation from the source on how GPT-5.5 was trained. It was on GB200 not G/B300. Not even maxing out Blackwell generation yet. It's worth noting they are using the $ARM Grace chip for this training run, not x86; need to think through the implications if any. So much more to come. $NVDA $MU $SNDK $LITE $ORCL
Here’s my 15-stock AI infrastructure watchlist AI INFRASTRUCTURE CORE $MRVL → Custom AI silicon for hyperscalers like Google and AWS → Data center now the core growth engine → Strong XPU positioning in AI compute $CRDO → Critical connectivity layer inside AI clusters → Expanding into silicon photonics and optical transceivers → Direct beneficiary of hyperscaler GPU scaling $ALAB → PCIe/CXL connectivity solving AI server bottlenecks → Key enabler for GPU communication efficiency → Strong execution and AI infrastructure leverage $AAOI → Riding the 800G and 1.6T optical upgrade cycle → Vertical integration gives margin and supply edge → Hyperscaler demand remains strong $MXL → Emerging optical DSP player in AI infrastructure → Pivoted from broadband into data center growth → Early in hyperscaler qualification cycle MEGA-CAP AI COMPOUNDERS $MSFT → Enterprise AI leader via Copilot and Azure → Massive distribution advantage through software ecosystem → AI monetization still in early innings $GOOG → Search funds AI innovation and cloud expansion → Strong custom silicon and AI infrastructure strategy → Multiple growth engines beyond search $AMZN → AWS remains the AI cloud backbone → Aggressive AI infrastructure spending → Retail and ads fuel long-term AI investment SEMICONDUCTOR CYCLE PLAYS $AMD → Leading Nvidia alternative in AI compute → Enterprise traction growing with MI300 → Multiple cycle tailwinds in AI and PCs $MU → HBM memory is essential for AI GPUs → Direct play on AI compute demand → Strong AI-driven memory cycle setup $INTC → Foundry turnaround with strategic US importance → Big upside if execution improves → High risk, high reward setup $ARM → Royalty model across global chip ecosystem → Expanding into AI edge and data center → Benefits from industry-wide chip growth CONNECTIVITY, POWER & INFRA $SIMO → Storage controllers powering AI data growth → NAND cycle recovery adds tailwind → Undervalued storage infrastructure play $NOK → Optical and fiber backbone for data traffic growth → Beneficiary of telecom and hyperscaler upgrades → Defensive infrastructure exposure $BE → On-site energy for power-hungry AI data centers → Solves grid bottleneck challenges → Direct energy infrastructure AI play AI is not one stock. It’s chips, memory, optics, networking, storage, and power. Follow the infrastructure. That’s where the real compounding happens. Not financial advice.
The full CPU stack: understands each layer, you can find many play Architecture - $ARM Design- $AMD $INTC $QCOM Foundry- $TSM $GFS Memory- $MU Storage- $SIMO Connectivity- $ALAB $TMBS Custom silicon- $AVGO $MRVL Future Architecture (RISC-V via Tenstorrent, $CAN)
@__visionxry__ Bro $ARM is up 45% this month, what are you complaining about lol
3. $ARM - Arm Holdings The architectural backbone of every custom AI server CPU being built by hyperscalers - Google Axion, Amazon Graviton, Microsoft Cobalt. CEO said CPU core demand could rise 4x per GW of AI capacity as agentic workloads scale. Just launched the AGI CPU, Arm's first-ever production silicon for AI data centers. Every new Arm-powered AI server design means more licensing revenue.
Semis and photonics are seeing a healthy pullback — that’s often just a technical reset after strong momentum. If you missed the last run and had the patience not to chase, this is the time to build your watchlist and map out the levels that make sense for your entries. Good setups come to those who wait for price, not emotions. $CRDO $MRVL $AMD $ARM $AAOI $COHR $GFS $RMBS
$AMD $ARM $MRVL $CRDO $AMKR $RMBS $TSEM all seeing pullbacks after strong recent runs. Nothing unusual — healthy consolidation after momentum can create the next setup. Keep them on the watchlist. Strong names often give better entries when sentiment cools and price resets.
The full CPU stack: check for details in post Architecture - $ARM Design- $AMD $INTC $QCOM Foundry- $TSM $GFS Memory- $MU Storage- $SIMO Connectivity- $ALAB $TMBS Custom silicon- $AVGO $MRVL Future Architecture (RISC-V via Tenstorrent, $CAN)
TLDR of recent news + bottlenecks that go brr: 1. CPU bottleneck - $INTC CEO said AI inference pushed CPU Ratio From 1:8 to 1:1. CPUs go brr ( $AMD, Intel, $ARM) -> $AMAT / $TSM / $KLAC, etc. go brr. 2. PGME / PGMEA shortage. DuPont, Shiny Chemical, Daxin, San Fu, $DOW and others go brr? Photoresist bottleneck go brr? 3. Microcontroller potential bottleneck + price hikes (Arterytek/Arterychip) was weighing price hikes on AI capacity squeezes. MCU companies potentially go brr? 4. President invoked the "Defense Production Act" this week, it included: -Transformers - transmission components - advanced conductors - power electronics - substations - high-voltage circuit breakers - protective relays, capacitor banks - electrical core steel As "severe shortages". Stuff like $AMSC, $PLPC, $POWL, $VICR, $ATKR, $HPS.A go brr. 5. $GOOGL ramps new TPU servers. Google splits AI chips into training and inference TPUs. Taiwan happy. Mediatek and others go brr? 6. Samsung, Kingston lift SSD prices by over 10%. SSD prices keep going brrr? 7. T-glass fiberglass shortages keep getting worse? Nittobo and others keep going brrr? 8. Bromine, essential for etching circuits and flame retardancy, has surged to $12,000 per metric ton. ICL Group in Israel apparently controls 40% of the global supply? Not as familiar with this but questionable brrr? 9. "Epitaxy manufacturer LandMark Optoelectronics reporting output still far below customer needs". Uhh $IQE and others go brr? 10. "AI data centers hit interconnect limits, boosting optical module demand". "the bottleneck is no longer computing power alone, but how that power is connected." Photonics from $AAOI, $LITE, $COHR, Innolight and others keep going brr? next gen from $SIVE, $POET, $MRVL, Win Semi and others go brr? Basically AI semi supply chains go brr because there's widespread shortages everywhere due to AI hyperscaler demand.
The full CPU stack: Architecture - $ARM Design- $AMD $INTC $QCOM Foundry- $TSM $GFS Memory- $MU Storage- $SIMO Connectivity- $ALAB $TMBS Custom silicon- $AVGO $MRVL Future Architecture (RISC-V via Tenstorrent, $CAN)
CPU ecosystem map across every tier: TIER 1 — Pure CPU Giants $INTC — Intel. The turnaround is real. Q1 2026: revenue $13.58B, EPS $0.29 crushed $0.01 consensus. Data Center revenue +22% YoY. Stock surged nearly 25% in a single session. CEO Lip-Bu Tan declared: “The next wave of AI will bring intelligence closer to the end user — from foundational models to inference to agentic.”  Up 80%+ YTD before that gap. $AMD — EPYC CPU order book nearly sold out for 2026. Q4 2025 revenue $10.27B, up 34% YoY, with record Data Center revenue of $5.38B.  MI400 platform coming. 1-year return: +269%. The share-gainer. $ARM — The toll road of CPUs. Used in 99% of the world’s smartphone CPU cores.  In 2026, ARM announced the launch of its own CPU products on top of its existing royalty business.  Up +87% YTD before Intel’s earnings day gap. Capital-light, royalty compounder. TIER 2 — CPU-ADJACENT INFRASTRUCTURE $QCOM — Snapdragon X Elite attacking PC CPU market with ARM-based Oryon cores. Edge AI CPU momentum building across enterprise. Automotive + IoT CPU exposure via Snapdragon Cockpit Elite. $MRVL — Custom XPU + CPU silicon. 18+ socket wins. $75B design pipeline. Data center CPU-adjacent workloads = direct tailwind. $AVGO — Custom ASICs + CPU workload accelerators for hyperscalers. AI business grew 106% YoY to $8.4B last quarter. Projects $100B+ business by 2027.  The ASIC answer to x86. $TSM — Makes every CPU on the planet. TSMC fabs AMD EPYC, Apple silicon, Qualcomm Snapdragon. ~70% foundry market share. Trading 14% below Morningstar fair value of $428.  You can’t build a CPU without TSMC. TIER 3 — CPU ENABLERS & PICKS/SHOVELS $MU — Every CPU needs memory. Micron can only fulfill half-to-two-thirds of current medium-term demand. Revenue was $13.6B two quarters ago, $23.9B last quarter, guiding $33.5B next quarter.  Memory is the bottleneck. $SIMO — SSD controllers feeding CPU storage layers in AI data centers. 46% YoY revenue growth. PCIe Gen5 controllers showcased at NVIDIA GTC 2026. The quiet CPU enabler. $ALAB — PCIe + CXL connectivity silicon — the bus that CPUs talk to everything else on. 75%+ gross margins. Zero debt. Every AI CPU cluster needs Astera Labs. $RMBS — Rambus. Memory interface IP. LPDDR5X SOCAMM2 server memory chipset for AI CPUs. ~80% gross margins. Capital-light IP model — gets paid every time a CPU ships with their interface. $GFS — GlobalFoundries. Specialty foundry for RF, automotive, aerospace, and IoT CPUs. The mature-node CPU manufacturer for defense + industrial. TIER 4 — RISC-V REVOLUTION (THE FUTURE CPU WAR) Tenstorrent (private — Jim Keller’s company) — First-gen RISC-V CPU “Ascalon” delivers 10-20 SPECint2006/GHz, competing directly with ARM’s Neoverse V2.  IP licensees already include LG and Hyundai.  Backed by Hyundai, Kia, Samsung. Watch for IPO. $CAN — Canaan Creative. Launched the world’s first commercial edge AI chip based on RISC-V.  Public, listed, speculative. RISC-V + AI edge play in one ticker. SiFive (private) — The ARM of RISC-V. Intel tried to acquire for $2B. CPU IP licensing model. IPO candidate to watch. TIER 5 — HYPERSCALER CUSTOM CPUs (Own The CPU, Own The Cloud) $AMZN — Graviton ARM-based CPUs + Trainium AI chips. Trainium 2 and 3 at max capacity. Nearly all Gen 4 capacity already pre-sold 18 months out.  AWS custom CPU = lowest cloud compute cost. $GOOGL — TPU + custom ARM CPU for Google Cloud. Fastest-growing cloud CPU fleet. $MSFT — Azure custom ARM silicon + Cobalt CPU. AI inference at edge + cloud. Azure fastest-growing hyperscaler by AI workload. The GPU era is maturing. The CPU supercycle is just loading up. Every layer of this stack is a potential winner. Not financial advice.
