$SIVE $SIVEF $GF EXECUTIVE ASSESSMENT
The Sivers Semiconductors and GlobalFoundries announcement is best interpreted as a strategic ecosystem validation event, not as a conventional revenue-bearing customer award. The strongest read-through is that Sivers’ continuous-wave laser array technology has been positioned inside GlobalFoundries’ silicon photonics reference-design pathway at the same time GF is explicitly positioning SCALE as an OCI MSA-capable co-packaged optics platform. That makes the announcement materially more significant than a generic “AI optics” press release, because OCI is not merely another datacenter optical module standard; it is an attempt by leading XPU, switch, and hyperscale stakeholders to define a common optical physical layer for AI scale-up connectivity. However, the public evidence supports “Sivers is now a publicly named laser-array partner available in GF silicon photonics and SCALE reference designs,” not “Sivers has won OCI,” “Sivers is exclusive,” or “Sivers has secured hyperscaler production volumes.” The investment conclusion should therefore be asymmetric but conditional: the announcement increases Sivers’ strategic relevance and option value, while the commercial value remains dependent on qualification, customer adoption, second-source dynamics, production timing, and GF’s ability to convert SCALE from a platform announcement into volume optical engine deployments.
WHAT THE SOURCE MATERIAL ACTUALLY SHOWS
The screenshots correctly link 3 separate pieces of information: 1) Sivers announced that its laser arrays will be integrated into reference designs built on GF’s silicon photonics platform and will also be available in GF’s SCALE platform; 2) GF’s own SCALE slide and press release present SCALE as a CPO solution for OCI and as an OCI MSA-capable platform; and 3) the OCI MSA was formed by AMD, Broadcom, Meta, Microsoft, NVIDIA, and OpenAI to establish an open, interoperable optical interconnect specification for AI scale-up. This triangulation is meaningful because it places Sivers in the reference-design layer of a GF platform that is publicly mapped to a standard backed by the most relevant AI infrastructure buyers and silicon vendors. The tweet’s broad thesis is directionally valid, but the wording “Optical Scale-Up initiative” should be tightened: the official organization is the Optical Compute Interconnect MSA, while the Business Wire release headline uses “Optical Scale-up Consortium” to describe the consortium’s purpose.
The Sivers announcement states that Sivers laser arrays will support GF’s silicon photonics platform and SCALE optical engine solutions, targeting a $25B pluggable optics market by 2030. It also states that the collaboration supports CPO, LPO, and other emerging datacenter interconnect architectures. This is important because the release does not restrict Sivers to a single point product; it frames Sivers as a light-source supplier across GF’s broader silicon photonics stack. Still, the announcement is non-regulatory, does not disclose purchase commitments, does not name a hyperscaler or switch customer, does not include a production revenue forecast, and does not claim exclusivity. The evidence therefore supports a positive strategic interpretation but not a de-risked revenue ramp.
WHY OCI MATTERS
OCI matters because AI scale-up connectivity is moving from a board-and-copper problem to a system-level optical architecture problem. The OCI MSA release states that copper connectivity is reaching physical reach limitations that affect AI cluster scale-up domain architectures, and that OCI is intended to enable migration from copper-based to optical-based scale-up architectures. The specification combines NRZ modulation and WDM optical technology and shifts the paradigm from a module-centric model to a silicon-centric model. This language is central: the consortium is not simply trying to standardize faceplate optics; it is trying to move optical connectivity closer to compute and switching silicon so that high-bandwidth accelerator domains can scale across larger physical footprints without the power, reach, and signal-integrity penalties of copper.
The OCI membership mix is unusually important. AMD, Broadcom, and NVIDIA represent critical accelerator, ASIC, and networking silicon interests; Meta, Microsoft, and OpenAI represent hyperscale and frontier AI demand. That combination makes OCI a potential counterweight to proprietary scale-up fabrics, because a common optical PHY could allow XPUs, scale-up switches, optical engines, and system builders to interoperate over a shared physical layer. The consortium explicitly frames the objective as an open specification, reduced integration risk, shorter development cycles, multi-generational deployment, and a multi-vendor supply chain. From an investment perspective, this is the key reason the GF/Sivers announcement has potential ecosystem significance: a component qualified into an OCI-aligned reference platform can have more leverage than a component qualified into a single proprietary module family.
TECHNICAL READ-THROUGH
The GF SCALE slide in the source material is technically consistent with the OCI 1.0 specification. The screenshot calls out 53Gbps signaling, bidirectional signaling, 2 CWDM wavelength groups per fiber, and 4 DWDM wavelength groups per CWDM group. The OCI technical specification defines a 200G Optical Compute Interconnect line interface using 53.125 Gbaud NRZ signaling on 4 wavelengths, with tightly spaced DWDM channels inside CWDM wavelength groups and bidirectional transmission over the same fiber using separate A and B wavelength groups. It also states that the external laser source provides light to the group A and group B transmitters for modulation. This is precisely the architectural niche where a supplier of multi-wavelength CW DFB laser arrays can matter.
