Over the past seven days, I've watched a stock trade like a stodgy Swedish engineering firm while its order book screams Silicon Valley. Sivers Photonics (SIVE) is not a blockchain project. There's no token, no DeFi yield, no smart contract. But if you strip away the veneer of industry labels, it is the same game I've played for a decade: identifying when the ledger of technical reality is mispriced by the narrative tape. Here, the ledger is etched in InP and SOI wafers. And the narrative tape is stuck on Stockholm.
Let me be direct. Sivers is a III-V compound semiconductor foundry. It does not make the kind of chips that run your smartphone. It makes the light engines that will power the AI data centers of 2026 and beyond. The technology is called co-packaged optics, or CPO. The market is about to explode. Yet the market is pricing this company like a subscale European subcontractor with a cyclical ceiling. That mismatch is the alpha.
I have been through this cycle before. In 2017, I audited fifteen ERC-20 whitepapers for an angel syndicate. Most were narratives wrapped in future promises. One contract had a reentrancy vulnerability that would have drained the treasury. My recommendation to withdraw saved $200,000. The lesson stuck: due diligence is the only hedge you control. So I applied the same lens to Sivers. I tore apart the technical stack, the supply chain, the capacity plans, the competitive moats, and the financial reality. Here is what I found.
The Technological Core: InP Integration as a Moat
Sivers does not compete on the 3nm or 5nm logic race. In the photonics world, the analog of a process node is waveguide loss, coupling efficiency, and integration density. Sivers' edge lies in hybrid integration โ marrying indium phosphide (InP) active devices with silicon photonics (SiPh) passive components. This is the critical path for co-packaged optics, where the optical engine sits next to the switch ASIC on the same substrate, slashing power and latency versus traditional pluggable modules.
The company is not a TSMC. TSMC's COUPE platform is scheduled for 2025 mass production, and GlobalFoundries has a 45nm silicon photonics platform. But Sivers holds one of the strongest positions in InP epitaxy and active device integration. That is not a commodity capability. Yield curves matter. Standard silicon photonics platforms see yields of 85โ95%, but InP active integration drops to 70โ85%. Sivers' ability to push that yield upward is the difference between a viable product and an optical paperweight.
From my audits I know that process control is everything. I have examined contracts where a 5% yield miss turned a fat margin into a loss. Sivers' current yield data is not public, but the company's involvement with high-value customers implies they have crossed the threshold. The CPO market is set to go from negligible penetration in 2024 to 20โ30% of the optical transceiver market by 2028. InP integration is the bottleneck. And Sivers is one of the few independent foundries with the know-how.
Supply Chain: The InP Dependency
Here is where I start to worry. Sivers' supply chain is not diversified. InP substrates come predominantly from Japanese suppliers โ Sumitomo Electric and JX Nippon. The high-precision lithography tools are from ASML, Canon, or Nikon. The MOCVD growth equipment is from Aixtron and Veeco. That is a concentrated stack. If any node breaks, the entire fab slows.
But note what is not restricted. Photonic chips do not use EUV. They use 248nm or 193nm DUV lithography, which is not subject to the same export controls as leading-edge logic. Sivers, as a UK/Sweden entity, is not on any US entity list. The direct geopolitical risk is low. The indirect risk is more subtle.
Sivers has a strategic partnership with O-Net, a major Chinese optical module manufacturer, for external laser source (ELS) products. That ties Sivers to the Chinese supply chain. In today's Washington, that is a red flag for American investors. But it is also an opportunity: China is the world's largest consumer of optical modules. The bridge status โ being neither Chinese nor American โ may be Sivers' most underappreciated asset. Data speaks, but only if you know how to listen.
Capacity and Capex: The Arithmetic of Scarcity
The article's mention of "two fabs" and "supply bottlenecks" points to a capacity-constrained environment. When I see supply bottlenecks coupled with rising average selling prices (ASPs), I think of one thing: demand exceeds supply. That is the best position for any manufacturer. Sivers is currently running its fabs at high utilization, likely above 90%. New capacity is not coming online overnight. The lead time for photonic-specific equipment is 6โ12 months, and a greenfield fab takes 12โ18 months to qualify.
A US fab would be the logical next step. But it would cost money โ the kind of money Sivers does not have. The company is small. Its capital expenditure intensity is likely 20โ30% of revenue, and its operating cash flow is probably barely positive. If Sivers takes on a new fab, depreciation will weigh on gross margins by 3โ5 percentage points. The new capacity will not break even until it reaches 60โ70% utilization. That is a high hurdle for a company with limited financial depth.
However, the alternative is worse. If Sivers waits, the window closes. By 2027, TSMC will have ramped its COUPE platform, and the independent foundry space will become a commodity. Sivers needs to strike while demand is still constrained. The yield is not the prize, the exit is. The question is whether management can execute on both fronts: secure funding for the US expansion without diluting existing shareholders to death, and land enough CPO design wins to fill the new fabs.
Market Demand: The AI Spine
Let me quantify the opportunity. AI clusters like NVIDIA's GB200 NVL72 are pushing optical interconnect requirements through the roof. A single rack can carry tens of thousands of dollars in optical modules. That is a 10x jump from the old 100G world. LightCounting projects the CPO market to reach tens of billions by 2028. Sivers sits at the top of the value chain because the optical engine is 30โ50% of the BOM cost of a module.