The AI trade is evolving — and CPUs may be the next breakout. For the last two years, everyone chased GPUs. That made sense — training models needed massive parallel compute. But the next phase of AI looks different: - Agentic AI needs orchestration, memory handling, workflow execution, and sequential reasoning — CPU-heavy workloads. - Inference at scale isn’t just about GPUs. CPUs manage the stack, coordinate workloads, and keep AI systems moving. - Data center CPU demand is tightening, and supply constraints are starting to show. Names to watch: $AMD — EPYC demand remains strong, with enterprise and cloud adoption accelerating. Earnings could be a major catalyst. $INTC — Turnaround momentum is building. Better-than-expected guidance has put Intel back in the conversation. $ARM — The architecture behind modern computing. More AI devices = more royalty leverage. $MRVL — Custom silicon and infrastructure exposure make it a strong secondary AI beneficiary. $QCOM — Quietly building CPU momentum through Snapdragon and edge AI. Big picture: AI is shifting from training → inference → autonomous agents. Phase 1 rewarded GPU leaders. Phase 2 could reward CPU infrastructure. The market always rotates before the crowd notices. Keep CPUs on your radar.
10 Stocks Built for LT Returns - $MU — The HBM supercycle play. Entire 2026 HBM output already sold out. Forecasts 40% CAGR in HBM market to ~$100B, surpassing all of 2024 DRAM. Q1: 57% YoY sales growth, 56.8% gross margins, EPS up 167%. This is the memory backbone of AI. $MRVL — AI data center powerhouse. 74% of revenue now from data centers. 18+ XPU socket wins, $75B design pipeline, targeting 20% of a $94B TAM by 2028. Acquiring Celestial AI for photonic fabric tech. Custom silicon + optics = long runway. $CRDO — The picks & shovels play on AI networking. Stock up 1,700% since 2022 IPO. Active Electrical Cables dominating hyperscaler deployments. $10B+ TAM across 5 product pillars. AEC adoption still early innings — long growth runway ahead. $ARM — The architecture that runs everything. Every smartphone, AI edge device, and custom chip from Apple to NVIDIA to AWS runs on ARM IP. Royalty + licensing model = recurring revenue with zero fab risk. As custom silicon proliferates, ARM gets paid more per chip. The toll road of semiconductors. $ALAB — Elite AI connectivity pure play. 75.7% gross margins, 25.7% profit margins, $1.1B+ cash, zero debt. PCIe + CXL connectivity for hyperscalers. 17 of 19 analysts rate Buy with median PT of $205. Premium valuation but premium business. $RMBS — The memory IP kingpin. ~80% gross margins, ~40% EBIT margins. Just launched SOCAMM2 LPDDR5X server memory chipset targeting AI data centers. Capital-light, IP-driven model = durable cash flows. Underrated in the AI memory stack conversation. $AAOI — AI optical networking rocket ship. Up 1,200%+ in one year. Supplying 800G and 1.6T fiber-optic modules to hyperscalers. 97% revenue concentration with top customers — high risk but massive growth. Earnings May 7. Valuation stretched at 180x+ forward. $APH — Amphenol. The quiet compounder. Makes connectors, sensors & cables used in AI data centers, EVs, defense, aerospace. Boring product, extraordinary execution. Long-term hold for diversified tech infrastructure exposure. $MXL — Just surged 80% in a day on Q1 earnings beat. Optical data center business tied directly to AI infrastructure. Needham upgraded to Buy, peers it with MRVL/CRDO/ALAB on valuation multiples. Increased 2026 outlook. The laggard that caught up fast. $GFS — GlobalFoundries. Specialty foundry with focus on mature nodes — RF chips, automotive, aerospace, IoT. Less exposure to bleeding-edge AI race but massive structural demand from defense + auto electrification. Stable, strategic, long-duration play. all positioned at the intersection of AI infrastructure, high-speed connectivity, and semiconductor supply chain. This is where capital is flowing for the next decade. Time in market > timing the market. Not financial advice.
$GOOGL $ARM $NVDA $LITE This is an outstanding interview. Lots of great visibility into Google's cloud and TPU business, as well as color on some of their key customers. EXECUTIVE OVERVIEW The Kurian interview is best understood as a systems-level disclosure, not a conventional model-layer discussion. The central message is that Google’s AI roadmap is being organized around full-stack control: proprietary TPU silicon, Nvidia GPU optionality, Axion Arm CPUs, Intel and AMD CPU support, custom networking, storage, data-center design, energy procurement, enterprise distribution, Gemini models, and agent orchestration. The most important inference is that Google is positioning itself as an integrated AI utility: it can compete at the model layer while also selling capacity to model competitors; it can use Search, YouTube, Cloud, and enterprise cash flows to fund infrastructure; and it can improve unit economics by owning meaningful portions of the hardware and software stack. Kurian’s clearest strategic point was that AI capacity scarcity is expected to persist for roughly 10 years, and that owning silicon in a constrained market creates structurally better unit economics than reselling someone else’s accelerators. The most important ecosystem implication is that generative AI is moving from a model race into an infrastructure supply-chain race. Model quality remains critical, but the binding constraints are increasingly memory bandwidth, accelerator utilization, interconnect latency, storage throughput, VM orchestration, local disk, power availability, data-center deployment cycle time, and energy efficiency per token. Google’s disclosed 8th-generation TPU architecture validates this shift: TPU 8t is optimized for frontier training and TPU 8i is optimized for low-latency inference, reasoning, reinforcement learning, MoE routing, KV cache, and agent workflows. Google Cloud’s Next ’26 materials state that TPU 8t and TPU 8i were created because infrastructure requirements for pre-training, post-training, and real-time serving have diverged, while AI Hypercomputer now integrates accelerators, CPUs, storage, networking, software frameworks, orchestration, and GKE runtime improvements into a unified stack. (Google Cloud) Google’s roadmap points toward 3 simultaneous monetization loops. The 1st loop is internal: Search, Gemini app, Workspace, YouTube, advertising, and DeepMind frontier models consume the infrastructure. The 2nd loop is external: enterprise customers, AI labs, capital markets, HPC users, and government labs rent or consume the infrastructure. The 3rd loop is ecosystem control: Google can attract external workloads onto TPUs, drive higher TPU volumes, amortize silicon R&D, improve supply-chain bargaining power, and reduce relative dependence on Nvidia economics. This does not eliminate Nvidia demand; Google continues to offer Nvidia GPUs and will support Vera Rubin NVL72 through A5X. It does, however, create a credible long-term substitution path for captive hyperscale workloads and model providers willing to optimize for TPUs. Google Cloud explicitly says Nvidia GPUs remain a core part of its accelerator portfolio, while Virgo Network will support both TPU 8t and A5X powered by Nvidia Vera Rubin NVL72. (Google Cloud) GOOGLE’S STRATEGIC POSITION: AN AI FACTORY WITH MERCHANT CAPACITY The interview makes clear that Google is not behaving like a pure model lab, a pure cloud distributor, or a pure chip vendor. It is behaving like a vertically integrated AI factory with merchant capacity. Kurian described a model in which Google monetizes tokens directly through Gemini, monetizes other parties’ models running on Google infrastructure, monetizes TPU capacity with external labs, and increasingly places TPU systems in customer venues for latency-sensitive workloads such as capital markets. The capital markets example is particularly notable because it expands TPUs beyond classic LLM training and inference into inference-style numerical workloads, low-latency quantitative research, and potentially venue-proximate AI infrastructure. This suggests Google is trying to convert TPU from an internal accelerator into a broader domain-specific infrastructure platform. The key strategic distinction is ownership of IP. Kurian repeatedly contrasted Google’s position with resellers that must buy accelerators, package them into cloud offerings, and compete primarily through capacity allocation and software services. Google still buys and sells Nvidia capacity, but the TPU stack gives it an owned path to differentiated price-performance. This matters because AI infrastructure has entered a supply-constrained regime where the value pool can migrate upstream toward scarce components: accelerators, HBM, advanced packaging, networking, power, and data-center real estate. In that environment, a cloud vendor without its own accelerator is more exposed to gross-margin compression when component prices rise. Google’s TPU path reduces that exposure and provides a mechanism to monetize scarcity without fully surrendering economics to external silicon suppliers. This does not imply Google should hoard every TPU for Gemini. Kurian’s answer was economically rational: the company needs recurring cash flow to fund training and infrastructure, and venture capital cannot indefinitely fund model providers if inference margins fail to cover training costs. That logic is a direct challenge to model-only companies with weak gross margins, subsidized usage, and rising compute obligations. The implication is that the AI market will increasingly favor companies that control 1 of 3 funding sources: a massive cash cow, merchant cloud infrastructure, or strategic hyperscaler balance-sheet support. Google has all 3 through Services, Cloud, and the ability to supply Anthropic and other labs. FINANCIAL AND CAPEX READ-THROUGH Alphabet’s financial disclosures support the Kurian thesis that Google can sustain an infrastructure arms race more credibly than most standalone labs. In Q4 2025, Google Cloud revenue grew 48% to $17.7B, Cloud operating income reached $5.3B, Cloud operating margin expanded to 30.1%, and Cloud backlog rose 55% sequentially to $240B. Alphabet generated $164.7B of operating cash flow and $73.3B of free cash flow for FY 2025, and ended the year with $126.8B of cash and marketable securities. The company guided to 2026 capex of $175B-$185B, explicitly tied to frontier model development, Google Services, Cloud customer demand, and strategic investments. (Alphabet Investor Relations) The investment implication is 2-sided. On the positive side, Google Cloud is no longer a low-margin strategic side project; it is a fast-growing, high-backlog, high-margin enterprise AI infrastructure business with meaningful operating leverage. On the negative side, the capex step-up is enormous, and the P&L will face pressure from depreciation and energy costs. Alphabet’s CFO specifically called out higher depreciation and data-center operating costs such as energy as consequences of technical infrastructure investment, with depreciation up 38% in 2025 and expected to accelerate in 2026. The central equity debate is therefore not whether demand exists; it is whether Google can convert backlog and AI usage into durable returns on $175B-$185B of annual capex without triggering pricing compression, utilization volatility, or material energy-cost inflation. (Alphabet Investor Relations) Cloud is also becoming a larger share of Alphabet capital allocation. Kurian