The architecture is structurally different from conventional EML-based pluggable optics. In an EML architecture, the laser and modulator are integrated in the optical source device; in a silicon photonics architecture, an external or remote CW light source can feed a silicon photonic integrated circuit, where micro-ring resonators or other silicon photonic devices perform modulation and routing. OCI explicitly points to DWDM grids and cascaded micro-ring resonators for low-power, high-density AI backend interconnects. GF’s SCALE announcement similarly emphasizes CWDM/DWDM, integrated photonic devices, micro-ring modulators, coupled ring resonators, photodiodes, TSVs, fine copper pad pitches, and 2.5D/3D stacking. This indicates that the critical value pools migrate toward silicon photonic integration, advanced packaging, fiber attach, laser sourcing, test, and thermal/mechanical reliability.
The external laser source is not a secondary component in this architecture. It supplies the optical carriers that the silicon photonic engine modulates, and its output power, wavelength stability, linewidth, relative intensity noise, polarization handling, thermal stability, and field serviceability become system-level constraints. OCI’s external laser source section specifies a polarization-maintaining fiber connection to the OCI chiplet, group A and group B wavelengths around the 1308nm to 1335nm range, laser RIN of -144 dB/Hz, linewidth of 1 MHz, SMSR of 30 dB, polarization extinction ratio of 16 dB, and output reflectance requirements. Sivers’ value proposition therefore depends not only on having DFB laser arrays, but on meeting the exact noise, thermal, wavelength, reliability, coupling, and packaging requirements of OCI-class deployments at volume yield.
Sivers has credible technical adjacency. The company previously demonstrated CW-WDM MSA-compliant DFB laser arrays with Ayar Labs’ SuperNova light source, including an 8-wavelength DFB laser array built on Sivers’ InP100 platform, output power above 65 mW per channel, 400 GHz channel spacing around 1300nm, and compatibility with silicon photonics and CPO. Sivers also announced a 16-wavelength DFB laser-array demonstration with Ayar Labs for 16 Tbps of bidirectional bandwidth, and Ayar describes SuperNova as a 16-wavelength, 16-port external light source capable of supplying light for 256 data channels or 16 Tbps bidirectional bandwidth. These prior demonstrations increase the plausibility that Sivers’ laser-array technology is relevant to GF SCALE and OCI-like architectures, although public materials do not prove final qualification against GF customer requirements.
ASSESSMENT OF THE “FIRST PUBLICLY NAMED LASER PARTNER” CLAIM
The claim that Sivers is the first publicly named laser partner integrated into GF SCALE reference designs is plausible based on available public announcements, but it should not be treated as proven in the strong sense. The Sivers release explicitly states that Sivers laser arrays will be integrated into reference designs built on GF’s silicon photonics platform and will be available in SCALE. Public GF SCALE materials reviewed here do not identify another named laser-array partner in the same manner. However, absence of public evidence is not evidence of exclusivity, and hyperscaler-aligned optical ecosystems generally require multi-sourcing for cost, yield, reliability, and supply assurance. The correct framing is therefore that Sivers appears to have achieved a publicly visible reference-design position inside GF’s SCALE/silicon-photonics ecosystem, which is strategically valuable, but not equivalent to sole-source status or a production award.
The distinction matters because reference design inclusion can be a powerful funnel but is not automatically monetizable. In semiconductors and optical components, a reference design can reduce customer engineering friction, establish a default bill-of-materials path, and create early qualification leverage. It can also remain a technology showcase if customers choose alternate sources, if performance margins are insufficient, if system vendors require different thermal or mechanical integration, or if the platform does not enter volume. The Sivers announcement therefore improves probability-weighted strategic value, but the investment case still requires evidence of design wins, purchase orders, production ramps, and customer concentration resilience.
IMPLICATIONS FOR SIVERS
For Sivers, the announcement is potentially material because the company’s current revenue base is small relative to the scale of the AI optical interconnect opportunity. Sivers reported Q1 2026 net sales of SEK 61.9m, down 22% YoY, with Photonics revenue of SEK 17.8m, down 32% YoY. The company attributed the quarterly weakness to FX headwinds and revenue timing issues linked to defense spending delays, while also highlighting opportunity-pipeline growth to $799m and multiple product ramps into 2027. In that context, a GF reference-design relationship is not just another logo; it can be a path for Sivers’ photonics business to access a much larger customer and platform ecosystem than it could access through direct selling alone.