The inventory cycle is currently in restocking mode. The supply bottleneck will persist until new capacity comes online, which I estimate will be 2025โ2026. This is a perfect storm for pricing power. During my DeFi farming days, I saw what happens when retail rushes into a yield farm without checking the smart contract. It ends badly. Here, the impulse is reversed: the market is ignoring a real, sustained demand wave because it does not know how to read InP yields.
Competition: The TSMC Shadow
TSMC's COUPE platform is the elephant in the room. With TSMC's scale, customer relationships, and capital, it can throw billions at silicon photonics. When it ramps, the competitive landscape will shift. But TSMC is not an expert in InP active integration. That is an acquired taste, not a core competency. Intel and Broadcom have their own silicon photonics efforts, but they are mostly internal, not open foundry services.
The independent foundry space is fragmented. IQE is a leader in III-V wafers, but it does not offer full device or photonic integration. Sivers' niche is the hybrid integration layer โ the place where the laser meets the silicon. It is a thin slice, but it is the most valuable slice of the stack.
Now, the contrarian angle. The activist investor Serenity criticized Sivers for over-indexing on Sweden. On the surface, that seems like a criticism of geographic focus. Dig deeper: Sweden's investor base is dominated by retail and local institutions who value stable dividends, not growth IPOs. They give Sivers a pension-fund multiple. The US market would give it a semiconductor-growth multiple. The difference is 2โ3x on price-to-sales, all else being equal.
That is the real thesis. Sivers is not operationally broken. It is structurally mispriced because of where it lists and who trades the stock. This is not a fundamental problem โ it is a clearinghouse problem. In my experience, that is exactly the kind of mispricing that active managers pray for. Liquidity evaporates when trust hits the floor. But trust in the technology is solid. It is the financial market that has not caught up.
Let me add a layer of my own: in 2022, during the Terra collapse, I had to make a decision in minutes. The algorithm that guided me was simple: sell first, ask questions later. Here, the same principle applies. When a stock is this undervalued relative to its forward order book, the question is not whether to hold, but how to size the position before the rest of the market discovers that the yield curve has turned.
Financial Metrics: The Hard Truth
The gross margin is likely in the 30โ40% range. That is healthy, but not spectacular. R&D intensity is high, around 15โ20% of revenue. The company is probably not free cash flow positive. The valuation looks expensive on trailing metrics โ a price-to-sales of 5โ8x, an EV/EBITDA of 20โ30x. But those trailing numbers are meaningless in a hypergrowth phase. What matters is the trajectory of the order book and the ability to convert that into revenue with sufficient gross margin.
If Sivers lands the US expansion and CPO ramps as expected, the revenue growth could triple over the next two years. The gross margin could expand as CPO products gain mix. The market cap would re-rate dramatically. But this is a binary bet: either management executes, or it doesn't. The risk of dilution is high. There is a finite chance that Sivers will need a rights issue, which would temporarily suppress the share price. But in the world of asymmetric bets, that dilution is a small price to pay for the option on the entire CPO market.
Geopolitics: The Bridge Play
Let me address the geopolitical chessboard. The US is pushing for domestic semiconductor independence. Europe has its own Chip Act. China is pouring billions into optical chip localization. Sivers does not fit neatly into any of these boxes. It is British-headquartered with a Swedish fab, a potential US site, and a key Chinese partner. That makes it a neutrality play. It can sell to everyone without alienating anyone. It is the equivalent of a Swiss bank account for photons.
In a world of fractured supply chains, that neutrality is valuable. The US wants to control AI infrastructure, but it cannot do so without optical components. China wants to build its own, but it needs advanced InP epitaxy. Sivers can play both sides. That may sound cynical, but cynicism is a trader's asset.
Now, the contrarian angle, more sharply: The market is treating Sivers as a cyclical foundry with no recurring revenue. But CPO is not a one-time order. Once a design is locked in, it stays for the life of the switch ASIC โ typically three to five years. That creates a sticky, annuity-like revenue stream. I know this from my work auditing smart contracts: the best protocols are those with deep liquidity and sticky user behavior. Sivers' orders have the same property. They are not one-off purchases; they are embedded in a hardware generation. The market does not see this, because it is focused on quarterly earnings. Profit is the receipt, not the purpose.
But let me not be Pollyanna-ish. There is a failure case. If TSMC's COUPE platform launches early and wins all major hyperscaler designs, Sivers may be left with meager crumbs. If the US fab takes too long to fund, the company could miss the entire CPO wave. If the Swedish market continues to dictate valuation, the stock may languish for years while the fundamentals improve. That is a compound option on the patience of capital.
The Takeaway: Two Levels, One Check
The first level is the technology. Sivers has a defensible moat in InP active integration. The second level is the market perception. The activism from Serenity is a catalyst. They are pushing for a strategic re-centering toward the US โ both in sales and in listing. If that happens, I expect the stock to re-rate significantly. I would look for a breakthrough above the current consolidation range on strong volume, and a trailing stop to protect against a false breakout. Alpha is found in the friction, not the flow. The friction here is the gap between the company's order book and its share price.
As an investor, you do not need to own every AI stock. You need to own the one that the market has mispriced. Sivers Photonics is that stock today. The ledger of physics is clear. The ledger of capital is distorted. I have seen this pattern before: in 2017, in 2020, in 2022. When the market finally adjusts the discount rate, the trade will be done. The question is whether you have the nerve to enter before the mark-to-market.
I do.
