The CPO Lottery: Why Sivers Photonics' Supply Bottleneck Is Not a Moat

The logic held until the ledger lied.

Here is the setup. A specialty foundry in Sweden. Indium phosphide substrates. Hybrid integration of active and passive photonics. A supply bottleneck that has average selling prices rising. Six new pluggable customers walking through the door. A roadmap that places the company inside the co-packaged optics revolution that every AI hyperscaler is racing toward. The demand is real. The scarcity is real. The pricing power is real.

And yet the market prices this company as a Swedish small-cap with a cyclical order book.

That disconnect is not an inefficiency. It is information.

When I spent forty hours decompiling the Golem v0.9 smart contracts in 2017, I identified three integer overflow vulnerabilities in the token distribution logic. The anonymous team had raised $8.6 million on a whitepaper that described a decentralized supercomputer network. The bytecode described something much more fragile. I learned a rule that has guided my analysis ever since: whitepaper promises rarely match bytecode reality.

The same rule applies to Sivers Photonics. The narrative is the whitepaper. The financial statements, the customer concentration tables, and the competitive timeline are the bytecode. Trace the hash, ignore the hype.

Sivers Photonics trades on London's AIM market under the ticker SIVE. It is a III-V compound semiconductor (InP/GaAs) and silicon photonics foundry. It sits upstream of the co-packaged optics supply chain — the technology that AI data centers need to move data between accelerators without exhausting the power budget. The market narrative says this is a structural winner in the AI infrastructure buildout. The technical reality says something more complicated.

I have been here before. In mid-2021, I reverse-engineered the Bored Ape Yacht Club smart contract and discovered that the JSON metadata pointing to 10,000 token images was hosted on a centralized server with no IPFS backup. One server outage could render the entire collection inaccessible. I published the forensic breakdown. Trading volume in unrelated blue-chip NFTs dropped 40% as the market realized the underlying infrastructure was fragile.

The lesson: the market prices the story, not the infrastructure. Sivers is the infrastructure. The question is whether the infrastructure holds when the demand wave arrives.

Part II: The Company

Let me establish what Sivers actually is.

The company operates as a specialty foundry. It processes InP and GaAs wafers, plus silicon photonics. Its value proposition is hybrid integration — combining InP gain media with silicon photonic passive devices on a single platform. This is the technical path that co-packaged optics depends on.

CPO solves a specific problem. AI clusters need massive bandwidth between accelerators. Traditional pluggable optical modules consume too much power and occupy too much board space. The solution is to place the optical engine on the same substrate as the switch chip. That is CPO. Near-package optics (NPO) is the transitional step. Sivers provides the optical engine chips that CPO manufacturers need.

The demand story is genuinely strong. LightCounting projects the CPO market will reach multi-billion dollar scale by 2028. AI compute is in a super-cycle running 2024 to 2028. Optical interconnect is the nervous system of the AI cluster. A single NVIDIA GB200 NVL72 rack carries optical module value in the tens of thousands of dollars, versus thousands for traditional configurations.

The customer list includes Ayar Labs — the most prominent CPO startup, with Intel lineage — and O-Net, a Chinese optical module manufacturer. The company references six new pluggable customers. The supply situation is tight. ASPs are rising. Utilization is reportedly high.

The critique comes from Serenity, an investment research firm. Their argument: Sivers is over-focused on the Swedish market. The longer it spends time in Sweden, the more its valuation is constrained by that market's investment logic. The recommendation: pivot to the United States, where institutional investors understand CPO and will pay growth multiples.

The Serenity critique is partly correct. It is also partly a distraction. The market shift is a real consideration. It does not address the structural questions that will determine whether Sivers survives the competitive wave it faces.

Part III: The Technology — Where the Moat Actually Is

Let me start with the technical claims.

Sivers is not a CMOS logic foundry. Process node concepts do not apply. The relevant metrics are waveguide dimensions, coupling efficiency, and integration density. This is a different language from the 3nm/5nm race that dominates semiconductor discourse.

The core strength is InP active device integration. This matters because InP is the material that emits light. Silicon can guide light, modulate light, and detect light. It cannot efficiently emit light. That requires InP. Hybrid integration — bonding InP gain media onto silicon photonic circuits — is the path that CPO takes.