The strategic upside for Sivers is that GF can serve as a distribution and validation layer. GF has customer relationships across optical transceiver, datacenter, networking, and emerging CPO architectures; if Sivers becomes a qualified laser option in GF reference designs, Sivers may benefit from customer pull-through without needing to win every optical engine design independently. This is especially important in CPO and LPO, where the customer does not simply buy a discrete laser chip; the customer buys a system-level solution involving silicon photonics, EIC/PIC integration, fiber attach, packaging, firmware/management, thermal design, test, and reliability. A laser vendor embedded in a foundry-backed reference flow can become part of the default engineering path, which can be highly valuable if that reference flow becomes widely adopted.
The risk is that Sivers remains one of several technically capable laser suppliers rather than the dominant supplier. Lumentum is showcasing an 800 mW 1310nm super-high-power laser, a 400 mW UHP laser family, and a 16-channel DWDM UHP laser source for next-generation CPO architectures; Coherent is sampling 400 mW 1311nm CW lasers with volume production expected in Q3 2026 and is expanding InP capacity. These are formidable competitors with scale, manufacturing history, customer relationships, and credibility in high-volume optical components. Sivers may have attractive multi-wavelength DFB-array capability, but the GF relationship should be evaluated against the reality that hyperscalers and CPO platform owners will prefer qualified second and third sources where feasible.
Sivers also carries execution and financing risk. The company’s 2025 annual report disclosed accounting restatements and corrections across several areas, including development expenditure, government grants, inventory valuation, revenue timing, contract assets, contract liabilities, and provisions. It also disclosed a USD 17.0m loan facility and a directed share issue of approximately SEK 125m after the reporting date. These items do not negate the technology thesis, but they do matter for institutional underwriting: the optical opportunity is high-conviction at the industry level, while Sivers’ conversion of that opportunity into durable, audited, high-margin product revenue remains unproven.
IMPLICATIONS FOR GLOBALFOUNDRIES
For GF, SCALE is strategically important because it moves GF from being a silicon photonics process supplier toward being a platform and ecosystem orchestrator. GF’s SCALE launch states that the platform is the industry’s first OCI MSA-capable CPO platform, built with GF silicon photonics technology, using CWDM and DWDM for bidirectional transmission over each fiber. GF says it has demonstrated 8λ and 16λ bidirectional DWDM natively on its platform, offers qualified photonic devices including 50Gbps and 100Gbps micro-ring modulators, and supports TSVs plus copper pad pitches from 110μm down to sub-45μm for 2.5D/3D stacking. These are not generic foundry claims; they are claims about platform-level readiness for AI optical interconnect integration.
The SCALE announcement fits a broader GF silicon photonics strategy. GF acquired Advanced Micro Foundry in Singapore in 2025, describing the transaction as expanding its silicon photonics technology portfolio, production capacity, and R&D footprint, while establishing GF as the largest pure-play silicon photonics foundry by revenue. Reuters also reported that the AMF acquisition positioned GF as a major silicon photonics player in AI datacenters and quantum computing, with a new research center planned in Singapore. This context makes the Sivers partnership more credible: GF is not casually attaching itself to AI optics; it is building a silicon photonics franchise around process technology, capacity, customer access, packaging, and reference designs.
The financial significance for GFS is meaningful but not yet dominant. GF reported Q1 2026 revenue of $1.634B, 27.6% gross margin, $104m net income, $227m non-IFRS net income, and $3.8B of cash, cash equivalents, and marketable securities. Its Q1 release also highlighted SCALE as a recent business development. Separately, GF investor materials point to a line of sight toward more than a $1B silicon photonics revenue run-rate, with silicon photonics reported within Communications Infrastructure and Data Center. For a company of GF’s size, silicon photonics can become a multiple-relevant growth vector, but it is not yet the whole company. The key debate for GFS is whether photonics becomes a durable high-growth, high-margin franchise large enough to re-rate the equity, or remains one differentiated technology among several cyclical specialty-foundry end markets.
MARKET CONTEXT
The optical interconnect market backdrop is highly favorable. McKinsey estimates that data centers may require $6.7T of worldwide investment by 2030 to keep pace with compute demand, and states that networking performance has not scaled at the same pace as compute performance, making networking a pivotal differentiator for AI customers. McKinsey also estimates that intra–data center optics above 1.6Tbps could exceed a 30% CAGR and reach about $24B by 2029, with hyperscalers shifting about 87% of back-end optics to 800Gbps and above by 2029 and 1.6Tbps transceivers accounting for more than 40% of that demand. This supports the broader Sivers/GF thesis that optical links are becoming strategic rather than peripheral in AI infrastructure.
