The numbers surged, but the room felt empty. Largan Precision's stock ticked upward on the news, a familiar pattern in a market that rewards narrative over substance. The announcement was brief, almost clinical: Largan, the world's largest smartphone lens maker, was partnering with TSMC on co-packaged optics. CPO. Three letters that promise to rewire the data center, and by extension, the entire digital economy that blockchain networks depend upon.
I've spent twenty-seven years watching infrastructure stories like this one unfold. The pattern is always the same. A technology emerges from the lab, the market assigns it a narrative, and the narrative outruns the physics. But every once in a while, the physics catches up. And when it does, the quiet companies — the ones that have been grinding on tolerances and yields for decades — become the most important players in the room.
This is one of those moments. And it matters for reasons that go far beyond a Taiwanese lens maker and a Taiwanese foundry. It matters because the next wave of decentralized infrastructure — the AI networks, the compute markets, the verifiable data layers — will be built on physical substrates that most crypto natives have never thought about. The graph spikes. The soul remains quiet. But the soul is where the real work happens.
The Context: Two Giants, One Unlikely Intersection
Let me be precise about who we're talking about, because the contours of this partnership define its significance.
Largan Precision is not a household name outside of supply chain circles, but it is one of the most important companies in the world that you've never heard of. Based in Taichung, Taiwan, Largan controls roughly 30% of the global smartphone camera lens market. It is the primary lens supplier for Apple's iPhone — a relationship that accounts for more than half of Largan's revenue. The company's gross margins have historically hovered between 60% and 70%, a testament to the extraordinary difficulty of manufacturing precision optical elements at scale. A single lens module for a flagship smartphone requires tolerances measured in nanometers — the curvature of a lens element, the alignment of multiple elements within a barrel, the coating uniformity across a surface measured in millimeters. Largan has spent four decades perfecting this craft.
TSMC needs no introduction to anyone reading this. The Taiwan Semiconductor Manufacturing Company is the world's largest contract chipmaker, producing the advanced logic chips that power everything from iPhones to NVIDIA's AI accelerators. TSMC's dominance in advanced packaging is even more pronounced than its dominance in logic. The company's CoWoS (Chip-on-Wafer-on-Substrate) technology — which enables the 2.5D and 3D packaging of multiple chiplets onto a single substrate — commands more than 90% of the global market. Every NVIDIA H100 and GB200 GPU relies on CoWoS packaging. Every AI data center built in the last two years depends on TSMC's ability to package increasingly complex chip stacks.
The partnership between these two companies is, on its face, an odd one. Largan makes lenses. TSMC makes chips. But the intersection of their capabilities — optical design and semiconductor packaging — is precisely where co-packaged optics lives.
CPO is the next evolutionary step in data center interconnect technology. Today, data centers use pluggable optical transceivers — small modules that convert electrical signals to optical signals and back again, connecting switches and servers across racks. These modules work, but they consume significant power and introduce latency. As AI workloads scale, the power and latency costs of pluggable optics become prohibitive. NVIDIA's GB200 platform, for example, requires thousands of optical connections per rack. The energy budget for interconnect alone threatens to overwhelm the compute budget.
CPO solves this problem by integrating the optical engine directly onto the same substrate as the switch or compute chip. Instead of plugging a transceiver into a faceplate, the optical engine — lasers, modulators, photodetectors, and the lenses that couple light between them — is co-packaged with the silicon. This eliminates the electrical-optical-electrical conversion losses, reduces power consumption by 30-50%, and cuts latency by an order of magnitude.
The catch is that CPO is extraordinarily difficult to manufacture. It requires silicon photonics, micro-ring modulators, fiber coupling, thermal management, and test methodologies that span two entirely different engineering disciplines. Optical engineers think in terms of wavelengths and diffraction limits. Semiconductor engineers think in terms of doping profiles and critical dimensions. Bridging these worlds requires a company that understands both — or a partnership between two companies that each understand one side deeply.