Sivers has established IP in InP active devices and silicon photonic passive devices. This is the foundry's real asset. It is not a processor design house. It does not license ARM cores. Its IP library is photonic design elements that enable customer-specific integration.

The gap analysis is where the narrative starts to fray.

The industry benchmark is TSMC's COUPE platform — Compact Universal Photonic Engine — scheduled for mass production in 2025. GlobalFoundries has a 45nm silicon photonics platform that is already production-ready. Intel and Broadcom have their own silicon photonics teams with multi-billion dollar R&D budgets.

Sivers is a small specialty foundry. Its total R&D spending is in the millions of dollars. Intel and Broadcom each spend billions. The absolute scale gap is not bridgeable.

In silicon photonics integration density — the number of on-chip devices — Sivers lags TSMC by roughly one to two generations. The company does not have a COUPE-equivalent platform. It does not have a 45nm silicon photonics process. What it has is InP active integration expertise that places it in the first tier alongside Intel and Broadcom in that specific sub-domain.

The yield question remains unanswered. Silicon photonics foundry yields typically run 85 to 95 percent. InP active integration yields are lower — 70 to 85 percent. Sivers has not publicly disclosed yield data. For a specialty foundry, yield is the difference between profitable orders and value-destroying ones. The supply bottleneck narrative assumes the company can convert its capacity into revenue at acceptable yield. That assumption is unverified.

Here is where I bring in the 2025 custody audit experience. When I audited the cold-storage protocols of the top three ETF custodians, I found that two firms used 3-of-5 multi-sig wallets but shared the same private key generation seed. The security theater was impressive. The actual security was a single point of failure. The same principle applies to yield claims in specialty foundries. The press release says the process works. The wafer maps tell the real story. Without access to the wafer maps, the yield claim is marketing.

Code does not lie; auditors do. The same is true for process engineers.

There is also the question of the technology roadmap. The article references CPO manufacturers expanding through 2028. This implies Sivers is positioning its process platform for high-bandwidth, low-power optical engines designed for the CPO era. That is the right direction. The question is whether the roadmap can be executed before the competitive window closes.

The next-generation substrate question is critical. InP-on-silicon — heterogeneously integrating thin-film InP onto silicon photonic chips — is the key technical path for the CPO era. Sivers' accumulated expertise in this specific area is its core asset. It is also the area where TSMC, Intel, and Broadcom are all investing heavily. The technical edge is real. The durability of that edge is unproven.

Part IV: The Industry Chain — Squeezed From Both Sides

Sivers sits in the foundry segment of the optical communications value chain. This is a high-value position. Optical chips account for 30 to 50 percent of the bill of materials for optical modules and CPO devices. The profit pool distribution favors the chip layer: approximately 35 to 40 percent of industry profit sits at the optical chip and device tier, versus 30 to 35 percent for modules and 20 to 25 percent for system integration.

But the position is not comfortable.

Upstream, Sivers depends on InP substrates. The supply is dominated by Japanese firms — Sumitomo Electric, JX Nippon Mining. This is a concentrated supplier base. The company also depends on high-precision lithography equipment from ASML, Canon, and Nikon, and epitaxial growth equipment from Aixtron and Veeco. None of this equipment has a mature alternative source.

The upstream dependence is structural. InP substrate supply is not a commodity market. It is an oligopoly. The number of qualified suppliers is small. The qualification process for a new substrate supplier takes years. Sivers cannot easily switch suppliers. It cannot pressure suppliers on price. The input cost structure is largely dictated to the company.

Downstream, the customer base is concentrated. The reference to six new pluggable customers suggests the company is diversifying, but the reality is that a handful of customers — Ayar Labs, O-Net, and a few others — dominate revenue. My estimate: the top five customers account for 70 to 80 percent of revenue. The largest single customer may represent 20 to 30 percent.

This is the same structural vulnerability I identified in the Terra/Luna collapse. When I mapped the $40 billion depeg through wallet clusters in May 2022, I found three insiders who had exited positions hours before the crash. The lesson was not about those specific wallets. It was about what concentrated positions do to a system. Concentrated exposure is a bomb with a delayed fuse. It works perfectly until it detonates.