said Cloud is about 50% of Alphabet’s capital and growing because Cloud is growing faster. Reuters also reported that Pichai reaffirmed $175B-$185B of 2026 capex and said just over 50% of Alphabet’s ML compute investment would be dedicated to the Cloud business. That means Alphabet’s infrastructure cycle is no longer only a defensive investment to support Search and YouTube; it is increasingly an external revenue engine. The positive scenario is that Google becomes a scaled AI infrastructure utility with better-than-peer cost structure. The negative scenario is that capex grows faster than monetizable demand, depreciation rises before revenue recognition, and competitive pressure from Microsoft, Amazon, CoreWeave, Oracle, xAI, and sovereign-cloud providers compresses returns. (Reuters) WORKLOAD EVOLUTION: FROM CHAT TO MEDIA TO AGENTS The most important technical roadmap signal in the interview is Kurian’s 3-phase framing of model workloads. Phase 1 was chatbot/search-style Q&A, where prompts were often long and output responses relatively shorter. Phase 2 was multimodal content generation, where simple prompts could generate long image, audio, or video outputs, increasing output-token intensity and stressing generation latency. Phase 3 is agents, where models interact with CRM, ERP, supply chain systems, browsers, code interpreters, APIs, databases, and computers over long-running workflows. This is a fundamentally different compute pattern because the model is no longer only producing a response; it is maintaining state, invoking tools, preserving memory, executing steps, reading and writing data, and coordinating other agents over potentially 6, 7, or 12 hours. This workload shift explains the 8t/8i split. Training infrastructure wants the largest possible compute pools, massive memory, deterministic scaling, high bisection bandwidth, checkpoint resilience, and high goodput. Agentic inference wants low latency, high concurrency, KV cache residency, memory bandwidth, CPU orchestration, local storage, low-cost sandboxing, and geographically distributed inference capacity. Google Cloud’s official description of the agentic era is consistent with Kurian’s interview: a single intent can trigger a chain reaction in which a primary agent decomposes goals into tasks for specialized agents that collaborate, preserve state, and use reinforcement learning to deliver outcomes. (Google Cloud) The adoption metrics indicate that agentic AI is moving from proof-of-concept into production. Google Cloud disclosed that nearly 75% of Cloud customers use its AI products, 330 customers each processed more than 1T tokens over the prior 12 months, 35 customers reached the 10T-token milestone, and first-party models process more than 16B tokens per minute via direct API usage, up from 10B last quarter. Gemini Enterprise paid monthly active users grew 40% quarter-over-quarter in Q1, and Google highlighted production deployments across GE Appliances, KPMG, Macquarie, Citi, Signal Iduna, ASCO, Virgin Voyages, Unilever, and others. (Google Cloud) The key inference is that token growth will not be linear with human usage. Agents multiply compute intensity per human request because a single request can invoke 10s, 100s, or 1,000s of intermediate steps. A travel-planning agent, code-repair agent, procurement agent, or SOC remediation agent can use accelerators for reasoning, CPUs for tools and VMs, SSD for local state, object storage for retrieval, network bandwidth for API calls, and identity/security systems for authorization. This creates a broader semiconductor and infrastructure demand basket than the 2023-2024 “GPU = AI” framing implied. TPU ROADMAP: THE 8T/8I SPLIT VALIDATES INFRASTRUCTURE SPECIALIZATION TPU 8t is Google’s training-focused system. Google states that a single TPU 8t superpod scales to 9,600 chips, provides 121 exaflops of compute, and includes 2 PB of shared high-bandwidth memory. Google also says TPU 8t delivers nearly 3x higher compute performance than prior generations, doubles ICI bandwidth, and can turn months of training into weeks with 1M+ TPU chips in a single logical cluster orchestrated by JAX and Pathways. The architecture is built for frontier model development, embedding-heavy workloads, large-scale pretraining, and near-linear scaling. (Google Cloud) TPU 8i is Google’s inference and reasoning system. Google states that TPU 8i includes 288 GB of HBM and 384 MB of on-chip SRAM, with the SRAM explicitly sized for KV cache footprints in reasoning models at production scale. It doubles ICI bandwidth to 19.2 Tb/s, reduces network diameter by more than 50% through Boardfly, uses a Collectives Acceleration Engine to reduce on-chip latency by up to 5x, and delivers 80% better inference performance per dollar versus the prior generation. This design targets high-concurrency reasoning, MoE models, chain-of-thought processing, reinforcement learning, and multi-agent workflows. (Google Cloud) The inference is that Google sees inference no longer as a byproduct of training chips, but as a separate economic domain large enough to justify custom silicon. That is a significant semiconductor signal. Historically, accelerators were evaluated mainly on training FLOPs and memory capacity. Agentic inference shifts the metric set toward tokens per watt, time-to-first-token, tail latency, KV cache efficiency, SRAM capacity, HBM bandwidth, interconnect hop count, utilization, and system-level goodput. Google’s 8i architecture is effectively a statement that the AI inference market is becoming sufficiently large, latency-sensitive, and memory-bound to support distinct product families. The 8t/8i split also implies that future AI capex will be less homogeneous. Training clusters will remain extremely dense, liquid-cooled, network-intensive, and site-concentrated. Inference clusters will be more distributed, more latency-sensitive, more utilization-sensitive, and more dependent on CPU and storage orchestration. Kurian’s comment that 8i can run in non-water-cooled mode is strategically important because it suggests Google wants inference deployability in a wider range of existing data centers. The ability to place inference capacity closer to users, exchanges, enterprises, and sovereign jurisdictions can reduce latency and expand addressable deployments beyond mega-campus training sites. GPU IMPLICATIONS: NVIDIA DEMAND REMAINS STRONG, BUT CUSTOM ASIC SHARE GAINS ARE REAL The interview should not be read as anti-Nvidia. Google has a dual-track strategy: maintain Nvidia access for customers and workloads that prefer CUDA, while expanding TPU adoption where Google can offer better cost, latency, or energy efficiency. Google Cloud’s Next ’26 announcements explicitly state that Nvidia GPUs are a core part of the AI accelerator portfolio and that Google will be among the 1st to offer Nvidia Vera Rubin NVL72, in addition to Blackwell and Hopper-based instances. Virgo Network will also support A5X powered by Nvidia Vera Rubin NVL72, with Google saying it can support up to 80,000 GPUs in a single data center and up to 960,000 GPUs across multiple sites. (Google Cloud) The Nvidia implication is nuanced. Near-term GPU demand remains supported by frontier training, CUDA inertia, customer portability, open-source ecosystem maturity, Nvidia’s full rack-scale roadmap, and hyperscaler desire to offer GPU choice. However, the long-term risk is mix shift, not demand collapse. Captive hyperscaler workloads and partner labs can increasingly move toward custom ASICs when model architecture stabilizes, software layers mature, and economics justify porting. Google’s native PyTorch support for TPUs, optimized vLLM support across GPUs and TPUs, and bare metal access are strategically important because they directly address the biggest adoption friction: CUDA lock-in and developer workflow inertia. (Google Cloud) The investment read-through is that Nvidia remains a core beneficiary of the AI capex cycle but may face relative share and margin pressure in workloads where hyperscalers can substitute internal ASICs. The more standardized inference becomes, the more attractive ASIC optimization becomes. The more frontier training shifts to novel architectures, dynamic kernels, and research-heavy experimentation, the more valuable Nvidia’s generality and software ecosystem remain. Google’s roadmap therefore creates a barbell: GPUs dominate flexible, broad, developer-led workloads; TPUs gain in high-scale, repeatable, cost-sensitive, Google-integrated workloads. CPU IMPLICATIONS: AGENTS RE-ACCELERATE GENERAL-PURPOSE COMPUTE DEMAND The interview contains a critical but underappreciated CPU point. Kurian said that agent computer use informed Google’s CPU strategy because an agent operating a computer is still using traditional compute. Agents need CPUs for tool calls, browser use, sandboxed VMs, API orchestration, data preprocessing, reward calculation, code execution, visualization, identity checks, logging, security scanning, and workflow state management. This means AI does not eliminate CPU demand; it changes its role. CPUs become the control plane and execution substrate around accelerator-driven reasoning. Google’s official infrastructure announcement reinforces this. Google said GPUs and TPUs must be complemented by high-performance CPU services for complex logic, tool calls, and feedback loops around the core AI model. Axion-powered N4A CPU instances are positioned for agent runtimes, while 4th-generation Google Compute Engine VM families powered by Intel and AMD are optimized for RL reward calculation, agent orchestration, and nested visualization. Google also said Axion N4A provides up to 30% better price-performance than agent workloads on other hyperscalers, while a separate Next ’26 keynote transcript says Axion N4A delivers 100% better price-performance than comparable x86 instances for sustained agent operation. (Google Cloud) The key inference is that agents are likely to create a CPU shortage or at least a CPU demand renaissance in cloud. The accelerator is the expensive part of the system, but the agent runtime can bottleneck on VM spin-up, sandbox density, tool latency, CPU memory, local disk, and network egress. Kurian explicitly identified consumer VM economics as the next major bottleneck: consumers cannot afford VMs running indefinitely, so infrastructure needs rapid activation/deactivation, local storage, and oversubscription models. This points to a new category of infrastructure competition around serverless agents, secure sandboxes, fast cold starts, low-cost local disk, and CPU utilization management. For Intel and AMD, this is a constructive but mixed signal. The positive is that agentic AI increases general-purpose compute consumption alongside accelerators. The negative is that Google is aggressively moving to Axion Arm CPUs for internal optimization and margin capture. Intel and AMD remain relevant for broad enterprise workloads, x86 compatibility, RL, orchestration, databases, and network-heavy instances, but Arm share gains inside hyperscalers are likely to continue wherever software portability and cost targets permit. NETWORKING IMPLICATIONS: INTERCONNECT IS NOW A PRIMARY BOTTLENECK Networking is moving from supporting infrastructure to strategic differentiation. TPU 8t requires massive training scale, while TPU 8i requires low-latency all-to-all communication for MoE and reasoning. Google’s Virgo Network is designed as a collapsed AI fabric with 4x the