Sivers Photonics trades on London's AIM as though it were a penny stock in a thematic ETF — visible enough to attract attention, small enough to be ignored. That is a dangerous combination for a company with real technology. It is not a crypto token. It is a III-V compound semiconductor foundry producing indium phosphide and silicon photonics for datacentre optics. And that is exactly why I began this assessment with a stress test, not a narrative.
The fracture line appeared in the recent investor update. Six new pluggable customers. An external laser source product shipped with O-Net. Supply bottlenecks and rising ASPs. All textbook bullish. But when I mapped that order book against the TSMC COUPE timeline, the arithmetic produced a less comfortable picture. The market is pricing Sivers at an estimated 5–8x price-to-sales, while its return on invested capital sits somewhere between 3% and 5% against a cost of capital above 10%. Valuation is a fiction; exposure is the reality.
The Foundry Context
Let me define the terrain for readers who normally watch smart contracts, not wafer fabs. In the AI datacentre, light is the new electricity. Every GPU cluster needs high-speed interconnects. Pluggable optical modules are the current workhorse; co-packaged optics (CPO) are the promised upgrade — putting the optical engine on the same substrate as the switch chip. The CPO roadmap is being drawn by Ayar Labs, Broadcom, Intel, and eventually TSMC. Sivers sits upstream. It grows InP epitaxial layers, integrates active components with silicon photonics, and sells the resulting chips to module and CPO integrators.
On InP active integration, Sivers is genuinely close to first tier. On overall silicon photonics integration density, it lags TSMC by perhaps two to three years. That gap is the entire thesis. It is also the entire risk.
Process: The InP Edge and the Scale Gap
Sivers does not compete on 3nm. It uses DUV lithography at 248 and 193 nanometres. The relevant metrics are waveguide loss, coupling efficiency, and InP-on-Si hybrid integration yield. On that last metric, Sivers is credible; it has deep, defensible know-how in epitaxial growth and active-device integration. But hybrid integration only matters when it is produced at scale. Industry yields for InP active integration hover around 70–85%, compared with 85–95% for mature silicon photonics. Sivers does not publish yield data. In my experience auditing hardware-dependent infrastructure, silence on yield is a data point. A foundry with good yields puts a chart on the front page. If they don't, they publish customer names instead.
The real technological problem is not the InP, it is the scale. TSMC's COUPE platform is expected to reach production in 2025. GlobalFoundries already has a 45nm silicon photonics platform. Intel and Broadcom have products in the field. Sivers has a niche, not a market. In a head-to-head procurement process with a hyperscaler, the procurement team will ask a simple question: who has the most wafers shipped? That is not a technical question. It is a scale question.
Supply and Demand: The Bottleneck Mirage
The supply chain is brutally concentrated. InP substrates come largely from Japanese suppliers. DUV steppers come from ASML, Nikon, and Canon. MOCVD epitaxy tools come from Aixtron and Veeco. Sivers has limited pricing power against those upstream monopolies. Its leverage exists only on the downstream side, where a genuine supply bottleneck has emerged. AI clusters from NVIDIA's GB200 generation to the next wave all need more optical bandwidth. Pluggable modules are migrating to 800G and 1.6T. CPO is moving from pilot to early production. That demand is real.
ASP increases are real too. And when a specialty foundry runs at high utilisation — the phrase 'supply bottleneck' strongly implies utilisation above 90% — pricing power temporarily improves. That improvement flows straight to gross margin. I model Sivers' current gross margin in the 30–40% range, below TSMC's 55% but above IQE's 25–30%. The upside is there if CPO and ELS products become a larger share of revenue. The problem is that revenue growth is capped by wafer starts, not demand. To break the cap, Sivers must spend.
My estimate from comparable specialty fabs puts capital expenditure at 20–30% of revenue. A new US facility would add 12–18 months before meaningful revenue. Depreciation would shave 300 to 500 basis points off gross margin. The breakeven for a new fab is around 60–70% utilisation. That is a high bar for a company that is still small.