For Sivers, the concentration risk is existential. Lose Ayar Labs — or lose O-Net — and the revenue impact is immediate and severe. The six new pluggable customers are a positive signal. They are not yet a structural fix.

The company's bargaining power is moderate. In a supply-constrained market — which is the current situation — Sivers has temporary pricing power. ASPs are rising. The bottleneck gives the foundry leverage. But this is cyclical, not structural. When new capacity comes online — when TSMC's COUPE scales, when Chinese competitors mature — the bargaining power will shift back to the customers.

The supply chain security assessment reveals a medium-high vulnerability rating. The core equipment and materials are import-dependent. The good news: optical communications chip manufacturing does not fall under advanced process export controls. Sivers is not subject to EUV or sub-3nm restrictions. The equipment acquisition path is accessible. But the InP substrate dependency is a latent vulnerability. If Japan ever restricted InP exports — an unlikely but not impossible scenario — Sivers' capacity would be severely impacted.

There is also the China angle. O-Net is a Chinese company. The collaboration between Sivers and O-Net on ELS products represents a deep binding with the Chinese optical communications supply chain. In a geopolitically tense environment, this is both an opportunity and a risk. The opportunity: access to the Chinese market. The risk: American investors may be wary of a company with significant Chinese exposure.

Part V: The Capacity Problem — No Room to Grow

The supply bottleneck narrative has a dark side. High utilization means the company has no spare capacity. Revenue growth is capped by what the existing fabs can produce.

The article references two fab capacity allocation — implying Sivers has or plans dual-site capacity in Sweden and the United States. The Swedish fab is operational. The US fab is unconfirmed — it exists as inference from the Serenity recommendation to pivot to the US market.

If a US fab is in the plan, the capital expenditure burden is significant. Specialty foundry capex-to-revenue ratios run 20 to 30 percent. A new fab requires equipment purchases — DUV lithography, MOCVD epitaxial tools — with delivery lead times of 6 to 12 months. From equipment installation to volume production typically takes 12 to 18 months. A US fab announced today would not contribute revenue until 2025 or 2026 at the earliest.

And here is the depreciation problem. Semiconductor equipment depreciates on a 5 to 7 year straight-line schedule. A major capex program would suppress gross margins by 3 to 5 percentage points during the ramp. The breakeven utilization for a new fab is 60 to 70 percent. Until the new fab reaches that level, it is a margin drag, not a profit center.

The capital expenditure question intersects with the cash flow question. Specialty foundries typically operate with tight cash flow. High capex plus long receivable cycles produce negative free cash flow. Sivers' operating cash flow to net income ratio is likely in the 0.5 to 1.0 range. Free cash flow is likely negative. If the company pursues a US fab, it will need external financing. That means equity dilution.

Silence in the logs is the loudest scream. In financial terms, silence in the cash flow statement is the loudest scream. A company that cannot generate free cash flow cannot fund its own growth. The supply bottleneck narrative assumes the company can monetize scarcity. The financial reality is that the company may need to dilute shareholders to capture the opportunity it claims to have.

The capacity allocation question is also strategic. If Sivers has limited capacity and strong demand, it must choose which customers to prioritize. High-value CPO customers or pluggable optical module customers? The choice determines the revenue mix and the long-term positioning. Prioritize CPO and the company locks into the growth market. Prioritize pluggable modules and the company maximizes near-term revenue in a market that is transitioning to CPO. This is a strategic decision that the company has not fully disclosed.

Part VI: The Market — Real Demand, Wrong Assumptions

The demand side is the strongest part of the thesis.

AI compute spending is in a genuine super-cycle. The optical interconnect requirement scales with cluster size. CPO penetration is projected to move from under 5 percent in 2024 to 20 to 30 percent by 2028. Ayar Labs' expansion plans through 2028 indicate the transition from pilot to scale.

This is structural growth. The optical communications industry historically grew at 8 to 10 percent CAGR. The AI-driven forecast is 15 to 20 percent CAGR for 2024 to 2028. CPO is the incremental driver.