This is where Largan and TSMC come in. TSMC brings the packaging expertise, the CoWoS platform, and the manufacturing scale. Largan brings something equally valuable: decades of experience designing and manufacturing precision optical elements. The lenses that couple light from a laser into a fiber, or from a fiber into a photodetector, are not off-the-shelf components. They require the same kind of nanoscale precision that Largan has been perfecting for smartphone cameras since the early 2000s.
The partnership is not a coincidence. It is a recognition that CPO is a systems problem, not a component problem. And systems problems require systems thinkers.
The Core: A Seven-Dimensional Analysis of the CPO Partnership
I want to take you through this partnership the way I would analyze any infrastructure play — through the lens of technical capability, supply chain dynamics, capacity planning, market demand, geopolitical exposure, competitive positioning, and financial sustainability. This is the framework I've used for two decades, and it has never failed to reveal the truth beneath the narrative.
Technical Capability: The Art of the Possible
Let me start with what Largan and TSMC actually bring to the table, technically speaking.
On the TSMC side, the relevant technology is the COUPE platform — Compact Universal Photonic Engine — which TSMC first showcased at its 2024 North America Technology Symposium. COUPE is TSMC's answer to the question of how to integrate optical engines with advanced logic. The platform leverages TSMC's CoWoS packaging infrastructure, but extends it with silicon photonics capabilities. The idea is to create a standardized platform that can host optical engines from multiple partners, much the way CoWoS hosts chiplets from multiple suppliers.
The technical challenges are formidable. Optical coupling — the process of aligning a laser's output to a fiber or waveguide with sub-micron precision — is one of the hardest problems in manufacturing. A misalignment of even a few hundred nanometers can reduce coupling efficiency by 50% or more. Thermal management is another challenge. Lasers are temperature-sensitive; their wavelength shifts with temperature, and if the wavelength drifts too far, the entire optical link fails. Co-packaging a laser with a switch chip that dissipates hundreds of watts creates a thermal nightmare that requires innovative cooling solutions.
This is where Largan's expertise becomes critical. Largan has spent decades manufacturing lens elements with sub-micron tolerances. The company's optical design IP — the mathematical models that describe how light bends, focuses, and couples — is among the most sophisticated in the world. Transferring this expertise to the design of optical coupling lenses for CPO is not a stretch; it is a natural extension of what Largan already does.
The yield question is the elephant in the room. CPO is a new technology, and new technologies have terrible yields. TSMC's CoWoS packaging yields have matured to above 90%, but CPO involves entirely new processes — optical coupling, laser integration, fiber attachment — that have not yet been optimized. If Largan's optical engines yield below 90%, the cost structure of the entire CPO solution becomes uncompetitive. My assessment, based on the trajectory of similar technologies, is that yields will reach manufacturable levels — above 95% — by 2026. But the path there will be painful, and companies that cannot tolerate the early yield losses will drop out.
The deeper point is that CPO is not a single technology. It is a stack of technologies — silicon photonics, micro-ring modulators, fiber coupling, thermal management, test and measurement — each of which has its own learning curve. The companies that succeed will be the ones that can integrate the entire stack, not just the individual components. Largan and TSMC, together, have the best shot at this integration because they bring complementary expertise to the table.
Supply Chain Dynamics: The Fragility of the New
Every infrastructure story has a supply chain dimension, and CPO is no exception. The CPO supply chain is a hybrid of two existing supply chains — optical components and semiconductor packaging — and it inherits the vulnerabilities of both.
On the optical side, the critical material is SOI — silicon-on-insulator — which is the substrate for silicon photonics. SOI wafers are produced by a small number of suppliers, led by France's Soitec and Japan's Shin-Etsu. This is a concentrated supply chain, and any disruption — a natural disaster, a trade dispute, a capacity shortage — would ripple through the entire CPO ecosystem.