bandwidth of previous generations, and Google says Virgo can connect 134,000 TPU 8t chips into a single fabric in 1 data center and more than 1M TPUs across multiple data-center sites into a single training cluster. Virgo will also support Nvidia-based A5X, with up to 80,000 GPUs in 1 data center and up to 960,000 GPUs across sites. (Google Cloud) The key technical point is that AI networking requirements are diverging by workload. Training needs high bisection bandwidth, deterministic latency, resilient checkpointing, and cross-cluster scaling. MoE inference and reasoning need low-latency all-to-all communication, reduced hop count, fast collectives, and predictable tail latency. Agentic workloads add another network layer because agents call tools, APIs, databases, storage, other agents, and enterprise SaaS systems. A single human request can fan out into many networked tasks, making network topology, congestion control, routing, and gateway design major levers of cost and user experience. This is structurally positive for optical components, high-radix switching, co-packaged optics, 800G/1.6T interconnect, NICs, DPUs, Ethernet fabrics, optical circuit switching, RDMA, retimers, and high-speed cabling. It is also strategically important for Google because network co-design can reduce dependence on merchant networking stacks. Google’s ability to optimize network layers for TPUs and its willingness to make Virgo available for Nvidia systems suggest that networking can become a cloud differentiation layer, not only a data-center cost item. MEMORY, DRAM, AND HBM: THE MEMORY WALL IS THE CORE SEMICONDUCTOR CONSTRAINT The 8t and 8i disclosures make clear that AI is increasingly memory constrained. TPU 8t’s 9,600-chip superpod with 2 PB of shared HBM implies roughly 216 GB of HBM per chip. TPU 8i’s 288 GB of HBM and 384 MB of on-chip SRAM show that inference and reasoning are being optimized around memory bandwidth, KV cache residency, and reduced data movement. The phrase “memory wall” is not marketing; it is the bottleneck that determines whether expensive accelerator FLOPs sit idle. (https://t.co/AUBGFz9nBz) HBM demand is therefore a direct beneficiary of Google’s TPU roadmap. TrendForce estimates HBM demand grew more than 130% YoY based on 2025 AI chip shipments and expects HBM consumption to rise by more than 70% YoY in 2026, driven by Nvidia, AMD, Google TPU, and AWS Trainium moving toward HBM3e. Reuters reported that SK Hynix said client requests for HBM over the next 3 years already far exceed production capacity, while DRAM contract prices jumped nearly 83% sequentially in Q1 2026 and some NAND prices rose around 160%. (TrendForce) The Anthropic-Google TPU deal illustrates the magnitude of memory demand. Anthropic announced access to up to 1M Google TPUs, worth 10s of B dollars and expected to bring well over 1 GW of capacity online in 2026. If future TPU deployments carried memory content comparable to TPU 8t or TPU 8i, a 1M-chip fleet would imply 216 PB-288 PB of HBM-class memory content on an illustrative basis, before host DRAM, SSD, networking buffers, redundancy, and spares. The exact mix and generation are not disclosed, so this is a scale illustration rather than a contract specification. (Anthropic) The DRAM impact is broader than HBM. Agentic inference requires host DRAM for CPU runtimes, VM sandboxes, tool execution, databases, local caches, retrieval systems, security logs, and orchestration. Google’s TPUDirect RDMA and TPUDirect Storage reduce host CPU and DRAM bottlenecks by moving data directly between TPU HBM, NICs, and storage, but they do not eliminate system memory demand. They shift the high-value bottleneck toward HBM bandwidth and direct data paths while still expanding overall data-center DRAM needs. The likely beneficiaries include HBM suppliers, DRAM suppliers, advanced packaging, TSV, CoWoS-like capacity, HBM test, memory controllers, and packaging equipment. The risk is that memory inflation becomes a server BOM headwind for cloud providers and AI labs. STORAGE: SSD, HDD, OBJECT STORAGE, LUSTRE, AND LOCAL DISK ALL MATTER Storage is becoming a 1st-order AI bottleneck. Training workloads need data ingest, checkpointing, model restore, multimodal corpus access, and failure recovery. Inference workloads need model weight loading, retrieval, KV cache tiers, prompt context, vector search, logs, and state. Agents add local disk because VMs and sandboxes need fast read/write storage for browser use, code execution, files, intermediate artifacts, and tool outputs. Kurian’s named “next bottleneck” around consumer VMs and local disk is therefore highly material for storage vendors. Google’s storage announcements are directly aligned with this bottleneck. Managed Lustre now delivers 10 TB/s of bandwidth, a 10x improvement versus the prior year and up to 20x faster than other hyperscalers, with capacity increased to 80 PB. Rapid Buckets on Google Cloud Storage offer sub-millisecond latency and 20M operations per second for checkpoints and recovery, with a target of maintaining 95%+ accelerator utilization. Z4M instances scale to 168 TiB of local SSD capacity and can be deployed in RDMA clusters of 1,000s of machines. TPUDirect Storage allows direct data movement between accelerators and high-speed managed storage, bypassing the host and reducing CPU bottlenecks. (Google Cloud)
@coreyagonzalez I mentioned 30 US names that I liked recently. A lot of my thesis posts talk about Japanese or Taiwanese names, but I talk about US stocks a lot like $AMSC, $ARM recently or stuff like $NBIS and $RDDT?
Semiconductors are on fire in premarket — this is what momentum expansion looks like. $INTC leading with a massive +30% move, signaling aggressive buying and possible short squeeze. $AMD up 10%+, continuing to show leadership in AI and data center space. $ARM gaining 8%+, strength across chip design names. $MRVL and $TSM both up ~4%, confirming broad sector participation. This is not just a bounce — this is sector-wide momentum driven by AI demand and FOMO chasing. When leaders and laggards move together, institutions are buying the whole space. Stay alert — volatility will be high, but this is where big trends start forming.
Just putting out there... Would have been +15.02% in 2W equal-weighted return. On 30 different stocks... mostly medium-large cap. 1. $INTC +29.62% 2. $MRVL +40.95% 3. $TSM +4.72% 4. $COHR +18.9% 5. $RKLB +26.76% 6. $DRAM +12.29% 7. $AVGO +18.32% 8. $AMZN +9.17% 9. $ARM +36.6% 10. $TSEM -1.25% 11. $IBIT +7.68% 12. $NBIS +15.22% 13. $GOOGL +6.41% 14. $AMKR +32.25% 15. $HOOD +19.14% 16. $CRCL +17.58% 17. $META +4.9% 18. $LITE -5.28% 19. $LPTH +20.23% 20. $FN +11.54% 21. $JBL +15.45% 22. $MP +17.48% 23. $HIMS +42.53% 24. $SMTC +18.83% 25. $POWL +9.26% 26. $VPG +17.44% 27. $MOG.A -3.96% 28. $MSFT +11.44% 29. $CVX -1.47% 30. $XLU -2.29% Obviously short timeframe, but I expect many of these to keep going up more. And probably would have been higher if you time the drop on specific names, rather than going long all at once. Not too shabby?
Everyone’s out there sharing new YTDs after $MRVL, $INTC to $ARM went up… I think I’ll stop sharing mine just to let others feel better?
Bruh it’s only been one day with $ARM. How does a $200B+ company go up 12% https://t.co/4BNSjjk88c
$ARM working and one way to play the CPU demand explosion. I am long 1/21/28 calls. https://t.co/gdHQKL9YKV
If you are a new connection or new to my timeline, I have attended every NVDA GTC in San Jose and Washington, DC since 2023. One of my highest signal takeaways from GTC San Jose last month was the massive indication of CPU demand growth. You have undoubtedly been hearing about this in recent weeks. I am a meaningful consumer of agentic GAI applications, predominantly @openclaw , and I see CPU usage spikes during heavy agentic workloads. The question is, what is the cleanest, highest beta expression to trade? $NVDA $INTC $ARM $AMD
Bullish on $ARM, given the new bottleneck shifting back to CPUs. MS shows stuff like Orchestration/RAG requiring CPUs. But I'm predicting parts of localized inference to be handled by CPUs more and more... as models like Gemma get lightweight in the future. Not every robot needs to be able to solve the mysteries of the universe. Data centers will need an astronomical amount of traditional CPU compute (AWS Graviton, $GOOGL Axion, and $MSFT Cobalt), which are all ARM based. $META + OpenAI are also buyers of the AGI CPU. And AI will flow down to edge. $15B annual revenue target.. Starting to look reasonable?
@beauty_oe $ARM の株価が大きく上がってよかったです！AI向けCPUの予測が凄まじかったので、いずれ株価もそれに追いつくだろうと思っていました。
Frontrunning 1.6T/CPO within the broader photonics supercycle is the most compelling investment to me. I have high conviction in that statement. Which is why I'm long the entire supply chain (+1 extra bottlenecK) 1. $SIVE - Their laser revenue scales aggressively with $JBL, $MRVL, Ayar, O-Net. And I do think CPO/1.6T will blow away any conservative analyst projections from how hard $NVDA, $GOOGL, and others have been pushing photonics architectures. Downside risk is multi-sourcing, but there's a reason Jabil chose Sivers. When you compare $MTSI, $LITE, $COHR, Furukawa, and others. There's genuinely not many laser suppliers in the entire world... they're all $10B+, then you have this mini CHIPS act chokepoint trading at <$1B MC. 2. Shunsin (6451) - I don't see how it's possible Foxconn's optical foundry for testing, packaging, and assembly is valued at $1.5B MC less than $LWLG. When they look extremely derisked piggybacking off of Foxconn's photonics volume. $TSM's optical arm VisEra example is ~$5B, but they scale H2 2028 from Gen-3. Foxconn looks to be ramping up just next year. They're just scaling low fwd p/e multiples off of $NVDA CPO supply chain demand in Taiwan and all public indicators point to capacity expansion + extreme demand. 3. Win Semi - They're the foundry for Sivers to scale up DFB laser production. As well as $AVGO, SpaceX supply chains and others. When I do supply chain mapping and Win Semi pops up in every single frontier supply chain I see. There's probably something markets are not pricing in. 4. $MRVL - I find this genuinely compelling as a mini-Broadcomm. Their potential design with with $GOOGL today, helps the case past 2028. But the catalyst I was looking at was $MSFT Maia ramp, which happens H2 2026, and likely keep scaling up exponentially into 2027, 2028, 2029. Celestial acquisition was probably the smartest thing in the world for them. Maybe on next drop or CSP? 5. $HPS.A - Transformers/Switchgears are commodities + boring parts of the DC supply chain. However, when the bottleneck is 2-5 years, and you have backlog increasing 100%+... causing extreme shortages. It's only up 20%+ since my thesis post, but I do see this being de-risked given massive backlog visibility (even though it's inferred, they don't give exact #). I do think markets are missing something, especially with potential gross margin expansion from price hikes if they pull it off.... Again backlog + demand just de-risks this company, and it seems like a high growth compounder post facility expansion last year. There's many others like $NBIS, $JBL, $RPI, $TSEM, $LITE, $ARM, $SOI, $AXTI, $IQE, $ALRIB, Fittech, PCL, and others that I'm very fond of, but just mentioning 5 off the top of my head from today's prices... if I'm creating a new portfolio. Of course, it's good to barbell with other uncorrelated companies to AI supply chains, but these are just 5 I liked.