The application distribution is instructive. Pluggable optical modules for data centers — 800G and 1.6T upgrades — represent the largest revenue share at 50 to 60 percent, growing 15 to 20 percent. CPO is 10 to 15 percent of revenue but growing at over 50 percent. ELS — external laser source — the companion technology to CPO, is 10 to 15 percent growing at over 30 percent. This is a portfolio that benefits from both the current pluggable market and the future CPO transition.

The inventory cycle is in the restocking phase. AI demand plus supply constraints equals downstream customers actively building inventory. Optical module manufacturers have healthy inventory levels — actually low inventory days, because supply is tight. The supply bottleneck is expected to persist until 2025 or 2026, when new capacity comes online. The historical inventory cycle in optical communications runs 2 to 3 years: 2019 de-stocking, 2021 re-stocking, 2023 de-stocking, 2024 re-stocking.

But there is a mismatch between the demand narrative and the company's ability to capture it. The question is not whether CPO will grow. It is whether Sivers will be the supplier that captures that growth. That brings us to the competitive landscape.

Part VII: The Competitive Siege — TSMC's Shadow

TSMC's COUPE platform is the existential threat.

TSMC is the world's largest foundry. It has unmatched scale, capital, and customer relationships. Its COUPE platform — targeted for 2025 mass production — is designed specifically for co-packaged optics. When TSMC enters a market, it does not enter quietly. It enters with the capacity to dominate.

The competitive landscape for silicon photonics is already crowded. Intel and Broadcom have production silicon photonics. GlobalFoundries has a 45nm platform. IQE dominates III-V foundry with over 30 percent market share. Win Semiconductors has over 20 percent. Sivers ranks third or fourth in III-V foundry and fifth or lower in silicon photonics foundry.

In InP active integration, Sivers is in the first tier. But the first tier is a crowded space. Intel, Broadcom, and TSMC all have InP integration capabilities. The differentiation is in the specific hybrid integration process — the know-how that makes InP-on-Si work reliably at scale.

The R&D comparison is brutal. Sivers likely spends 15 to 20 percent of revenue on R&D — a healthy ratio for its size, but the absolute amount is in the millions. TSMC's R&D budget is in the tens of billions. Intel and Broadcom each spend billions on silicon photonics alone. Sivers cannot outspend its competitors. It can only out-focus them.

Every exploit is a history lesson in slow motion. The competitive dynamic here is the same pattern I see in crypto markets: a small player with a niche technology gets a window of opportunity, and then the large players arrive with the resources to absorb the niche. The question is whether the small player has built enough of a moat before the big players arrive.

The TSMC COUPE timeline is the countdown clock. If COUPE hits its 2025 mass production target and lands major CSP customers — NVIDIA, Google, Microsoft — the CPO optical engine market will consolidate quickly. Sivers' differentiation in InP active integration may not be sufficient to survive that consolidation.

The probability assessment: 60 to 70 percent that TSMC's COUPE platform succeeds as planned. The hedge for Sivers is to deepen its binding with early CPO customers like Ayar Labs and to accelerate US capacity to stay close to key customers. Neither hedge is guaranteed to work.

There is also the new entrant threat. Chinese optical chip manufacturers — Yuanjie Technology, Accelink — are advancing with policy support from China's National Fund III (344 billion yuan). NVIDIA and Broadcom may develop in-house optical engines. The competitive pressure is not just from TSMC. It is from every direction.

The five forces analysis is sobering. Industry competition: intense. TSMC, Intel, Broadcom, IQE all present. Buyer bargaining power: strong. Large customers with concentrated purchasing. Supplier bargaining power: strong. InP substrate and equipment suppliers with oligopolistic positions. Substitution threat: medium. VCSELs can replace silicon photonics in short-reach scenarios, though CPO favors silicon photonics. New entrant threat: high. TSMC's entry into CPO reshapes the landscape.

Sivers operates in a niche — InP active integration. The niche is defensible in the short term. The long term is a siege.

Part VIII: The Financial Reality — A Story Without Numbers

The financial disclosure from the company is thin. The original article provides no specific margin data, no cash flow figures, no balance sheet details.

My estimates, based on industry benchmarks for specialty foundries: gross margin in the 30 to 40 percent range. R&D expense ratio of 15 to 20 percent of revenue — high for the industry, reflecting the company's innovation-dependent position. Operating cash flow tight. Free cash flow negative. ROE below 5 percent. ROIC in the 3 to 5 percent range, below the cost of capital of 10 to 12 percent.