On the semiconductor side, the critical dependencies are the equipment and materials that TSMC already relies on. ASML's lithography systems, Applied Materials' deposition tools, and a complex web of specialty gases and chemicals. These dependencies are well-understood and largely de-risked for TSMC, which has decades of experience managing them. But the combination of optical and semiconductor supply chains creates new failure modes. A shortage of optical-grade glass, for example, would not affect TSMC's logic business but would cripple Largan's CPO production.
The bargaining power dynamics are worth noting. Largan's current customer concentration — more than 50% of revenue from Apple — is a strategic vulnerability. The CPO partnership with TSMC is, in part, a hedge against this concentration. By diversifying into AI data center optics, Largan reduces its dependence on the smartphone market, which has been stagnant for years. TSMC, for its part, gains a trusted optical partner that can help it close the CPO ecosystem loop.
The supply chain assessment is, on balance, moderately fragile. The CPO supply chain is not yet mature, and the concentration of critical inputs — SOI substrates, optical-grade materials, specialized equipment — creates vulnerabilities. But the companies involved are sophisticated operators with deep experience managing complex supply chains. The risk is manageable, but it is real.
Capacity and Capital: The Price of Entry
Infrastructure stories are ultimately capital stories. The companies that win are the ones that can fund the capacity buildout before the demand arrives.
Largan's current capacity utilization for smartphone lenses is running at 70-80%, reflecting the softness in the global smartphone market. This is actually good news for the CPO transition — it means Largan has manufacturing capacity that can be repurposed for optical engine production. The company's capital expenditure intensity, historically 10-15% of revenue, will need to increase as it builds out CPO production lines. This will pressure free cash flow in the near term, but the long-term payoff — higher margins and a more diversified revenue base — justifies the investment.
TSMC's situation is different. The company's advanced packaging capacity is running at above 90% utilization, driven by AI demand. Adding CPO capacity means adding to an already strained system. TSMC's capital expenditure intensity, at 35-40% of revenue, is among the highest in the semiconductor industry. The company has committed to expanding CoWoS capacity to 80,000 wafers per month by 2025, with CPO accounting for an estimated 10-20% of that capacity.
The equipment delivery timeline is a critical constraint. CPO production lines require specialized equipment — lithography, coating, coupling, and test systems — with lead times of 6-12 months. The equipment is not currently subject to export controls, but this could change. The capacity ramp timeline is 12-18 months from equipment installation to full production, which puts the first meaningful CPO volumes in the second half of 2025 and the first half of 2026.
The depreciation impact is worth noting. Both Largan and TSMC use accelerated depreciation schedules — 5-7 years for Largan, 5 years for TSMC. The initial CPO production lines will carry significant depreciation loads, which will pressure gross margins by 2-3 percentage points in the early years. The breakeven point is expected in 2026, when volumes reach scale and yields mature.
Market Demand: The Certainty of the Inevitable
The demand side of the CPO equation is the most compelling part of the story. AI compute demand is not a hypothesis; it is a fact. NVIDIA's GB200 platform, which began shipping in volume in 2025, requires CPO-grade optical interconnects. Each GB200 GPU needs multiple optical engines, and the total addressable market for CPO is enormous.
The market projections are striking. LightCounting, the industry's leading market research firm, projects CPO market growth from $500 million in 2024 to $5 billion in 2028 — a compound annual growth rate of approximately 60%. This is one of the fastest-growing segments in the entire semiconductor industry.
The demand is not just from AI training. AI inference — the process of running trained models on new data — is exploding as AI applications move from research to production. Inference workloads require massive data center interconnect capacity, and CPO is the technology that makes this feasible within power budgets.
The structural shift is important to understand. CPO is not an incremental improvement over pluggable optics; it is a fundamental re-architecture of the data center. Traditional optical module manufacturers — companies like Zhongji Innolight and Eoptolink in China — face an existential threat. Their products will be replaced by CPO solutions over the next 3-5 years. The window for these companies to adapt is narrow, and the technology gap is wide.
For Largan, the market opportunity is transformative. The company's CPO revenue could grow from essentially zero in 2024 to $1 billion by 2028, representing a significant share of the projected $5 billion market. This would transform Largan from a smartphone lens company into an AI optics company — a re-rating that the market has not yet fully priced in.