Wow, majority of these 30 stocks I’ve liked are up a lot in just two weeks (just a recap to new folks) By the way, my long term opinion doesn’t change on any of them from $MRVL, $AMD, $ARM and others. Short term entry points do though with names like $AAOI to $AEHR. And they make the difference between +10-20%. I focus a lot about the “undiscovered” ones like Riber or $SIVE or $RPI or $IQE in analysis when I make a new entry -> wait for it to play out. But the same thesis around $LITE or $NBIS or $AXTI from last year is still the same. And I don’t need to post that same thesis multiple times, since it’s not new anymore. But the reason they’re not new is because markets have validated the thesis and are repricing the stocks live because of them.
Physical AI play Silicon / Compute (The Brain) $NVDA — Owns the full AI stack $ARM — Royalties on every advanced chip $AMBA — Edge vision processing $QCOM — Low-power AI decision chips $NXPI — Control + sensor fusion backbone Sensing / Memory (Real-Time Awareness) $ADI — Converts physical signals into data $MU — High-speed memory for instant decisions $OUST — Affordable LiDAR enabling scale Materials (The Constraint) $MP — Rare earth monopoly = motor supply control Simulation (Build Before Reality) $CDNS — Every robot is simulated before deployment Proven Monetization $ISRG — Physical AI already generating recurring revenue Deployment Layer (Real-World Execution) $ONDS — Drone + defense infrastructure $SERV — Autonomous delivery at scale Software / Interface (Control Layer) $PATH — Orchestrating humans + AI + robots $SOUN — Voice + vision interface for machines
$ISRG — The Blueprint For How Physical AI Gets Monetized Surgical robotics is Physical AI that already prints recurring revenue. With an installed base exceeding 10,763 systems globally, da Vinci 5 continues gaining momentum with AI-powered force feedback and in-console video replay for real-time surgical decision-making. Every humanoid robotics company is trying to build what ISRG already has. $ONDS — Drones Are Physical AI. And Ondas Is Building the Sovereign Security Grid. Drones are the first large-scale deployment of Physical AI — whoever controls drone swarm technology controls the next generation of warfare and commercial autonomy. Management raised 2026 revenue guidance to $375M — an 840% YoY growth trajectory — driven by Roboteam ground robotics and drone-in-a-box deployments across UAE and Saudi Arabia. Most speculative on this list. Highest upside if execution holds. $PATH — The Orchestration Layer Between Human Workers and Robot Workers Maestro coordinates AI agents, software robots, and human workers across complex enterprise workflows — the central nervous system for autonomous business operations. 90% of U.S. IT executives see agentic potential, 77% plan investments in 2026, but only 37% are live. PATH is where PLTR was in mid-2023. The re-rating hasn’t happened yet. $SOUN — Every Physical AI Machine Needs a Voice. SoundHound Wants to Own That Layer. Robots need to take commands. AVs need to understand context. Machines need ears. At CES 2026, SoundHound unveiled Vision AI for vehicles — uniting visual with voice AI so an in-vehicle assistant can listen, see, and interpret the world simultaneously. Named a leader in the Aragon Research Globe for Agent Platforms 2026. Speculative — but 400+ patents and 217% revenue growth says the moat is real. $SERV — Jensen Huang Called Them Out By Name. The Robots Are Already Delivering. Not a concept. Not a demo. Serve has deployed over 2,000 robots across the U.S. — the nation’s largest sidewalk delivery fleet — completing thousands of deliveries weekly for Shake Shack, Little Caesars, Uber Eats, and DoorDash. The 2026 acquisition of Diligent Robotics expanded Serve into hospitals — Moxi operates in 25+ U.S. hospitals generating $200K–$400K annual revenue per deployment. Nvidia spotlighted. T-Mobile partnered. Real deployments, real data, real moat forming. Stack Map: → Silicon/Compute: $NVDA $ARM $AMBA $QCOM $NXPI → Sensing/Memory: $ADI $MU $OUST → Materials: $MP → Simulation: $CDNS → End Market: $ISRG → Deployment: $ONDS $SERV → Software/Interface: $PATH $SOUN Not financial advice. DYOR.
Physical AI Playbook-  Wave 1 was digital AI — data centers, GPUs, LLMs. Wave 2 is Physical AI — robots, drones, AVs. $NVDA — The God of Physical AI Every robot OS runs on Nvidia. Jetson Thor delivers 7.5x the performance of its predecessor — already adopted by Amazon Robotics, Boston Dynamics, Figure, and Caterpillar.  Simulation, foundation models, edge compute — Nvidia owns the entire stack. You don’t beat the platform. $ARM — The Invisible Tax On Every Robot Ever Built You’ll never see Arm’s logo on a robot. You’ll pay them every time one ships. Arm’s cores sit inside virtually every chip in edge AI — Nvidia’s Jetson, Qualcomm’s Dragonwing, NXP’s S32, Ambarella’s CVflow. As chips get smarter, Arm captures a larger royalty per unit.  Pure leverage play on the entire Physical AI buildout. $AMBA — Robots Need Eyes. Ambarella Makes Them. When a robot drops a glass, it can’t wait for a signal to go to the cloud and back. Ambarella makes low-power AI chips that process vision on the edge instantly.  Real-time edge vision is a hard engineering problem — Ambarella has the lowest power, highest accuracy solution. No vision, no robot. $QCOM — The Phone Chip Company That’s Actually a Robot Chip Company The market still prices Qualcomm as a smartphone play. Qualcomm’s Dragonwing IQ10 SoC enables low-power decision-making for drones, robots, and AVs beyond the cloud.  When Physical AI re-rating hits this ticker, the gap closes fast. $NXPI — The Unsexy Chip That Every Humanoid Robot Will Need Nobody talks about NXP. They should. In March 2026, NXP announced foundational robotics solutions developed with Nvidia — combining Nvidia’s Holoscan Sensor Bridge with NXP SoCs to enable sensor fusion, machine vision, and precision motor control for humanoid form factors.  NXP’s processors are purpose-built for the zonal architecture shift — from hundreds of isolated control units to a few centralized superchips. These hit volume production in 2026.  Once designed in, stays for a decade. $ADI — The Nervous System No One Is Investing In Every robot needs to feel the world before it can act on it. Analog Devices converts real-world pressure, motion, and temperature signals into data robots can use. ADI stands out for resilient profitability, dividend safety, and growth exposure to robotics.  The boring bottleneck with a blue-chip balance sheet. $MU — Speed Is Life When Your Robot Is Walking A robot walking and making a decision can’t afford a 1–2 second delay — a one-second lag means it drops something, crashes, or hurts someone.  Fast low-power memory at the edge closes that gap. LPDDR5X and HBM are Micron’s answer. Robotics demand here is structural, not cyclical. $MP — No Rare Earths, No Robot Motors. Full Stop. You can’t build a million robots without the magnetic materials that make their motors spin — and the rare earth supply chain is severely constrained. If the Pentagon wants U.S.-made robots, they need MP’s California mine.  The geopolitical moat is the thesis. $OUST — LiDAR Fell 99% in Cost. The Mass Deployment Era Just Started. Ouster is benefiting from a 99%+ cost decline in LiDAR since 2019, helping expand AV and robotics beyond R&D into real-world deployment.  Cheap LiDAR is the unlock — robots finally have affordable spatial awareness at scale. $CDNS — You Can’t Ship a Robot You Haven’t Simulated First Cadence’s $3.18B acquisition of Hexagon’s Design & Engineering business brings industry-standard multibody dynamics simulation tools essential for Physical AI and robotics development.  Every robot company is a Cadence customer before it ships a single unit. Invisible infrastructure, zero competition.
Here's a bunch of random 30 US-available random stocks I like today and why: 1. $INTC - America's hope for foundry, national security 2. $MRVL - scales rev from future maia asics and add ons like cpo, they do everything lost count 3. $TSM - backbone of semis/ai 4. $COHR - They do everything vertically integrated + captures optical cycle 5. $RKLB - the final frontier of space will be around 5 years from now and 20 years from now. 6. $DRAM - memory exposure for samsung/sk hynix 7. $AVGO - hyperscalers dont like nvidia gpu tax 8. $AMZN - nobody can compete against the overnight shipping of toilet paper. robotics will lower opex over time 9. $ARM - AGI CPUs scale revenue quite a bit over the next decade 10. $TSEM - you're going to need a foundry for light based stuff 11. $IBIT - bitcoin, we all know by now 12. $NBIS - i think it's the next AWS. Also they do self-driving cars with uber, own scaling DB companies, data labeling. It's almost like a mini Google. 13. $GOOGL - youtube is not going away, gemini is great. they're vertically integrated with TPUs and fund buildout with operating income so i like it. 14. $AMKR - super facilities coming online in late 2027-2028. benefits from made in america 15. $HOOD - i dont like short term, but long term i'm a fan of Robinhood since they captured retail + have more products like banking, etc that they're scaling up. product innovation is wild. 16. $CRCL - I happen to really like stablecoins and see them as the future for both payments/holding (depends on clarity act) 17. $META - people aren't going to stop using instagram or whatsapp, or others anytime soon. 18. $LITE - $GOOGL TPU exposure decently high part of BOM. As long as Google's AI program keeps running I think $LITE will do well. 19. $LPTH - Germanium and China export controls will always be an issue so US made engineered alternatives will always be important 20. $FN - Someone needs to assemble optical stuff 21. $JBL - same as above, but added with ip from Intel's SiPh acqusition so might end up like innolight? 22. $MP - American rare earths program is extremely important, similar to $INTC national security risks 23. $HIMS - Okay here me out they just acquired a ton of companies, and at $19 they have global DTC channel. short sellers really hate this company, but I think it's actually promising as a contrarian long 24. $SMTC - LRO/LPO transition 25. $POWL - US alternative to hammond for switchgear DC type bottleneck 26. $VPG - Humanoids will be a thing down the road maybe 2027-2028, this makes the sensors. 27. $MOG.A - Feels like i see them everywhere in robotics, to spacex supply chains 28. $MSFT - At $375, one day we'll look back and see this as a buying opportunity. 29. $CVX - oil might crash after war but these oil companies are going to be extremely important, especially when Venezulea is a goldmine. 30. $XLU - i think rate cuts might be back online, we need power/grid for AI so these names will always be improtant from $CEG to $NEE Just throwing out other thoughts aside from $AAOI and $AEHR.