The valuation picture is uncomfortable. Price-to-sales in the 5 to 8x range. EV/EBITDA in the 20 to 30x range. Price-to-book of 3 to 5x. These multiples exceed IQE, the III-V foundry comparable, which trades at roughly 2x book and 3x sales. The market is pricing in significant CPO-driven growth.

The question is whether the growth justifies the multiple. The company is barely profitable — possibly loss-making. The cash flow is negative. The competitive threats are material. A 5 to 8x price-to-sales multiple on a company with negative free cash flow and a major competitive wave approaching is a bet on execution, not a valuation of current reality.

The gross margin trajectory is the key metric to watch. If the supply bottleneck persists and ASPs continue to rise, gross margins should expand. If the mix shifts toward high-value CPO and ELS products, gross margins could reach 40 percent or higher. But this requires execution. It requires yield improvement. It requires customer retention.

I audited the cold-storage protocols of ETF custodians in 2025 and found that two of the top three firms had compromised key generation processes. The institutional entry into crypto had not solved the fundamental security hygiene problems. The same pattern applies to growth valuations in the photonics sector. The narrative has moved ahead of the operational reality. The market is paying for the story. The story has not yet been validated by the numbers.

The R&D capitalization question is worth examining. Small-cap tech companies often capitalize development costs to boost reported profit. If Sivers is capitalizing R&D, the profit quality is lower than reported. If the company is expensing R&D in full, the profit quality is higher. My assumption: the company is conservative, expensing R&D in full. But this is an assumption, not a verified fact.

The working capital dynamics are also a concern. Specialty foundries carry inventory for customer commitments. They extend payment terms to large customers. The operating cash flow conversion is typically below 1.0. If Sivers is growing revenue, working capital absorbs cash. Growth is expensive.

Part IX: The Geopolitical Angle — The US Pivot

The Serenity recommendation to move to the US market has a geopolitical dimension that deserves examination.

Sivers is a UK company with Swedish operations. It is not on the US Entity List. Its technology — optical communications chips — is not in the core scope of US export controls, which target 3nm logic, GAA, and EUV-related technology. The company uses DUV lithography, not EUV. Its equipment supply chain is accessible through European and American suppliers.

The InP substrate supply is a vulnerability. Japanese firms dominate. If geopolitical conditions deteriorate — if Japan were to restrict InP exports — Sivers' capacity would be severely impacted. That scenario is unlikely, as InP substrates are not military-sensitive items. But the vulnerability exists.

China's export controls on gallium and germanium do not directly affect Sivers, since its core material is InP. However, China's National Fund III is targeting advanced packaging and optical communications chips. This will cultivate Chinese competitors. Companies like Yuanjie Technology and Accelink are moving into the same space with policy support.

The regional production trend is clear. The US CHIPS Act provides $52 billion for domestic semiconductor manufacturing. The European Chips Act provides 43 billion euros. Japan has its semiconductor revival plan. China has the National Fund III. Every major economy wants to strengthen domestic semiconductor supply chains. Sivers, as a UK-Swedish company, sits in the European camp. Its Swedish base benefits from European policy support.

The US pivot makes sense from a risk perspective. The US offers CHIPS Act subsidies, a more stable policy environment, and proximity to key customers like Ayar Labs. A US fab would provide a hedge against European and Chinese risks.

But the pivot is not a solution to the fundamental problems. Moving headquarters or adding US capacity does not fix the customer concentration issue. It does not address the TSMC threat. It does not improve cash flow. It changes where the company operates, not how the company performs.

The decoupling risk assessment is medium-low. Optical communications chips are not in the core scope of advanced process export controls. A full decoupling is unlikely. But if the US expands controls to include AI-related infrastructure — and if CPO is classified as part of AI compute infrastructure — Sivers could face indirect compliance requirements as a CPO supply chain participant. As a non-Chinese company, the risk is manageable. But it is not zero.

The deeper issue is the investor structure argument. Serenity's critique — that the Swedish market's investment logic constrains Sivers' valuation — is essentially an argument about who prices the stock. Swedish retail and local institutions may not understand CPO. American institutional investors may pay a growth multiple.