Silicon photonics is also gaining share. LightCounting expects more than half of optical transceiver sales in 2026 to come from modules based on silicon photonics modulators, up from 10% in 2018 and 33% in 2024. LightCounting also expects LPO and CPO to double silicon photonics market share from 30% in 2025 to 60% in 2030, while emphasizing that CPO adoption requires realistic expectations because manufacturing, power, cost, and end-user acceptance remain barriers. This is the correct market lens for the announcement: the long-term direction is strongly supportive of silicon photonics and CW light sources, but adoption will be staged and uneven rather than instantaneous.
CPO timing remains a critical controversy. McKinsey states that large-scale CPO adoption is unlikely until 2035, although the CPO supplier segment could still grow at a 20% CAGR through 2029 and 1.6Tbps CPO testing began as proof of concept in 2024. LightCounting is more constructive on scale-up CPO specifically, arguing that even modest CPO adoption in scale-up networks would require millions of ports because scale-up has roughly 9x higher bandwidth requirements than scale-out. The apparent discrepancy is reconcilable: large-scale universal CPO adoption across datacenter switching may take much longer, while selective CPO adoption in high-value AI scale-up domains can still become commercially significant earlier. Sivers’ GF/SCALE exposure is more levered to the second interpretation than the first.
COMPETITIVE AND ECOSYSTEM RISK
The most important competitive risk is that OCI and CPO are ecosystem standards, not single-vendor procurement guarantees. NVIDIA has already announced Spectrum-X and Quantum-X silicon photonics networking switches using CPO, with an ecosystem including TSMC, Coherent, Corning, Foxconn, Lumentum, and SENKO. NVIDIA claims its photonics switches deliver major power, signal-integrity, resiliency, and deployment benefits, and its Quantum-X and Spectrum-X photonics platforms create a vertically integrated alternative pathway. The presence of NVIDIA as an OCI founding member does not imply NVIDIA will use GF/Sivers for its own photonics roadmap; large vendors can support standards while still deploying proprietary or captive implementations.
Broadcom is another major axis of competition and influence. Broadcom announced its 51.2Tbps Bailly CPO switch in 2024, citing a 70% improvement in optical interconnect power, and later described third-generation CPO developments and ecosystem partners. Broadcom is also an OCI founding member and an editor of the OCI 1.0 technical specification. This creates both opportunity and risk for GF/Sivers: Broadcom’s participation increases OCI credibility, but Broadcom is also a sophisticated CPO platform owner with its own supplier preferences and integration strategy.
A further risk is that silicon photonics foundry competition is intensifying. LightCounting notes that GF’s AMF acquisition places GF in the top position by silicon photonics revenue but also re-energizes competitors, including Tower Semiconductor, which is expanding silicon photonics capacity. Reuters reported that STMicroelectronics is considering further Crolles expansion driven by rising AI datacenter silicon photonics demand, with ST targeting a much larger share of the market through an AWS-linked strategy. This competitive response is rational because the entire industry sees optical I/O as a future control point. The implication is that GF’s early SCALE positioning is important, but not structurally unassailable.
THE 800G EML TO 1.6T/3.2T SIPH + CW LASER THESIS
The tweet’s statement that OCI drives a transition from 800G EMLs to 1.6T/3.2T silicon photonics plus CW laser architectures is directionally right for scale-up and CPO, but too broad if applied to all optical modules. EML-based pluggables remain competitive and will continue to coexist with silicon photonics across multiple reaches and form factors. Lumentum’s OFC 2026 disclosure explicitly includes a 1.6T DR4 OSFP pluggable prototype using 400G differential EMLs as a stepping-stone to 3.2T, illustrating that EML innovation is not stopping at 800G. The more precise statement is that OCI and CPO increase the strategic value of silicon photonics plus external CW laser architectures in high-density, short-reach AI scale-up domains where bandwidth density, energy efficiency, serviceability, and integration close to compute are more important than backward compatibility with conventional module architectures.
This nuance is investment-critical. A simplistic “EML loses, CW laser wins” framing risks overestimating Sivers’ addressable market and underestimating incumbents. A better framing is that optical interconnect architectures are fragmenting by use case: EMLs can remain attractive in pluggable 1.6T/3.2T modules; silicon photonics can gain share in LPO, NPO, and CPO; external laser sources become more important as modulation moves onto silicon photonic engines; and scale-up optical fabrics may become a distinct market with different latency, reach, thermal, and packaging requirements than conventional Ethernet pluggables. Sivers’ GF relationship is valuable because it attaches Sivers to the architecture with the highest potential strategic change, not because it eliminates legacy optical approaches.