@TD_btc24 Most recent 5 thesis posts I've shared: 1. $HPS.A ($1.77B) - Transformer/Switchgear DC bottleneck 2. $ARM ($152B)- AI CPU ramp 3. Win Semi ($5.7B) - Foundry for CW lasers and other supply chains from SpaceX to humanoids 4. $SIVE ($295M) - CW Laser ramp for H2 2026 and 2027. 5. $TSEM ($22B) - photonics foundry Apart from those, names I've positively mentioned like $MRVL, $AAOI, $RDDT, $NBIS, $RPI, $AEHR, $LITE, $COHR, SK Hynix, $LASR, $SOI, $IQE, and others might be decent additions.
Who is even writing this? We don't need more leverage over our allies like Japan, Europe, or South Korea. We need them over Russia and China. - In SEA over 40% of gas stations in Laos have closed, - Cambodia and Thailand have started rationing and price controls - India, Pakistan and Bangladesh face rising prices and emergency conservation - SK/Japan remain exposed to disruptions. China can get just get their oil from Russia. And Russia got their export control removed... And China is playing arms dealer like what US did in WW1 right now. Why are we screwing over our own supply chains to help Russia/Israel out? They're called allies for a reason, and we don't need more leverage over them. If you really wanted to weaponize this: -> Secure supply chains first, pour funding into rare earths/precursors like Vietnam/South America and other places. -> Pour funding into refinery processing from SK/Japan/Canada/US -> Develop crude/processing in Venezuela First -> Build out alternative trade route: Which will take 3-5 years min. Then you can go blowing stuff up then go threatening European $ASML, Japanese $ARM, US $NVDA, Taiwan $TSM type chokepoints over China/Russia all together. Again: America First relies on: -securing own supply chains in Assembly Thailand, -Semiconductors in South Korea, -Chemical from Japan, -Foundries in Taiwan, -Rare Earths from Canada, - high-end equipment from Europe and others to make United States stronger. Then you weaponize that all together against our enemies. You can't just blow OUR OWN SUPPLY CHAINS and Relations around the world up then say America is becoming more dominant?
@VJNCapital Nope. $SIVE photonics is too small as an upstream optical component supplier so would likely pass. $ARM would raise antitrust due to size.
Faster compounds: $AAOI - 10x revenue ramp from optical transcivers h2 2027 $NBIS - 10x revenue ramp Q4 2026 $ARM - 5x revenue growth from their new AI CPU $MRVL - 2-3x revenue growth from $MSFT Maia Ramp. $AVGO - Long hyperscaler ASIC $LITE - Long OCS / Google TPU Win Semi - Foundry exposure to frontier industries $TSEM - Long photonics, backlogged SK Hynix - Memory exposure, extreme operating income ramp With some barbell exposure away from Hyperscaler capex aside from Amazon: $VNP - Long term rare earths for Western Supply chains $NEO (TCX) - Robotics Supply chains $AMZN - Robotics/AI cutting opex $CRCL - Stablecoin long $RDDT - Ridiculously high profit $GLD - Safe Hedge $IBIT - Halving 2028 $CVX Calls - Oil Hedge And maybe long term (you know it's coming): $INTC / $AMKR- Made in America supply chains $SOI - Silicon Photonics / CPO substrates. $RKLB - Long term call on Space industry Then pick one or two small cap moonshots: $SIVE - CW Laser Chokepoints or $IQE for Landmark rerating on restructuring were my two favorites. There's others I've mentioned like $AEHR for testing or $VPG for Optimus. How I actively manage my own stuff from $AXTI and others is a lot different risk profile than what others should do. Going full port into high-beta in this macro environment is not the best idea.
@gude57856 I would personally not just invest in one sector for diversification sake. I talk about photonics like $AAOI recently because I see it as highest short-term upside. But others like $NBIS as Jensen accurately said "will take care of you" long term. Maybe figure out high growth longs for example: $ARM - 5x revenue from new AI CPU $NBIS - 10x revenue to q4 $7-9B ARR ramp $AAOI - 10x revenue ramp from optical transceiver demand $MRVL - 2-3x revenue ramp from $MSFT Maia ASICs Pick one or two moonshots: I mentioned $SIVE as my favorite, but given it's small size, I wouldn't put too much concentration into them and then barbell with some "safer plays" $AMZN long term I'm bullish on even from robotics/AI cutting opex though it moves like a slug $RDDT long term I'm bullish on from on just because it's ridiculously high profit and generating massive FCF today. and maybe some "long, long term players" that have deep national security benefits eg: $INTC for Made in America $AMKR for Made in America, etc. Just a made up example.
@RiskAdjustedMe Tbh I like all of the names I’ve mentioned. Win Semi and $ARM were the newest ones out of the bunch.
Google's TurboQuant... And it's effect on $SNDK, $MU, SK Hynix, and others: What it does: -> 6x reduction in KV cache memory footprint -> 8x Speedup on H100 GPUs It's a compression algorithm. Now... Will it beat down memory? -> Prob not. Implications might be bullish for $ARM and others though where you can run AI locally, rather than DRAM heavy DCs. However: ->This is basically DeepSeek round 3. You can make algorithms more efficient. But that doesn't replace either memory or GPUs. -> It could structurally (and slightly) reduce DRAM demand. -> think it's only been tested on small models so far like Gemma, Mistral, and Llama-3.1 (and paper's been out for a year) Also, markets conflated DRAM with NAND... this algo compresses the KV cache (DRAM). Doesn't do anything to NAND storage? Regardless: Algorithms will always get more efficient. People keep saying Jevons Paradox, which is true since this just scales use cases. Main thing to look out for is hyperscaper capex projections, not Google Algorithms that made things more efficient. Feels more like a narrative headwind than anything material to earnings.
@Jornka329996 Thanks! $SIVE is high conviction for me personally. Even so at these levels around ~$400m MC. $ARM and $AAOI seems promising long term at these levels too! I did have two more large cap companies I'm looking at this week.
I'm telling you guys... Even if I buy small cap $170 Billion dollar companies like $ARM. They just go up 20% like $TSEM or $SIVE the next day as well? https://t.co/ULZkv65d27
I agree, and how will the stock respond? I feel like $ARM can take out the all-time high of ~$185 on this TAM expansion over the next two years. Or you simply buy $NVDA here??
@Rationalmind__ Bro $ARM is a $140B MC company
Looking to take down $ARM 1/21/28 calls. ARM looks increasingly well positioned in the GAI CPU stack, with real architecture momentum building as x86 loses ground in more conversations. ARM has multiple ways to win here: potential upside from its own silicon efforts and royalty leverage through third-party CPUs like $NVDA Vera. The bigger question is supply. It’s hard to see how $TSM keeps up with all of this demand without a major capacity ramp—Arizona is starting to look like one giant semiconductor complex. That setup should be a clear positive for $AMKR .
$ARM EXECUTIVE OVERVIEW Arm’s March 24, 2026 launch of the Arm AGI CPU is best understood as a business-model transition, not merely a product introduction. The company disclosed its 1st production silicon product and simultaneously set a 5-year objective of roughly $15B of annual chip revenue, roughly $25B of total revenue, and more than $9 of non-GAAP EPS. Against a current revenue run-rate of about $5B, that target implies roughly 38% compound annual growth over 5 years. Management also indicated that meaningful AGI CPU revenue is not expected until FYE28, which means the market reaction is capitalizing long-duration earnings optionality rather than near-term reported results. Strategically, the move is coherent because the CPU layer in AI infrastructure is becoming more valuable as inference becomes persistent, multi-agent, and power-constrained. Financially, however, the bridge is unusually steep and should be treated as an aggressive target path rather than a de-risked base case.  The most important contextual point is that Arm is making this move from a position of strength, not distress. In Q3 FYE26, revenue increased 26% YoY to $1.242B, non-GAAP gross margin was 98.3%, and non-GAAP operating margin was 40.7%. On the February 2026 earnings call, management stated that Neoverse had surpassed 1B deployed cores, that cloud AI data-center royalty revenue continued to double YoY, and that Arm share among top hyperscalers was expected to reach 50%. The launch therefore is not a rescue of a weakening IP model. It is an attempt to capture a materially larger share of the economics in a part of the stack where absolute dollar pools have expanded far beyond handset-era royalty streams.  PRODUCT AND WORKLOAD POSITIONING Technically, the disclosed part looks like a purpose-built AI host CPU rather than a generic enterprise server CPU. Arm disclosed up to 136 Neoverse V3 cores, Armv9.2, dual 128-bit SVE2 units per core, up to 3.7 GHz boost, TSMC 3 nm or N3P manufacturing, 12 DDR5 channels at up to DDR5-8800, more than 800 GB/s of memory bandwidth, 96 PCIe Gen6 lanes, CXL 3.0, sub-100 ns memory latency, and a 300W TDP. The company also described dense reference platforms with 8,160 cores per air-cooled rack and 45,000+ cores per liquid-cooled rack. That specification set is optimized for control-plane scheduling, data preprocessing, API and task hosting, networking and storage services, and accelerator orchestration across heterogeneous AI systems. The product is therefore aimed at a genuine bottleneck in AI data centers: feeding, coordinating, and balancing increasingly large accelerator clusters within fixed power envelopes.  The strongest claims around the part should still be treated as provisional. Arm says the AGI CPU can deliver more than 2x performance per rack versus x86 and up to $10B of capex savings per GW of AI data-center capacity, but those are company estimates rather than independently validated field benchmarks. That does not invalidate the thesis, because power density and performance-per-watt are the correct axes for this category. It does mean the quality of proof will depend on third-party deployment data, OEM benchmarking, and customer testimony over the next 2 to 4 quarters. The near-term de-risking point is that test silicon has already returned and is functioning, early systems are available now, and both Arm and Reuters said broader availability is expected in 2H 2026. That makes the launch materially more credible than a pure roadmap reveal.  The architectural logic behind the launch is also internally consistent. Arm’s public thesis is that agentic AI raises CPU intensity because CPUs remain responsible for coordination, data movement, I/O, storage, security, and latency-sensitive orchestration while accelerators generate tokens. In that framing, the AGI CPU is not trying to replace GPUs; it is trying to become the preferred CPU control layer around them. That distinction matters. It enlarges the addressable market and reduces direct overlap with the accelerator vendors that Arm still needs as ecosystem partners. It also explains why the chip is being marketed to use-cases such as accelerator management, control-plane processing, and cloud or enterprise API and task hosting rather than as a monolithic replacement for the existing server CPU market.  DEMAND AND TAM At the industry level, the demand backdrop is large enough to support a meaningful new merchant CPU franchise if the product performs as advertised. Arm argues that agentic AI data centers may require more than 4x the current CPU capacity per GW. That directional claim aligns with the capital budgets now being deployed across AI infrastructure. Meta has guided 2026 capital expenditures of $115B to $135B. Alphabet has guided 2026 capex of $175B to $185B. OpenAI, Oracle and SoftBank have described Stargate as approaching 7 GW of planned capacity and more than $400B of investment over the next 3 years. The implication is that the market is not constrained by demand scarcity. The market is constrained by proof of performance, deployment readiness, and the ability to win sockets without forcing customers to build their own CPU instead.  However, the critical distinction is between Arm architecture adoption and Arm merchant-silicon adoption. The architecture thesis is already heavily validated. AWS stated in Arm’s launch materials that the majority of compute capacity it added in 2025 was powered by Graviton. Google’s Axion is a custom Arm-based processor. Microsoft’s Cobalt is a fully in-house Arm CPU and its Cobalt VMs are already generally available. Arm itself said in February 2026 that all major hyperscalers are ramping Arm-based data-center chips and that Neoverse share among top hyperscalers is expected to reach 50%. The merchant-silicon thesis is much less mature because the largest cloud buyers already internalize CPU design. Supportive quotes from AWS, Google and Microsoft validate the ISA, the software ecosystem, and Arm’s data-center relevance. They do not automatically translate into purchase commitments for Arm AGI CPU.  