The TSMC Countdown
Demand is not Sivers' problem. The CPO market will move from well under 5% penetration in 2024 to perhaps 20–30% by 2028. Ayar Labs has publicly discussed expansion through 2028. External laser sources, a niche Sivers serves with O-Net, are growing at more than 30% annually. Any one of these lines would justify a growth multiple. But Sivers is not alone in the lane.
TSMC's COUPE is not a distant threat. It is a 2025 threat. If TSMC executes on schedule — and the probability is high, given its resources and customer relationships — it will capture a large share of the CPO optical engine market. Sivers' defence is its InP active integration know-how and its early co-design relationships with CPO makers. But those are thin walls when the neighbour is TSMC. The risk is not that Sivers loses all its orders; the risk is that its orders become lower-margin, second-source leftovers.
Customer concentration amplifies the TSMC risk. The top five customers likely represent 70–80% of revenue. Ayar Labs and O-Net are not optional; they are structural dependencies. The recently announced six new pluggable customers will help, but a new customer in a pre-production phase is not revenue certainty. It is a non-binding expression of interest.
Geopolitics and the US Pivot
Geopolitical risk is medium-low but not absent. Sivers is a UK-headquartered, Sweden-based operator. It is not on the US Entity List. It uses DUV rather than EUV, so it escapes the harshest export controls. China's gallium and germanium restrictions do not directly affect InP. Japanese InP supply is unlikely to be weaponised in a way that would spare a British company. But the strategic drift is real.
US policy wants local foundry capacity. US clients prefer local supply. The CHIPS Act is a magnet. Activist pressure, exemplified by Serenity, is pushing Sivers to shift its centre of gravity toward the United States. That is less a suggestion than an acknowledgment of where the revenue will flow. Staying too long in Stockholm means absorbing the Swedish investor logic — a logic built around dividends, stability, and small-cap inertia. It has no vocabulary for semiconductor growth options. That is why the stock trades at a discount to what a comparable US-listed photonics pure-play would command. Market centre determines valuation architecture as much as earnings.
Financial Stress Test
The financials do not support complacency. Gross margins are not disclosed, but 30–40% is the right range for a specialty foundry with product mix between low-value pluggables and high-value external laser sources. R&D is likely expensed, which is conservative. Free cash flow is probably negative. With estimated ROIC of 3–5% and WACC of 10–12%, Sivers is destroying value on every wafer according to standard allocation logic. Bulls will say that accounting cannot capture the IP stack. I have heard that argument at every ICO, every SPAC, and every 'infrastructure token' I have audited. The IP stack is only real when it converts to contracted cash flow.
The order book does carry real economic value. Supply bottleneck plus ASP growth means new orders are worth more than those signed during a downturn. That is the most credible bullish argument. But an order book is not a balance sheet asset. It is a promise contingent on execution. Execution requires capital. Capital requires dilution. Dilution at this valuation is a silent tax on existing shareholders.
What the Bulls Got Right
Now let me give the bulls their due. Sivers is not a fraud. It is not a marketing vehicle. It has actual epitaxial knowledge and a seat in a supply chain that will not disappear. The external laser source product with O-Net is not a press release; it is a co-designed, ship-ready product. The six new pluggable customers suggest concentration will improve. And the UK-Sweden base gives Sivers a genuinely neutral position between the US-China semiconductor standoff. That neutrality is a form of optionality that no incumbent can copy by throwing money at a fab.
The market is right to see structural demand. The market is also wrong to price Sivers as if it will capture the full CPO wave. The capture rate is the variable that determines whether 5–8x sales is a discount or a delusion.
Takeaway: Find the Fracture Line
I have built my career on finding structural flaws before they become price drops. You do not need to be a forensic auditor to find this one. TSMC's roadmap is printed in public. The supply bottleneck is transparent in every hyperscaler datacentre report. The customer concentration is visible in every customer announcement. The fracture line was visible before the quake struck.
The next six quarters will determine whether Sivers becomes a niche king or a footnote in a CPO post-mortem. It cannot simultaneously diversify customers, build a US fab, and avoid dilution without making a choice. The order book says the demand is real. The balance sheet says the money is not there. The ledger balances today. The architecture is already bleeding. Sivers has three trimesters to prove that its orders carry economic weight before the market decides that a small-cap photonics foundry is just another asset to reprice at zero.
