The Photonics Bridge: Why Sivers Photonics Is the Silent Bottleneck in the AI Trust Layer

I remember watching the liquidity dry up in crypto markets last year, but the real liquidity crunch is happening in photonics, and nobody is talking about it. Over the past seven days, I've been sifting through the noise of AI mania and found a different kind of signal: a small British semiconductor foundry called Sivers Photonics (SIVE) is quietly sitting on a supply bottleneck that could decide the future of co-packaged optics. And here's the kicker: the market is pricing it like a local Swedish shop, not like the third-ranked InP integration player on the planet. The company doesn't make Bitcoin miners or GPUs. It makes the optical engines that connect them. As AI clusters scale to 1.6T interconnect speeds, the demand for high-bandwidth, low-power optical links is exploding. But the real untold story is the one Serenity Capital, the activist investor, is pointing at: Sivers' value is trapped under a Swedish valuation ceiling, and breaking that ceiling requires a strategic pivot to the US market. This is not another crypto narrative. This is the infrastructure layer where decentralized trust meets physical reality. Let me take you inside the silicon. Sivers is a specialty foundry focused on III-V compound semiconductors — indium phosphide (InP) and gallium arsenide (GaAs) — plus silicon photonics (SiPh). Unlike traditional CMOS logic fabs that chase nanometre nodes, its process metrics are waveguide dimensions, coupling efficiency, and integration density. The core technology is hybrid integration: bonding InP gain media onto silicon photonic wafers to create active optical components. That's the same architecture behind co-packaged optics (CPO) and external laser sources (ELS), the two hottest segments in the next generation of data centre interconnect. Here's where the technical analysis gets interesting. In InP active integration, Sivers ranks in the first tier, competing directly with Intel and Broadcom's silicon photonics teams. In silicon photonics overall, it lags TSMC's COUPE platform by roughly two to three years. But that's the wrong metric. The industry doesn't need another giant making everything; it needs a flexible, neutral, specialised foundry that can customise InP lasers and optical amplifiers for CPO designs. My audit of over 150 Uniswap V2 liquidity pools taught me to look for real economic value behind the noise. Sivers has that value — it's just hidden beneath a mountain of misunderstood financials. Let's break down the supply chain position. The optical chip is the beating heart of every transceiver, representing 30–50% of the BOM cost. In the profit pool of the optical communications industry, optical chips capture 35–40% of total profits, followed by modules at 30–35%. Sivers sits at the highest-margin segment. But its upstream dependencies are severe: InP substrates are dominated by Japanese suppliers like Sumitomo Electric and JX Nippon Mining. High-resolution lithography tools come from ASML, Canon, and Nikon. MOCVD and MBE epitaxial equipment are supplied by Aixtron and Veeco. This is a concentrated supply chain, and Sivers has limited negotiating power. Downstream, there's another concentration risk. The company works with a handful of customers — Ayar Labs, O-Net, and six new pluggable module clients — which means 70–80% of revenue could come from the top five. But here's the twist: the announced supply bottlenecks and rising ASRs (average selling prices) suggest demand is significantly outpacing supply. That flips the negotiation dynamics in Sivers' favour, at least temporarily. The real strategic question is capacity allocation. With two fabs — one in Sweden, one possibly planned in the US — Sivers is facing a choice: allocate scarce wafer starts to high-value CPO customers or to pluggable module clients. Serenity's push to shift focus to the US market suggests they see the American fab as the key to unlocking both customer proximity and a more sophisticated investor base. The market opportunity is hard to overstate. CPO penetration is projected to grow from under 5% in 2024 to 20–30% by 2028. This is a structural shift, not a cyclical blip. AI training clusters like NVIDIA's GB200 NVL72 have pushed optical interconnect value per rack from a few thousand dollars to tens of thousands. The demand for external laser sources — Sivers' bread and butter — is growing at 30% annually as CPO architectures adopt these discrete sources. LightCounting predicts the CPO market will reach tens of billions of dollars by 2028. Sivers is directly in the path of that rocket, yet its valuation remains stuck in the AIM market's low-liquidity orbit. Now, the contrarian angle. The common bear case is that TSMC's COUPE platform will crush Sivers. And yes, TSMC has scale, capital, and customer relationships. But TSMC's hybrid integration still relies on partnerships for InP active components. Sivers' InP epitaxial growth know-how is not a commodity. It's a wrapper of decades of material science, environmental controls, and process tuning. That's a defensible barrier. The bear case also points to weak financials — estimated negative free cash flow, a ROIC below WACC, and a price-to-sales ratio between 5x and 8x, well above IQE's 3x. But here's the blind spot: those numbers do not capture the embedded option value of an upcoming US expansion, potential CHIPS Act subsidies, or the strategic premium that American institutional investors place on AI supply chain purity. The quantitative metrics are trailing indicators; the order book is a forward indicator. And the forward indicator is screaming scarcity. The biggest risk isn't TSMC. It's customer concentration. Losing Ayar Labs or O-Net would be catastrophic. But even that risk is mitigated by the industry's shift to multi-sourcing. The myth of a single source for photonic components is fading. Every hyperscaler is desperate for redundant supply chains, and Sivers' neutrality — as a British/Swedish foundry not aligned with any geopolitical bloc — becomes an asset, not a liability. In a world of export controls on advanced logic, photonics remains a grey zone. But the real threat is the weaponisation of AI infrastructure. If CPO gets classified as critical to AI, Sivers could face indirect compliance hurdles. Yet that scenario also raises the value of a neutral, Western-friendly fab. So, what's the takeaway? This is not a call to buy a stock. This is a call to recognise the underlying reality: the future of decentralised, high-performance computing hinges on tools like Sivers. Open source is not a license; it's a state of mind. And the state of mind here is that we can't have a trust layer without hardware sovereignty. The company's move to pivot to the US markets is a strategic gambit, not to flee the EU, but to find a valuation that reflects the true economic value of its orders. If Sivers can execute the US expansion, deepen its relationship with Ayar, and diversify its pluggable client base, it won't just bridge the optical gap — it will become the missing link between the abstractions of software and the physics of light. The market will eventually catch on. But will it be too late? Liquidity isn't a measure of asset quality; it's a measure of attention. And right now, attention is flowing toward London and New York, not Stockholm. We didn't build a mirror for this market; we built a lens. And that lens focuses on one single question: can a small foundry out-engineer the giants by staying narrow? Mining for truth in the noise of AI mania, I suspect the answer lies not in the semiconductor roadmap, but in the willingness of investors to see beyond the noise and into the substrate. The next time you look at your GPU powering a zk-proof, remember that somewhere in a clean room in Sweden, a tiny InP laser is being lit to connect it all.