Geopolitical Exposure: The Taiwan Question
No analysis of a Taiwan-based partnership is complete without addressing the geopolitical dimension. Taiwan is the most strategically important piece of real estate in the global semiconductor industry, and the concentration of advanced chip manufacturing on the island is a source of profound geopolitical anxiety.
The current export control regime does not cover CPO technology. Neither Largan nor TSMC is on the U.S. Bureau of Industry and Security's Entity List. The equipment and materials required for CPO production are not currently subject to export controls. But the trajectory of U.S. export controls — which have expanded steadily since 2022 — suggests that CPO technology could eventually be added to the list, particularly if it is deemed critical to AI capabilities.
The irony is that export controls designed to limit China's AI capabilities could have the side effect of constraining the very companies that are building the infrastructure for the free world's AI ecosystem. This is the paradox of technological nationalism: the tools we use to protect our advantage can also undermine it.
The Taiwan strait question is the elephant in every room where Taiwanese technology is discussed. A disruption to Taiwan's semiconductor industry would be catastrophic for the global economy — and for the blockchain networks that increasingly depend on the physical infrastructure of data centers. The probability of a full-scale conflict remains low, but the tail risk is severe. Companies like Largan and TSMC are aware of this risk and are taking steps to mitigate it — TSMC is building fabs in Arizona and Japan, and Largan is exploring manufacturing options outside Taiwan. But the reality is that the concentration of advanced semiconductor manufacturing in Taiwan is a structural vulnerability that cannot be fully hedged.
Competitive Landscape: The Race to Scale
The CPO competitive landscape is taking shape, and Largan and TSMC are not the only players.
Intel has been investing in silicon photonics for over a decade and has a credible CPO roadmap. The company's integrated photonics research has produced some of the most advanced silicon photonics devices in the industry. Intel's advantage is its vertical integration — it designs and manufactures its own photonics, which gives it control over the entire stack.
Broadcom is another formidable competitor. The company is the dominant supplier of switch chips for data centers, and it has been developing CPO solutions that integrate optical engines directly with its switch silicon. Broadcom's CPO switch chips are expected to enter production in 2025, and the company has the customer relationships — with hyperscalers like Google and Meta — to drive adoption.
Marvell is focused on the DSP side of the equation, developing CPO solutions for the digital signal processing that underpins optical communications. Marvell's expertise in DSP and its relationships with optical module manufacturers give it a position in the CPO ecosystem.
The Chinese optical module manufacturers — Zhongji Innolight, Eoptolink, and others — are also investing in CPO, but they face significant technology gaps. The Chinese CPO ecosystem is supported by government funding through the National Integrated Circuit Industry Investment Fund, but the fundamental challenges of optical coupling, yield management, and advanced packaging remain.
The competitive assessment is that Largan and TSMC have a first-mover advantage, but not a moat. The technology is still early, and the competitive landscape will shift as the market matures. The key differentiator will be the ability to scale production while maintaining yields — a capability that favors companies with deep manufacturing experience.
Financial Sustainability: The Re-rating Story
The financial picture for Largan is the most interesting part of the story from an investment perspective.
Largan's current gross margins are approximately 60-65%, down from over 70% in 2019. The decline reflects intensifying competition in the smartphone lens market and pricing pressure from customers like Apple. The company's return on equity has declined from over 25% to approximately 15-20% over the same period.
The CPO opportunity changes the financial trajectory. CPO optical engines are expected to carry gross margins of 60-70% — comparable to Largan's historical smartphone lens margins. The CPO business could lift Largan's overall margin profile and reverse the declining ROE trend.
The valuation story is compelling. Largan currently trades at 20-25x trailing earnings, below its historical average of 25-30x. The market has been discounting Largan as a mature smartphone component supplier with limited growth prospects. The CPO opportunity challenges this narrative. If Largan successfully transitions to an AI optics company, the valuation multiple could re-rate to 30-35x — a 50% upside from current levels.
