โ Renewable-powered AI/HPC infrastructure thesis. Small cap, volatile, but meaningful catalysts are beginning to appear.
THE BIGGER THESIS
โ AI models need compute
โ Compute needs GPUs
โ GPUs need power + infrastructure
โ Power is now the scarce asset
The market still underestimates how important MW ownership becomes in an AI-driven world.
The next phase of the AI supercycle may not be won by the apps.
It may be won by the companies controlling:
โก Energy
๐๏ธ Infrastructure
โ๏ธ Compute capacity
This is not just a software cycle anymore.
Itโs a physical infrastructure cycle.
Not financial advice.
3/6. His Q1 2026 13F drops this week.
Based on running his exact framework forward against what changed in Q1.
Hereโs my prediction on what he MOVED.
New position positions:
$COHR โ Coherent corp
In march 2026, nvidia committed $2B to lumentum AND $2B to coherent.
Jensen Huang called them partners for โgigawatt-scale AI factories.โ
Aschenbrenner already owns Lumentum.
He understands the optical networking bottleneck better than anyone.
Coherent is the direct complement and nvidia just locked the supply chain.
If he didnโt add $COHR in Q1, iโd be surprised.
$GEV โ GE Vernova
Transformer lead times are 128-144 weeks right now. Two thirds of planned AI data center capacity is delayed, not because of chips, but because of ELECTRICAL EQUIPMENT.
GE Vernova controls grid-to-chip infrastructure. $52B revenue target by 2028. record backlog. This is exactly the kind of overlooked chokepoint he hunts.
$WULF โ Terawulf, low-carbon, nuclear and hydro-powered mining sites. co-led a $1B round for fluidstack, which has a $6.7B 10-year hosting agreement with terawulf starting 2026. This is exactly the kind of pre-priced structural position he builds.
$OKLO โ Advanced fast reactors for hyperscalers. $2.54B in cash after Q1 fundraise. natural gas is the bridge fuel. nuclear is the destination. watch for a new entry here.
@philipp317 Yeah. And $CIFR. Im well diversified into this sector. We are still in the innings. And some will also fail on deadlines and projects.
But my top 3 is $NBIS$IREN$CIFR. I also like $WULF a lot. But Iโm not in it.
๐คฏDO WE COOL HUMANS OR DO WE COOL COMPUTE?!?
Ha! We are not bullish enoughโฆ
$IREN$NBIS$CIFR$WULF$DGXX$AAOI$SIVE$AMD$NVDA
China is adding a nuclear reactor every two months at this point.
How much power do we need to beat China?
By 2030, 8% of global power will go to powering data centers.
That means we need about 2.000 GW to power these data centers.
Right now the entire U.S. of on grid energy is about 1.200 GW.
If Canada and U.S were responsible for powering 25% of the data center power needed by 2030, we would have to bring online 500 nuclear power plants in the next 4 years.
The last power plant put up in United States took 6 months.
The bottleneck is real.
The race for AGI is real.
Donโt miss it.
-BP
Please note: This is not financial advice.
What I've read this week:
1. Oil prices increase after Iran doubles down on Strait of Hormuz closure
Summary:
Brent crude rose above $107 on Sunday after Iran warned the Strait will return to normal "under no circumstances." Gas averaged $4.10 a gallon, up about 27% since the start of the war.
My take:
The market is pricing a clean resolution. Iran is not offering one. The part nobody is connecting: this oil shock is landing directly on the AI buildout's cost structure, data centers run on electricity, electricity runs on gas, and the margin math on $100-plus oil changes materially for anyone without locked-in power.
2. Dalio warns Warsh against cutting rates amid stagflation risk.
Summary:
Dalio told CNBC the US is "certainly in a stagflationary period" and that cutting rates now would destroy the Fed's credibility. Traders are pricing 100% probability of no move at this week's meeting, with core PCE running at 4.1% annualized.
My take:
Warsh inherits Powell's problem with no clean answer: cut into 3.3% inflation and lose credibility, hold all year into a softening labor market and face political pressure. Supply chains are tightening at their most widespread levels since 2022 and input costs are rising at their steepest rates in three years.