This is a real dynamic. I observed the same pattern in crypto markets: the same asset priced differently on different venues based on the investor base. The valuation is not a pure reflection of fundamentals. It is a function of who is trading and what they believe.

But moving to the US does not automatically create value. It changes the venue. The company must still execute. The market shift is a narrative adjustment, not a fundamental improvement.

Governance is just a slower attack vector. In this context, the investor structure is the governance. The Swedish market's inability to price CPO growth is a structural constraint. The US market's willingness to pay growth multiples is also a structural feature. Neither is a verdict on the company's operational quality.

Part X: Contrarian — What the Bulls Got Right

Now let me steelman the bull case. Because the bears — including myself — can be too clever by half.

The bulls argue that Sivers is a scarce asset in a supply-constrained market. The supply bottleneck is real. ASPs are rising. The company has six new pluggable customers. It is deeply embedded with Ayar Labs, the most credible CPO startup. The InP active integration capability is genuinely differentiated. TSMC's COUPE platform is a threat, but TSMC has not yet demonstrated the InP hybrid integration know-how that Sivers has.

The scarcity argument has teeth. In a market where demand exceeds supply, capacity is pricing power. Sivers can choose which orders to accept. It can prioritize high-value CPO customers. The bottleneck gives the company a window to build customer relationships and lock in long-term supply agreements.

The InP moat is also underappreciated. InP epitaxial growth is not a commodity process. It requires years of accumulated know-how. The defect density, the doping profiles, the quantum well design — these are not things that can be replicated quickly. Intel and Broadcom have deep pockets, but they are not guaranteed to match Sivers' specific InP integration yield.

The customer binding with Ayar Labs is a real asset. Ayar Labs is the leading CPO startup with Intel lineage. Its 2028 expansion plan suggests CPO is moving from pilot to scale. Sivers, as the optical engine chip supplier, is positioned to ride that wave.

The supply chain position is genuinely high-value. Optical chips are 30 to 50 percent of the optical module BOM. The profit pool sits at the chip layer. Sivers is upstream of the competition in the value chain.

And the valuation argument has merit. If the company executes on its CPO roadmap and the market shifts to the US, the re-rating potential is significant. The gap between the Swedish market's valuation and what a US institutional investor would pay for a CPO-linked growth company could be substantial. This is not a pure narrative play — the demand is real, the technology is real, and the scarcity is real.

The bulls are right about the direction. The question is the magnitude and the timing. The window is real. The question is whether Sivers can execute within the window before the competitive wave arrives.

The one thing the bulls have that the bears do not: a verifiable demand signal. The supply bottleneck is not a projection. It is happening now. ASPs are rising now. Customers are placing orders now. The question is whether this current strength translates into durable competitive advantage or whether it is a cyclical peak that will reverse when capacity catches up.

Part XI: The Takeaway

The Sivers thesis is a race. The company has a genuine technical position in InP active integration. It has a genuine window of supply scarcity. It has a genuine customer relationship with a leading CPO player. All of this is real.

But the race is against time and scale. TSMC's COUPE platform is the countdown. The customer concentration is the fragility. The cash flow constraint is the limiter. The company needs to convert its window of scarcity into durable customer relationships, expand its customer base, and secure the financial resources to fund its growth — all before the competitive wave arrives.

The US pivot is a useful narrative adjustment. It is not a solution. The market shift will not fix the yield question, the customer concentration, or the cash flow problem. It will simply change who prices the company.

My assessment: Sivers is a real technology company in a real growth market with a real competitive threat and a real financial constraint. The supply bottleneck gives it a window. The window will close. The question is what Sivers builds before it closes.

The logic held until the ledger lied. Here, the logic is the demand story. The ledger is the financial statements. The ledger has not yet validated the story. Until it does, the valuation is a bet on execution — and execution is the one thing that cannot be verified from a press release.

The market will eventually separate the story from the substance. In 2026, we will know whether Sivers converted its bottleneck into a moat or watched it evaporate as TSMC's COUPE scaled. The data will be in the wafer maps, the margin reports, and the customer diversification metrics. That is where the truth lives.

Trace the hash. Ignore the hype.