TSMC's financial picture is more stable but less exciting. CPO is expected to account for less than 5% of TSMC's revenue by 2028, so the financial impact on TSMC is modest. But the strategic impact is significant. CPO extends TSMC's advanced packaging franchise and deepens its relationships with AI customers. The financial markets have already recognized this — TSMC's stock has outperformed the broader semiconductor index over the past year.
The Contrarian Angle: What the Optimists Are Missing
Every infrastructure story has a contrarian angle, and the Largan-TSMC CPO partnership is no exception. Let me offer three uncomfortable truths that the market narrative tends to gloss over.
First, the yield problem is real, and it will not be solved quickly. CPO is not an incremental improvement over existing technology; it is a fundamental re-architecture of the data center interconnect. The yield challenges — optical coupling, laser integration, thermal management — are not the kind of problems that get solved in a quarter or two. They require years of process development and manufacturing learning. The companies that succeed will be the ones that can tolerate the early yield losses and the associated cost penalties. The companies that cannot will drop out, and the market will consolidate around a few winners.
Second, the "decentralization" narrative is a myth when it comes to physical infrastructure. The blockchain community talks a lot about decentralization, but the physical infrastructure that powers the digital economy is becoming more concentrated, not less. TSMC controls 90% of advanced packaging. Largan controls 30% of the global lens market. The CPO partnership consolidates this concentration further. The irony is that the decentralized networks we are building depend on physical infrastructure that is increasingly centralized in a single geographic region — Taiwan. This is a vulnerability that the crypto community has not fully grappled with.
Third, the geopolitical risk is underpriced. The market treats the Taiwan strait risk as a tail event with low probability. But the probability is not zero, and the consequences would be catastrophic. A disruption to Taiwan's semiconductor industry would not just affect chip prices; it would affect every digital system that depends on advanced computing — including blockchain networks. The crypto community's reliance on physical infrastructure that is concentrated in a geopolitically volatile region is a systemic risk that deserves more attention than it receives.
The contrarian view is not that the Largan-TSMC partnership will fail. On the contrary, I believe it is one of the most strategically important partnerships in the semiconductor industry. The contrarian view is that the path to success is longer and more painful than the market narrative suggests, and that the risks — yield, concentration, geopolitics — are more severe than the optimists acknowledge.
The Takeaway: Infrastructure Is Destiny
I have spent twenty-seven years watching infrastructure stories unfold. I have seen technologies rise and fall, companies succeed and fail, narratives outrun reality and reality eventually catch up. The pattern is always the same: the graph spikes, the soul remains quiet, and the quiet work — the grinding, the iterating, the failing and trying again — determines who wins.
The Largan-TSMC CPO partnership is a quiet story. It does not have the drama of a token launch or the excitement of a protocol upgrade. But it matters more than most of the stories that dominate the crypto headlines. It matters because the next wave of decentralized infrastructure — the AI networks, the compute markets, the verifiable data layers — will be built on physical substrates. And the companies that control those substrates will shape the digital economy for the next decade.
The question for the crypto community is whether we are paying attention. The infrastructure that powers our networks is being built by companies like Largan and TSMC — companies that most crypto natives have never heard of. The concentration of this infrastructure in Taiwan is a systemic risk that we have not fully priced in. The yield curves and capacity ramps of CPO production will determine the cost and availability of AI compute, which will in turn determine the economics of decentralized AI networks.
When the graph spikes, the soul remains quiet. But the soul is where the real work happens. The real work of building the infrastructure for the next decade is happening in cleanrooms in Taichung and Hsinchu, where engineers are solving problems that most of the world does not know exist. The market will eventually recognize this work. The question is whether we will be ready when it does.
The optics of power are shifting. The companies that understand the intersection of light and silicon will define the next era of computing. Largan and TSMC understand this intersection better than anyone. The rest of us — the builders, the investors, the dreamers of decentralized futures — would do well to understand it too.