3. IEA warns AI data centre electricity use will triple by 2030
Summary:
The IEA's new report finds AI data centre electricity demand grew 50% in 2025 and is set to triple by 2030. The pipeline of conditional offtake agreements between data centre operators and SMR nuclear projects has grown from 25GW to 45GW in roughly 18 months.
My take:
45GW of SMR commitments sounds like a solution. Not one of those reactors is in commercial operation anywhere in the world. What fills the gap until the 2030s is natural gas, which brings us straight back to article one.
The companies that locked in power before this crunch have a moat the market is still underpricing.
4. In 8 weeks, the Iran war has dented the US economy.
Summary:
Economists expect PCE inflation to hit 4% by year-end, double the Fed's target, with oil prices likely to remain above pre-war levels throughout 2026 even after a settlement. Supply chains, energy costs, and consumer confidence are all moving in the same direction.
My take:
The AI buildout is debt-financed at $660B this year. It assumes cheap energy, falling rates, and uninterrupted supply chains. All three assumptions are now under stress simultaneously. That is not in any return projection I have seen.
As of now, power is still the biggest constraint to AI. It's still worth watching; $CIFR$NBIS$IREN$WULF
-BP
This is the week the AI trade either gets extended or gets questioned.
$GOOG, $AMZN, $META, $MSFT all report Wednesday night.
$AAPL Thursday.
Five companies. One week. Combined market cap somewhere north of $14 trillion.
The market isn't asking whether they beat revenue estimates.
It's asking one question: is cloud capex still accelerating?
Because the entire bottleneck thesis rests on this. If hyperscalers pull back on data center spend on $IREN, $NBIS, $WULF, $CIFR, $APLD even slightly, even in guidance language, EVERY fโฆ.. AI infrastructure name gets re-rated downward.
Watch for:
> Cloud revenue growth rate vs. prior quarter
> Capex guidance language "accelerating" vs. "maintaining" vs. "moderating"
> Any mention of compute shortages, power constraints, or data center lead times
> OpenAI partnership commentary from Microsoft post-restructuring
The tone of five earnings calls this week will set the direction for AI infrastructure for the next 6 months.
This is the most IMPORTANT earnings week of 2026.
-BP
This is not financial advice, always DYOR.
I donโt expect a deal from $IREN around the corner or at earnings.
On the other hand I expect some kind of delays. I also expect to see the a drawdown in the stock due to sentiment.
But letโs see. I could be wrong.
Anyhow. Itโs important to note this most likely will hit the whole sector.
โSeveral U.S. data centers slated for completion in 2026 are at risk of being delayed as strict schedules encounter regulatory friction, supply chain bottlenecks, and the lack of available utility.
According to a report by the Financial Times, major data center projects involving Microsoft, OpenAI, and other tech companies will miss projected deadlines by more than three months.
The estimate is based on data from SynMax, a geospatial data analytics company that uses satellite imaging and AI to deliver real-time insights and predictive analytics on the maritime and energy sectors.โ
Iโm not shocked at all here. And expect to see delays on all project; $NBIS$CIFR$CRWV$IREN$WULF.
The Danish professor and world-leading expert on megaproject management who frequently cites high failure rates in project time and economy is Bent Flyvbjerg.
His research often highlights that 9 out of 10 (90%) of projects go over budget, over time, or fail to deliver on benefits
-BP
Please note: This is not financial advice.
The companies building AI data centers, ranked by market cap:
$CRWV $115.25 - ~$55B
CoreWeave. GPU cloud king. Acquired Core Scientific. The largest pure-play AI compute company on the market.
$NBIS $159.89 - ~$32B
Nebius. $17B+ Microsoft Azure deal. Targeting $7-9B annualized revenue by end of 2026.
$IREN $45.15 - ~$14B
4.5 GW pipeline. $9.7B Microsoft contract. Largest single-site AI data center buildout.
$WULF $20.00 - ~$8.4B
TeraWulf. Zero-carbon AI data centers. Lake Mariner facility. 100% nuclear and hydro powered.
