The anomaly jumps off the screen. A London-listed specialty foundry with a supply bottleneck, rising average selling prices, and a direct line into the AI optical interconnect boom — yet the valuation still carries a Swedish discount. Sivers Photonics (SIVE) is hiding in plain sight. While the market obsesses over NVIDIA's GPU shipments and TSMC's advanced packaging, the real constraint is the photons shuttling between accelerators. And Sivers holds one of the few independent InP active-integration platforms that matter.
I've spent years auditing smart contracts, looking for the invariant that breaks under stress. For a photonics foundry, the invariant is simpler: yield times price equals margin. Under current demand, that invariant holds — but only if capital expenditure doesn't blow a hole through the balance sheet. So let's do what I always do: ignore the narrative, trace the technical stack, and find the attack vector.
Context: The Photonic Layer of the AI Stack
Sivers Photonics is not a typical chip foundry. It operates in the III-V compound semiconductor space — indium phosphide (InP) and gallium arsenide (GaAs) — plus silicon photonics (SiPh). Its core competence is hybrid integration: combining InP gain media with passive silicon photonic circuits. That is the exact recipe needed for co-packaged optics (CPO), where optical engines sit on the same substrate as switching silicon. CPO is slated to become the default architecture for AI clusters because it slashes power and increases bandwidth density compared to pluggable modules.
The market context is blunt. AI clusters like NVIDIA's GB200 NVL72 require tens of thousands of dollars of optical interconnects per rack. Traditional pluggables are hitting a power wall. The industry is transitioning through near-package optics (NPO) toward full CPO. Sivers sits upstream, selling the laser and amplifier chips that make CPO work. It claims six new pluggable customers and a partnership with O-Net on external laser sources (ELS) — a key CPO enabler. The order book looks solid. But look closer, and the technical and financial layers start to reveal their own hidden complexity.
Core: The InP Ledger, Unaudited
Let's start with the technology. InP active components are hard to manufacture. Epitaxial growth, waveguide etching, facet coating — every step demands exacting control. Sivers has accumulated its own InP design IP, and that is the foundation of its value. It is not competing with TSMC at the nanometer level; it's competing in a niche where TSMC has no proven in-house capability. TSMC's COUPE (Compact Universal Photonic Engine) platform is scheduled for 2025 production, but it focuses on silicon photonics packaging, not necessarily the InP laser source. Sivers' expertise in InP-on-silicon heterogeneous integration is a genuine moat — but moats have a floodgate.
The company is reportedly running at high utilization. The original analysis references "supply bottleneck and ASP rise," which suggests the foundry is selling at premium prices. That is a rare position in a semiconductor industry that usually faces oversupply cycles. Yet the financials remain opaque. Gross margins for specialty III-V foundries typically range from 25% to 40%. IQE, the UK peer, sits around 25%. Sivers might do better, but there's no public proof. The real risk is on the balance sheet. A small foundry with <$50M revenue cannot fund a new US fab without dilution. The hidden insight: the "two fab allocation" mentioned in the original analysis hints at a Sweden/US split. A US facility would secure proximity to CPO leaders like Ayar Labs and potentially align with CHIPS Act funding. But that costs hundreds of millions. The current market cap likely doesn't price that capex cycle.
Customer concentration is the other attack vector. The top five customers probably account for 70–80% of revenue. Ayar Labs and O-Net are likely anchors. That dependency works both ways: a single win or loss can move the stock 30%. The original analysis suggests Sivers is diversifying, but six new pluggable customers is a small pool. Also, the O-Net collaboration ties Sivers to the Chinese optical supply chain. That is an opportunity and a geopolitical liability. US investors might shy away. The market is pricing Sivers as a niche foundry, not as a strategic CPO enabler. Serenity's critique — that too much time in Sweden keeps the valuation bound to a local retail investor base — is astute. A Nasdaq listing would rererate the STORY. But a listing doesn't change the fundamental law: yield times price equals margin, and margin divided by capex equals survivability.
Contrarian: The Protection is a Thin Waveguide
The consensus sees TSMC's COUPE as the primary threat. I see a smaller, more insidious risk: the durability of the InP pricing premium. The current "supply bottleneck" is real, but it's a cycle, not a permanent stratification. If Sivers raises ASPs too aggressively, it invites competition from Chinese III-V fabs like Yuanjie Technology and Accelink, which are gaining policy support. The Chinese supply chain is already building ELS and laser chips. They will undercut. The US is trying to reshore photonics, but China's domestic substitution is faster than anyone expects.
Another blind spot is the assumption that "order economic value" equals future revenue. In my audits, I've seen too many fat order books that die on delivery dates. Sivers' orders are only as good as its ability to ship. For a small foundry, a single yield issue on a new InP integration process can crater margins. The mathematical elegance of CPO doesn't guarantee manufacturing yield. Code is law, but bugs are the human exception — in photonics, it's contaminants that are the exception.
The ledger remembers what the wallet forgets: investors remember the AI story but forget the cash flow statement. Sivers is likely burning cash. If it starts a US fab, the cash burn intensifies. A desperate equity raise would dilute current shareholders precisely at the moment the CPO narrative peaks. The Swedish discount might be a feature, not a bug — it's pricing in the real risk of capital dilution. The market may be smarter than Serenity thinks.
Takeaway: Watch the Capex, Not the Photons
The future is not a straight line. Sivers has the InP edge, but edges erode. The next catalyst is not a headline order — it's the announcement of a US fab location, and the funding mechanism attached to it. If the company secures strategic investment from a US CPO player or qualifies for CHIPS Act funding, the valuation re-rates. If not, the supply bottleneck will eventually normalize, and the ASP premium will evaporate.
My verdict: the risk/reward is tilted toward the upside, but only for investors who can stomach the volatility of a small-cap foundry trying to catch a wave as big as AI. The technology is sound. The customer relationships are real. But the capital structure is fragile. I'm not looking for the perfect wafer — I'm looking for the moment when the company admits its growth story requires a bigger balance sheet. That's the clearest entry point.
Until then, I'll keep my position small and my eyes on the ledger. The machine is humming. The question is whether it's being amplifying or exploited. Every system has an attack vector. For Sivers, the vector isn't in the optical engine — it's in the financing.