Arm’s own reference architecture also illustrates why only a handful of very large wins could still matter disproportionately. The disclosed air-cooled design targets a 36 kW rack with 30 1U servers and 2 AGI CPUs per server, or 60 CPUs per rack. Applied mechanically, 1 GW of such capacity would correspond to roughly 1.7M CPUs, although real deployments will vary materially by cooling choice, server design, and utilization. That arithmetic does not prove the $15B revenue target, but it does show why the model can become large if AGI CPU becomes a standard host-CPU choice inside multi-GW AI clusters. The revenue bridge is therefore aggressive, but not obviously impossible in a world where only a small number of customers are building at gigawatt scale.  GO-TO-MARKET AND CUSTOMER MIX The initial customer and channel structure is logically chosen. Meta is the lead partner and co-developer. Arm also identified customers including OpenAI, Cloudflare, Cerebras, SAP, SK Telecom, F5, Positron and Rebellions, and channel partners including Lenovo, Quanta, Supermicro and ASRock Rack. Arm’s investor deck says the company is primarily selling AGI CPU to hyperscalers, larger enterprises, and tier-1 server OEMs, and describes the opportunity as serving an underserved customer base. That suggests the 1st commercial lane is not simply to displace AWS Graviton or Azure Cobalt. It is to serve AI neo-clouds, accelerator vendors, sovereign or enterprise deployments, and customers that want fast deployment without funding a multiyear custom-CPU effort. Strategically, the product can function both as a merchant chip and as a common platform that accelerates software enablement and system validation across the broader Arm ecosystem.  Meta’s role is highly valuable, but it should not be over-interpreted. It validates technical relevance and deployment seriousness. It does not imply exclusivity. NVIDIA stated on March 16, 2026 that Meta is also collaborating to deploy Vera, NVIDIA’s own purpose-built CPU for agentic AI. Reuters also reported that Arm plans additional AGI CPU designs at 12-18 month intervals. The correct inference is that the host-CPU category around AI accelerators is becoming strategically important and likely to be multi-sourced. A single design win is not enough. The business model requires repeat generations, sustained competitiveness, and proof that the installed base expands beyond a small number of marquee customers.  COMPETITIVE DYNAMICS Competition is broader and tougher than a simple x86-displacement narrative suggests. Intel positions Xeon 6 as a host-node CPU for AI systems and disclosed that Xeon 6 is being used as the host CPU in NVIDIA DGX Rubin NVL8 systems. AMD positions EPYC 9005 as a leading CPU for AI and offers high-frequency host-CPU variants for GPU-enabled servers, with published claims of better time-to-first-token and higher throughput than comparable Xeon-based systems in selected tests. NVIDIA has now launched Vera, a custom 88-core Arm CPU with claims of 2x efficiency and 50% faster performance than traditional rack-scale CPUs, and is coupling Vera tightly with its GPU stack via NVLink-C2C. Arm is therefore not only attacking legacy x86 incumbents. It is entering a market where Intel, AMD and NVIDIA all regard the AI host CPU as strategically important.  Arm’s differentiation is openness, density, and ecosystem breadth. Unlike NVIDIA, it does not require a closed accelerator stack. Unlike Intel and AMD, it can leverage the growing software and standards base already created by Graviton, Axion, Cobalt, Grace, BlueField and other Arm infrastructure deployments. That is real strategic leverage. The offset is that neutrality can also mean less lock-in. NVIDIA can use the CPU socket to reinforce an end-to-end AI factory platform. Intel can bundle the CPU into entrenched x86 workflows and enterprise compatibility. Large cloud buyers can continue to use Arm IP while preferring their own custom chips over Arm merchant silicon. That is the central competitive tension: Arm’s ecosystem success does not automatically convert into AGI CPU share.  ECONOMIC MODEL The economic rationale is straightforward and credible. Arm’s investor deck explicitly compares illustrative gross profit dollars on a $1,000 chip ASP: about $50 from IP, about $100 from CSS, and about $500 from selling the full chip, while also stating that the illustration is not representative of actual AGI CPU pricing. Management’s FYE31 framework calls for $10B of IP or CSS revenue at more than 65% non-GAAP operating margin and $15B of chip revenue at more than 30% non-GAAP operating margin. On those minimum thresholds alone, the framework implies more than $11B of non-GAAP operating profit before any upside from margins above the stated floors. This is the heart of the upside case. Arm is knowingly exchanging cleaner percentage margins for far larger absolute profit pools.  The key question is whether the existing IP and royalty engine can remain intact while the chip business scales. Current economics remain exceptional because the business is still overwhelmingly IP. In Q3 FYE26, Arm generated 98.3% non-GAAP gross margin and 40.7% non-GAAP operating margin. At the same time, its long-range model assumes the IP or CSS business still grows to $10B by FYE31 while chip revenue is additive rather than cannibalistic, and assumes mid-teens opex CAGR from FYE26 to FYE31 after a period of heavy engineering investment. The burden of proof is therefore 2-fold: first, that AGI CPU becomes a real merchant business; second, that the launch does not slow the existing licensing and royalty machine, which today remains the cleaner and more predictable earnings stream.  Financing the push does not appear to be the primary issue. As of December 31, 2025, Arm had $2.807B of cash and $735M of short-term investments, or about $3.54B in immediately available liquidity, and it reported trailing 12-month non-GAAP free cash flow of $893M. The larger issue is business-model complexity. Merchant silicon introduces foundry dependence, packaging and test execution, customer qualification cycles, warranty exposure, supply commitments, inventory, and working-capital needs that do not exist in a royalty-only model. Reuters reported that TSMC will manufacture the part on 3 nm, which is strategically positive but also means Arm now takes on a portion of the operational risk chain previously borne by its customers. The company can fund the effort. The more difficult question is whether it can execute it while maintaining the economics and relationships of the legacy model.  RISKS, RELATIONSHIPS AND GOVERNANCE Channel conflict is the central strategic risk, and Arm has already disclosed it explicitly. In its FY2025 annual report, the company warned that CSS, chiplets, complete end-chip solutions and other more integrated compute products may create real or perceived conflicts with important customers and partners, potentially causing them to reduce or terminate relationships and seek alternatives. That warning should be taken seriously because customer concentration remains high. Arm said its top 5 customers, including Arm China, represented approximately 56% of FY2025 net revenue. The company is therefore moving up the stack toward its customers’ profit pool while still depending on those same customers for the majority of its existing economics. That is the core reason the $15B chip target should not yet be treated as a base case.  There is also a credible mitigating case. Arm’s investor deck states that AGI CPU revenue is additional to CPU IP and CSS license and royalty revenue and explicitly frames the chip business as complementary to the existing CPU IP or CSS franchise. That can be true if Arm primarily serves customers that do not have hyperscaler-scale custom silicon programs, want rapid deployment, or value a common reference platform more than bespoke optimization. If the product stays concentrated in that lane, the company could widen the ecosystem without materially undermining its annuity streams. If it starts competing for sockets that otherwise would have been served by customer-designed Arm silicon, the conflict risk increases sharply. The dividing line between those 2 outcomes is not yet visible.  China is a possible upside option, but it is low-quality and should be discounted heavily until regulatory clarity improves. Reuters reported that management believes the current product likely would not trigger U.S. export controls. However, BIS has repeatedly tightened and revised its framework around advanced computing semiconductors, including January 2025 actions to strengthen restrictions and a May 2025 rescission of the prior AI Diffusion Rule while simultaneously strengthening chip-related controls. Combined with Arm’s existing revenue concentration that includes Arm China, any future China contribution from AGI CPU should be treated as optional rather than embedded. The policy regime remains too fluid for confident underwriting.  SoftBank’s ecosystem both helps and complicates the analysis. SoftBank completed the acquisition of Ampere in November 2025 and had already acquired Graphcore, while OpenAI and SoftBank also funded SB Energy and are participating with Oracle in the Stargate buildout. Those relationships can create adjacent demand, especially where OpenAI or SoftBank-controlled infrastructure wants an Arm-based host CPU quickly. However, Ampere itself is an Arm-based server CPU company. The ecosystem therefore contains both synergy and overlap. The cleanest interpretation is that SoftBank is assembling multiple layers of AI infrastructure and is willing to tolerate internal overlap where it accelerates platform reach. Even so, product boundaries and internal channel governance deserve monitoring because the portfolio is no longer cleanly segmented.  VALUATION AND CONCLUSION At the latest available quote, Arm traded at $134.96 after reaching an intraday high of $145.99 on March 25, 2026 UTC, indicating some retracement after the initial reaction. Using the 1.062B issued shares outstanding reported as of December 31, 2025, the equity value is roughly $143.3B. Based on the company’s implied FYE26 non-GAAP EPS of about $1.75, derived from 9-month non-GAAP diluted EPS of $1.17 and Q4 guidance midpoint of $0.58, the stock trades at about 77x current-year earnings. Using only the company’s minimum FYE31 EPS target of more than $9, the multiple compresses to roughly 15x on a 5-year-out basis, and the current market value is about 5.7x the company’s FYE31 revenue target. The market is therefore not assuming certainty, but it is already capitalizing a meaningful portion of the long-range bridge.  The stock implication is that proof velocity now matters more than narrative quality. If 2H 2026 systems ship in visible volume, if Arm can disclose repeat design wins beyond the initial customer set, and if the IP or CSS base continues compounding without visible partner backlash, the current premium can remain supported because the out-year earnings bridge is very large. If the product instead proves to be mainly a useful reference platform with limited merchant volume, or if it creates friction with major licensees, the valuation becomes vulnerable because near-term earnings still reflect an IP company while the equity increasingly discounts a silicon company. The announcement expands upside, but it also widens the distribution of outcomes.  The balanced judgment is therefore favorable on strategy and still cautious on underwriting. The strongest positive point is that Arm is attacking a real and growing bottleneck in AI infrastructure from a position of architectural strength, not from weakness. The strongest negative point is that it is attempting a difficult migration from pure IP monetization to merchant silicon while selling into a concentrated customer base that already licenses its IP and, in many cases, already designs its own Arm CPUs. The appropriate base case is that AGI CPU raises Arm’s long-term earnings ceiling and improves the strategic quality of the story, but that the full $15B chip target should be probability-weighted rather than capitalized at face value until 1) 2H 2026 shipments convert into visible revenue, 2) the 12-18 month roadmap produces a credible 2nd-generation part, 3) customer concentration broadens beyond a handful of names, and 4) the existing royalty trajectory remains intact. That is a strategically important positive development with a still-open execution question, not a fully de-risked financial transformation. 