$APLD $28.72 - ~$7.9B
Applied Digital. HPC hosting and GPU cloud. Building 400MW campus in North Dakota.
$HUT $72.30 - ~$7.7B
Hut 8. AI compute + managed infrastructure. Diversified across mining, hosting, and cloud.
$CIFR $18.30 - ~$6.7B
Cipher Digital. HPC and AI data center operator. Expanding capacity across Texas.
$CORZ $19.25 - ~$5.8B
Core Scientific. HPC hosting pioneer. Being acquired by CoreWeave.
$BTDR $11.69 - ~$2.7B
Bitdeer. ASIC chip design + AI compute. Building custom silicon alongside hosting.
$CLSK $11.15 - ~$2.5B
CleanSpark. Mining operations scaling into AI hosting infrastructure.
This space is moving fast. A year ago half these names were Bitcoin miners. Now they're selling compute to Microsoft, Meta, and the hyperscalers.
Bookmark this.
Wall Street talks about the hyperscalers: $AMZN, $GOOG, $META, $MSFT. That is where the headlines are.
But the money flows through the supply chain. Capex is tracking above $1 trillion this year.
That is six times the level of 2022
Here is the full stack, sector by sector, of who actually gets paid:
Power Infrastructure:
$IREN, $CIFR, $NBIS, $WULF
The compute layer requires power first. These are the companies building and owning the megawatt-scale facilities that hyperscalers cannot build fast enough themselves.
Optical Networking:
$AAOI, $COHR, $LITE, $POET
GPU clusters are useless without the interconnects that move data between them at 800G and 1.6T speeds. This sector is capacity-constrained and just had its export inputs tightened by China this week.
Space & Infrastructure:
$RKLB, $ASTS, $PL, $SATL, $FLY, $ONDS, $KTOS
Golden Dome. Autonomous systems. Satellite constellations. The government is funding the next generation of distributed compute and surveillance infrastructure from orbit.
Critical Materials:
$AXTI, $ALMU
The raw substrate layer that the entire photonics stack sits on. Severely undersupplied. Almost zero retail coverage.
Launch & Deployment:
$RKLB, $ASTS
Getting infrastructure into orbit requires reliable, affordable launch. One company owns this market. The public comp is scaling fast.
Every sector on this list has either hyperscaler contract backing, government funding, or a physical supply constraint that makes demand inelastic.
Save this for later.
-BP
Please note this is not financial advice.
THE AI SUPPLY CHAIN โ COMPLETE MAP
I've spent a few hours mapping the entire AI supply chain. Hyperscalers are spending $750 billion. These are the 42 companies building, powering, and deploying AI from start to finish.
Every layer. Every sector.
Bookmark it. Feel free to add companies in the comments.
Layer 1: Chip Equipment (the machines that build the machines):
$ASML$AMAT$LRCX$TSEM
Layer 2: Foundry & Fabrication (where chips are born): $TSM$GFS$INTC
Layer 3: GPU / ASIC / CPU (the AI compute engines): $NVDA$AVGO$AMD$MRVL
Layer 4: Memory & HBM (the bandwidth bottleneck): $MU$WDC
Layer 5: Photonics & Optical Interconnects:
$COHR$LITE$AAOI$POET$MTSI$ALMU
Layer 6: Data Center + Space (the physical home of AI):
$IREN$NBIS$CIFR$WULF$EQIX$RKLB$ASTS$PL
Layer 7: Cybersecurity (every new AI system is a new attack surface):
$CRWD$PANW$ZS$NET$S$FTNT
Layer 8: AI Software & Automation (where the ROI shows up):
$PATH$PLTR$NOW$HIMS$DDOG$SNOW$MDB$SOFI
Layer 9: Defense & End-Use (where AI becomes operational):
$ONDS$OSS$RKLB$AMPX$LHX$RTX$NOC$PNG.V
Every company on this list has either government budgets, hyperscaler contracts, or can potentially benefit from the AI build it somehow.
Save this. And if you found this valuable, you should follow me.
The market won't be red forever.
-BP
Please note: This is not financial advice.
