The Cost of Seeing Clearly: Why Largan's CPO Pivot Demands More Than Optics

We didn't see it coming. Not the technology itself—the roadmap was visible in plain sight. We didn't see the existential calculation behind it. When Largan Precision, the company that puts the glass on nearly every flagship smartphone, quietly announced a co-packaged optics partnership with TSMC, the crypto and AI communities treated it as another headline in the endless parade of hardware announcements. We read the specs, we noted the timeline, we moved on. But we missed the deeper signal: this is not a product launch. This is a survival story being written with silicon and light. For two decades, Largan owned the lens market. A 30% global share in mobile lenses, with Apple alone representing over half of its revenue. Its gross margins have hovered between 60% and 65%—numbers that would make most semiconductor companies weep with envy. Yet those margins are eroding. The smartphone market is saturated, growing at low single digits. The company's historical 70%+ margins from 2019 have drifted down to 60% as competition in optical design intensifies. This is the context of the CPO partnership. When Largan enters the co-packaged optics world with TSMC, it is not pursuing innovation for innovation's sake. It is running from a shrinking mirror. But the deeper context is technological. Co-packaged optics represents a fundamental shift in how data moves inside AI data centers. The current architecture uses pluggable optical modules—small, removable devices that convert electrical signals to optical ones at the edge of the switch. As AI training models scale exponentially, these pluggable modules become a bottleneck. They consume enormous power, generate significant heat, and limit bandwidth density. The solution, according to TSMC and its partners, is to place the optical engine directly on the same substrate as the computing chip. This is CPO. Light is no longer leaving the package to travel through cables; it is generated, modulated, and coupled entirely within a silicon photonics engine co-packaged with the switch or GPU. The result is a 30% to 40% reduction in power consumption for interconnects and a latency measured in pico-seconds rather than nanoseconds. This is not an incremental improvement. It is a fundamental shift in how data moves. TSMC has already shown its roadmap. In 2024, at the North American Technology Forum, the foundry presented its COUPE platform—Compact Universal Photonic Engine. The initial phase targets Ethernet switches with a potential extension to compute and GPU chips. The plan is to begin small-scale production in 2025. Largan's addition to this ecosystem completes a critical gap in optical design. TSMC controls the packaging, the process, and the silicon photonics elements. Largan brings over three decades of expertise in precision optical design—the lenses, the coupling mechanisms, and the tolerances required to push light into a fiber from a chip with minimal loss. This is not a simple supply agreement. It is the marriage of two very different engineering disciplines. Let me be direct about the core analysis here. Based on my years of observing supply chain shifts, the success of CPO will not be determined by the silicon. It will be determined by the optical engine. The value chain splits roughly into two: the optical engine itself, which includes the laser, the photodetector, and the lens assembly, represents 30% to 40% of the module cost, while the advanced packaging and substrate assembly represents the remaining 40% to 50%. In the traditional pluggable module market, the major players were Chinese companies like Innolight and Eoptolink, who achieved scale on mature 100G and 400G technologies. Their gross margins are 20% to 30%. CPO changes the competitive landscape completely. The optical engine for CPO is expected to command a price point of $500 to $1,000 per unit—2 to 3 times the price of a traditional pluggable module—with a projected gross margin of 60% to 70%. Based on my audit experience and my early investment analysis, here is the critical insight: this is not about who has the best technology. It is about who can control the interface between the lens and the substrate. The coupling tolerance is measured in microns. If the lens is misaligned by even one micron, the signal loss makes the entire system uneconomical. Largan's core competency—the ability to mass-produce glass and plastic lenses with sub-micron precision for smartphones—is directly transferable to this optical coupling challenge. This is why the TSMC partnership is strategic rather than convenient. TSMC does not have the optical design heritage to solve this coupling problem at scale. Largan does. And Largan cannot solve the substrate challenge without TSMC's CoWoS packaging, which holds over 90% of the advanced packaging market. The mutual dependency creates a moat that neither Intel nor Broadcom can easily cross. Now, here is the contrarian angle that most coverage misses. This partnership is not primarily about the technology. It is about the supply chain risk and the geopolitical reality of the island of Taiwan. We keep discussing CPO as a technical upgrade, but the entire AI supply chain is fundamentally fragile. The substrate used for silicon photonics—the SOI substrate—is not manufactured by TSMC or Largan. It is dominated by a French company. The supply chain for co-packaged optics involves a dual system: the semiconductor packaging system, which is dominated by Taiwan, and the optical system, which is dominated by China. The fragmentation of these two systems is the biggest risk to CPO adoption. If there is a conflict or if export controls are extended to include photonics, this partnership could be interrupted. Largan and TSMC are not just building a product. They are building a Taiwan-centric supply chain for the AI optical era, an attempt to secure strategic autonomy in a technology that will underpin every AI data center. The strategic significance of Taiwan as the manufacturing hub for both compute and light is a defense against a future where the US and China are decoupled. This is the hidden narrative. I also want to address the market size projection. LightCounting forecasts the CPO market will grow from $500 million in 2024 to $5 billion by 2028, a CAGR of about 60%. If this partnership captures 20% to 30% of that market, we are talking about a revenue opportunity of $1 billion to $1.5 billion for Largan by 2028. Compare that to Largan's current annual revenue of roughly $2 billion, which is overwhelmingly from smartphones. This is not a side business. This is a second growth curve that could double the company's revenue within five years. The market seems to be underpricing this optionality. Largan's current valuation is 20-25x P/E, which is at the low end of its historical range. If the market begins to value Largan as an AI optical company rather than a smartphone lens maker, the multiple could expand to 30-35x. That is a 50% upside from the multiple expansion alone, before the revenue growth is even factored in. But I want to be careful with the optimism. The biggest risk is not technology or demand. It is execution. CPO is an early-stage technology. The yield rate for the optical engine is a crucial factor. The industry standard for mature photonics is around 95%, but CPO's new processes will likely start below 90%. If Largan's optical engine cannot achieve a yield above 90% in early production, the entire cost structure of the CPO module will be undermined. The depreciation pressure will also be significant. Both companies will need to invest heavily. Largan will need to build new production lines for optical engines, which will reduce free cash flow in the short term, but the long-term opportunity is substantial. We should also consider the timeline. The 2025-2026 period is the true test. TSMC's COUPE platform is scheduled for 2025, and the AI roadmap of NVIDIA with the GB200 Blackwell platform will require photonic interconnect. If NVIDIA adopts CPO in 2025, this will be a major catalyst. If not, the timeline could slip to 2026. The window for traditional pluggable modules is two to three years, after which the structural shift will occur. So, what are we watching? We are watching the yield data. We are watching the qualification of Largan's optical engine. We are watching whether TSMC's CPO capacity reaches a meaningful percentage of the overall CoWoS capacity. We are watching whether the AI chips adopt this. But the fundamental story is clear. The convergence of optical and semiconductor is happening, and a company that was once defined by the lens in your hand is now betting its future on the light between machines. The question is not whether this transition will happen. The question is who will control the optics that connect the AI systems. Largan and TSMC are making their claim. We should be watching with the same precision.