Sivers Photonics: The InP Foundry Caught Between CPO's Promise and TSMC's Shadow

At block 1,000,000 of the AI infrastructure buildout, the market is finally realizing that the bottleneck isn't the GPU—it's the light. Sivers Photonics (SIVE), a London AIM-listed III-V and silicon photonics foundry, sits at the intersection of this realization. But the market's perception of this company is a study in structural mispricing, driven less by fundamentals and more by the investor ecosystem it happens to be listed in. The recent analysis from Serenity, which criticized Sivers for being 'overly focused on the Swedish market,' cuts to the core of a valuation disconnect that has nothing to do with the company's technical capabilities and everything to do with the lens through which it is viewed. Sivers is not a logic foundry. It does not compete on 3nm or 5nm nodes. Its currency is waveguide dimensions, coupling efficiency, and integration density. The company's core competency lies in hybrid integration—the ability to bond InP gain media onto silicon photonic passive structures. This is the foundational technology for Co-Packaged Optics (CPO) and the external laser sources (ELS) that will power the next generation of AI data centers. In this specific niche, Sivers is not a laggard; it is a first-tier player. The analysis correctly identifies that while Sivers trails TSMC's COUPE platform by 2-3 years in large-scale CMOS-compatible silicon photonics, it holds a leading position in InP active integration, a domain where the Taiwanese giant lacks core expertise. Tracing the supply chain back to its genesis, the structural advantage becomes clear. The photonics value chain places Sivers in the highest-margin segment. Optical chips account for 30-50% of the BOM cost of a transceiver, and the profit pool is concentrated at the chip level (35-40%) rather than in module assembly. The analysis notes a 'supply bottleneck and ASP increase,' which is the classic signal of a demand-supply imbalance. When capacity utilization exceeds 90% and pricing power is on the rise, the economic value of each order increases. This is the 'order economic value' that Serenity argues the market is failing to price. The company's expansion from a single or few large customers to 'six new pluggable customers' is a critical de-risking event, moving it away from a precarious dependency on Ayar Labs or O-Net. However, dissecting the atomicity of this growth narrative reveals a more complex picture. The competitive landscape is not static. TSMC's COUPE platform, slated for mass production in 2025, is the elephant in the room. With its scale, financial muscle, and existing relationships with hyperscalers, TSMC could commoditize the CPO optical engine market, squeezing the niche that Sivers currently occupies. The analysis rates this risk as high, with a 60-70% probability of TSMC's success. This is not a question of if, but when. Sivers' defense lies in its InP epitaxial growth know-how and its deep co-design relationships with CPO pioneers like Ayar Labs. But this is a defensive position, not an offensive one. The moat is real but narrow. The contrarian angle here is not about technology but about market structure. The analysis suggests that Sivers' valuation is constrained by its listing on the Swedish market, where the investor base—dominated by retail and local institutions—cannot properly value a high-growth, loss-making CPO play. This is a classic structural discount. The recommendation to pivot to the US market is not just about accessing deeper capital pools; it is about aligning the company with an investor base that understands the AI infrastructure narrative and is willing to pay a growth multiple. The hidden implication is that the 'two fabs' mentioned in the analysis—likely Sweden and a planned US site—are not just about capacity but about strategic signaling. A US fab would not only bring Sivers closer to customers like Ayar Labs but also unlock access to CHIPS Act funding and a more favorable valuation regime. Yet, the market's focus on the 'Swedish discount' may be missing a more fundamental risk. The analysis highlights that Sivers' ROIC is below its WACC, meaning the company is currently destroying value. The financial foundation is thin, with negative free cash flow and a reliance on external funding. A US expansion would require significant capital expenditure, likely leading to dilution. The depreciation from a new fab would suppress gross margins by 3-5 percentage points until utilization reaches 60-70%. This is the classic growth trap: the very expansion needed to capture the CPO opportunity could strain the balance sheet and delay profitability. The market is not just discounting for geography; it is discounting for execution risk. Looking forward, the key variable is not Sivers' technology but the pace of CPO adoption. The analysis predicts CPO penetration will rise from under 5% in 2024 to 20-30% by 2028, driven by the insatiable bandwidth demands of AI clusters. If this timeline holds, Sivers is in a prime position. But if TSMC accelerates its COUPE ramp or if the AI capex cycle cools, the window could close. The company's fate is tied to the broader AI infrastructure supercycle, and its current valuation is a bet on that future. The question is not whether Sivers has the technology—it does. The question is whether it can execute its expansion without being crushed by the very giants it seeks to serve. The bridge between promise and profitability is a pessimistic oracle, and right now, it is whispering a warning about the cost of capital.