@beauty_oe SKハイニックスやサムスン以来、大型株でこんなにワクワクしたのは久しぶり! $ARM のAI CPUの収益拡大予想、マジでヤバいね。
@piyush1337 I didn't long $ARM bc of RISC-V cannibalizing their model, but pivot to AI CPUs themself... It's an incredibly massive game changer, esp if you look at revenue projections.
@JOptionEngineer $ARM is a small cap $143B dollar company.
FYI: I went long on $ARM at $139. Genuinely a compelling long at $143B MC as markets shift more from training -> inference. Then $ARM AI CPUs cannibalize the market for inference and $NVDA market share. Especially as LLMs get more lightweight. The projections to $25B/revenue (5x revenue) are already insane to justify risk-reward.
@mi20483980476 Not exactly. I'm just saying $ARM probably deserves to be a re-rated a tad higher if they're 5xing their revenue from a new product line.
$ARM expects $15B in annual revenue from the the AGI CPU: "The company projects that the new chip business will generate over $15 billion in annual revenue" within the next 5 years. 5 times current revenues (~$25B revenue)... Arm probably deserves to be up more than 5% on this news if they're multiplying their revenue with a new product overnight?
@455BAG I'm actually personally pretty bearish on both $ARM and $QLCM. Probably not the best person to ask on $ARM since I did help out RISC-V quite a bit so I'm a little biased. As for Qualcomm... Mediatek is probably better long, especially with their high growth ASIC arm working with $GOOGL.
@GiovandomenicoC If $ARM does $4.67B annual revenue. Then $ARM introduces a new product that does billions in annual revenue. That would be very material. Of course need more time to look into details.
@B8trades I need more time to model in $ARM estimates, just saw the news recently.
$ARM announces new AI chip called "AGI CPU" According to Arm, the company "Expects it to add billions in annual revenue". - 136 Neoverse V3 cores, built on $TSM 3nm process - Custom-built for "agentic AI" workloads with OpenAI and $META as lead customers. $AMD and $INTC have recently received a tailwind from enterprise CPU shortage. You might be wondering: Does ARM Solve the Shortage? Architecturally, yes. Physically, no. Main beneficiary is $TSM, but this is a large tailwind for $ARM moving forward as they pivot from licensing.
THE ROBOTICS STACK EXPLAINED: 1. $NVDA & $QCOM provide the vision processors, & compute silicon to allow robots to process data, & make real time decisions. 2. $GOOGL, $MSFT & $META supply the foundation models to give robots reasoning, perception, & interaction abilities. 3. $PLTR, $ORCL & $MSFT convert massive robot fleet data into actionable intelligence while $PANW & $CYBR secure the systems. 4. $ARM, $SNPS & $CNDS design the chips powering robot brains. 5. $MBLY & $TDY provide the vision systems, cameras, & perception sensors to allow robots to understand the world. 6. $MP supplies rare earth magnets used in electric motors to move robotic systems. 7. $ADI, $ON & $STM provide the analog semiconductors that connect digital AI systems to motors, sensors, & power electronics. 8. $ST, $RRX & $TKR manufacture the bearings, motors, & mechanical components for robots. 9. $HON & $ROK deliver the industrial automation platforms that integrate robotics into factories. 10. $TSLA, $AAPL, $BABA & $AMZN are building competing robotic ecosystems across manufacturing, logistic, and mobility applications. It’s clear the expansion within the robotics layer is wide, & many names are responsible for this sector boom…
I have written a full article on the AI chip supply chain. The supply chain is structured into 4 different phases with 13 layers: 1. Raw Materials: $SHECY, $SUOPY, GlobalWafers, $WAF.DE, $SHWDF, $AXTI, $IQE 2. Manufacturing Equipment: $ASML, $ASM.AS, $AMAT, $LRCX, $KLAC 3. EDA & Core Intellectual Property: $SNPS, $CDNS, $ARM, $RMBS 4. Chip Design: $NVDA, $AMD, $INTC, $QCOM 5. Foundries: $TSM, Samsung Semiconductor, $SMIC 6. Memory and HBM: SK Hynix, Samsung Electronics, $MU 7. Packaging and OSAT: ASE Technology, $AMKR, JCET Group 8. Server and Rack Integration: $SMCI, $DELL, $HPE, Foxconn 9. Networking Silicon: $AVGO, $MRVL, $CSCO, $ANET 10. Photonics and Optical Components: Ayar Labs, $ALAB, $CRDO, $COHR, $LITE 11. Power, Thermal management and Grid: $VRT, $MOD, $NVT, $SU.PA, $IREN, $CIFR 12. Hyperscalers: $AMZN, $GOOGL, $MSFT, $META 13. AI Storage, platforms and Data: VAST data, Weka, NetAPP, $PLTR, Blue Yonder, $KXSCF The article covers it all.
I believe humanoid robots are going to become pervasive in society. Investors are drastically underestimating how big this market, and the leading companies, will grow to. In order to position myself correctly, I asked @cfosilvia to generate a list of companies around the world that I could invest in to profit from this mispriced opportunity. She suggested: $TSLA $ISRG $ROK $NVDA $AMD $MRVL $ARM $CGNX $AMBA $LITE $6324. T (Japan) $PH $NOVT $QS $ALB $FLEX $JBL Silvia then analyzed my personal portfolio to see where I already had some exposure to humanoids, while also suggesting specific names or allocations that could compliment my current portfolio. I don't make any financial decisions without checking with Silvia first. You can try her free: https://t.co/bMI7hLeciU
$GOOGL will become the most valuable company in the world very soon… One of its most notable features is the diversified portfolio which includes: ~ Search & Advertisement ($2T) ~ DeepMind ($900B) ~ Google Cloud ($540B) ~ YouTube ($500B) ~ Waymo ($160B) Not only this, but $GOOGL has assets in: ~ SpaceX (7.5% stake) ~ $ASTS (AST SpaceMobile) ~ $PL (Planet Labs) ~ $ARM (Arm Holdings) ~ $PATH (UiPath) ~ $MTSR (Metsera) This is just a few of the notable names within Googles portfolio. If you want to capitalize off the future, $GOOGL is your top pick…
$ARM CEO Rene Haas said the constraints on memory chips were “the most severe I have seen in at least two decades.” https://t.co/XgBMHFESEx
These 6 layers are responsible for the immense AI boom that we have seen… 1. AI Compute & Chips ~ $NVDA, $AMD, $ASML, $ARM, $AVGO These companies design and manufacture the processors that AI models run on. 2. AI Compute Operators ~ $IREN, $NBIS, $CIFR They operate large scale infrastructure that delivers AI compute capacity. 3. AI Applications ~ $PLTR, $SNOW, $NOW They sit at the top of the AI cycle and turn data into real world value. 4. AI Security ~ $MSFT, $CRWD, $PANW They turn AI systems, data, and infrastructure from cyber threats. 5. Cloud Platforms ~ $MSFT, $GOOGL, $AMZN, $ORCL They turn AI infrastructure into scalable compute capacity. 6. AI Networking and Connectivity ~ $ANET, $MRVL, $CIEN They move massive amounts of data between servers, racks, and data centers so AI systems function. Without these layers AI would not have scaled as rapidly as it did to reach today’s capabilities…
One of the three major licensing customers signed in this Q is from China, and $ARM described the demand in China as "as strong as we've ever seen." Both licensing and royalties are doing well in China, but licensing is a little bit of a bigger driver this quarter. https://t.co/KwwfSH2O4F
I found this part of the $ARM call quite interesting. It seems the revenue contribution of SB is quite significant and have notable increase compared to last Q. https://t.co/mbFJs6xDlx
Huh. This is super super interesting. Someone is trying to expand its dollar content in the data center. $ARM: "DreamBig Semis got a lot of interesting IPs....particularly around the ethernet area and Arm DMA controllers, which are very, very key for scale-up and scale-out networking."
$ARM: "In the data center, access to power has now become the bottleneck. And this is accelerated adoption of Arm's new verse compute platform. which has now surpassed 1 billion CPUs deployed ..... $GOOGL is migrating the majority of their internal workloads to run on Arm Demand for CSS continues to exceed expectations. During the quarter, we signed 3 new CSS licenses, 1 each in smartphone, tablets and data centers"
$ARM CEO Rene Haas: "Unprecedented compute demand has led to our data center new verse royalties to more than double YoY. Licensing revenue rose 56% to $515 million as companies continue choosing Arm to build their next-generation AI products."
+ 3 more
