Tracing the Optical Trail Back to the Genesis Block
The announcement landed with the quiet force of a tectonic shift. Largan Precision—the Taiwanese optics giant that has supplied the lion's share of smartphone camera lenses for over a decade—is partnering with TSMC on Co-Packaged Optics (CPO) technology. On the surface, this reads as another supply chain collaboration in the ever-expanding AI hardware ecosystem. But tracing the gas trail back to the genesis block, this is something far more consequential.
This is the moment when the semiconductor industry's center of gravity begins to shift from electron transport to photon transport. And the implications ripple far beyond the two companies involved.
The market reaction was telling. Largan's stock price moved on the news, but the broader semiconductor complex barely registered the signal. That's the tell. When a smartphone optics company with 30% global market share in camera lenses pivots toward AI datacenter infrastructure, and the world's most advanced foundry validates that pivot with a formal partnership, the market is looking at the wrong metrics. The real story isn't about lenses or packaging. It's about the fundamental architecture of AI compute—and who gets to own the interface layer between silicon and light.
Entropy increases, but the invariant holds. The invariant here is that AI compute density will continue to outpace the ability of traditional electrical interconnects to keep up. And the entropy is the growing chaos in the optical interconnect market as incumbents and newcomers scramble for position.
The Context: Why CPO Matters Now
To understand why this partnership matters, you have to understand the physics problem that AI datacenters are hitting. It's not a software problem. It's not even a chip design problem. It's a bandwidth and power problem at the interconnect level.
Current AI datacenters rely on pluggable optical transceivers—discrete modules that plug into switch faceplates and connect to compute nodes via fiber. These work, but they're hitting fundamental limits. The power consumption of driving electrical signals across a switch faceplate, through the PCB, and into the optical engine is becoming prohibitive as data rates climb from 400G to 800G to 1.6T per port. The electrical path between the switch ASIC and the optical module is a bottleneck that no amount of signal integrity engineering can fully solve.
Co-Packaged Optics changes the calculus. Instead of plugging optical modules into the faceplate, the optical engine—lasers, modulators, photodetectors—gets co-packaged directly onto the same substrate as the switch ASIC or compute chip. The electrical path shrinks from centimeters to millimeters. Power consumption drops by 30-50%. Latency drops. Bandwidth density increases.
This is the technology that TSMC has been developing under its COUPE (Compact Universal Photonic Engine) platform, first showcased at the 2024 North America Technology Symposium. And this is the technology that Largan is now helping to bring to market.
The timing is not accidental. NVIDIA's GB200 (Blackwell) platform, expected to ship in volume in 2025, will require massive optical interconnect density. Each GPU rack needs hundreds of optical channels. The traditional pluggable module ecosystem cannot scale to meet this demand without consuming unacceptable power and space. CPO is the answer.
Smart contracts don't govern physics. The laws of thermodynamics and electromagnetism are the ultimate arbiters here. And they're pointing squarely at CPO as the inevitable endpoint of AI datacenter evolution.
The Core Analysis: Deconstructing the Largan-TSMC CPO Partnership
The Technical Architecture
Let me be precise about what each party brings to this collaboration, because the division of labor reveals the strategic logic.
Largan Precision is not a semiconductor company. It's an optics company. For decades, it has designed and manufactured precision glass and plastic lens elements for smartphone cameras. Its core competency lies in optical design—understanding how light bends, focuses, and transmits through complex lens systems. This expertise is directly transferable to CPO, where the optical coupling between the laser source, the silicon photonic waveguide, and the fiber array requires precisely engineered micro-lens systems.
The specific contribution Largan makes to CPO is in the optical engine's coupling optics. When you co-package an optical engine with a switch ASIC, you need to couple light from the laser into the silicon photonic chip, and from the photonic chip into the fiber array. This coupling is one of the most challenging aspects of CPO. The alignment tolerances are measured in sub-microns. The thermal expansion of different materials must be managed. The optical loss at each interface must be minimized.
Largan's decades of experience in precision optical manufacturing—achieving sub-micron alignment in smartphone camera modules at scale—gives it a unique capability set for this challenge. The company has developed proprietary optical design IP, manufacturing processes, and quality control systems that are directly applicable to CPO optical engines.
TSMC brings the semiconductor packaging expertise. Its CoWoS (Chip-on-Wafer-on-Substrate) technology is the industry standard for 2.5D/3D advanced packaging, with over 90% market share. CPO is a natural extension of this capability. The optical engine needs to be integrated onto the same substrate as the switch ASIC, which requires the same kind of interposer technology that CoWoS provides, but with additional optical routing and thermal management considerations.
The COUPE platform that TSMC has been developing is specifically designed to address these challenges. It integrates the silicon photonic chip, the laser source, and the optical coupling elements into a single package that can be co-packaged with the switch ASIC. The platform leverages TSMC's expertise in advanced packaging, but it requires the optical design and manufacturing capabilities that Largan provides.
The Manufacturing Challenge
Here's where the analysis gets interesting from a technical risk perspective. CPO is not just a packaging problem. It's a multi-disciplinary challenge that spans silicon photonics, laser integration, fiber coupling, thermal management, and test methodology. Each of these areas has its own yield challenges, and the overall yield of a CPO module is the product of the yields at each step.
Based on my audit experience with complex multi-component systems, I can tell you that the yield math here is unforgiving. If the optical coupling step has a 95% yield, the laser integration has a 95% yield, and the packaging has a 95% yield, the overall yield is only 85.7%. And that's before you account for the test and burn-in steps, which typically add another 5-10% yield loss.
The industry benchmark for mature CPO production is likely to be around 90-95% overall yield. Getting there will require significant process development and iteration. TSMC's CoWoS packaging yield is already above 90%, but CPO adds new process steps that CoWoS doesn't have—specifically the optical coupling and alignment steps.
Largan's contribution here is critical. The company has developed manufacturing processes for precision optical components that achieve high yields at scale. Its experience with automated optical alignment, quality control, and process optimization is directly applicable to CPO manufacturing. This is not a trivial advantage. The companies that crack the CPO yield problem first will have a significant cost advantage that will be difficult to overcome.
The Cost Structure
Let me break down the economics of CPO versus traditional pluggable optics, because this is where the strategic implications become clear.
A traditional 800G pluggable optical transceiver costs roughly $800-1,200 per unit, depending on the vendor and the specific configuration. The bill of materials includes the optical engine, the DSP, the PCB, the housing, and the connector. The gross margin for established optical module vendors is typically 20-30%.
A CPO optical engine, by contrast, is expected to cost $500-1,000 per unit, but with significantly higher gross margins—potentially 40-60% for the companies that master the technology. The cost structure is different because the optical engine is integrated into the package rather than being a standalone module. This eliminates the housing, the connector, and the PCB, but adds the cost of the co-packaging process.
The value distribution is also different. In a traditional pluggable module, the DSP accounts for roughly 40-50% of the cost. In a CPO architecture, the DSP can be integrated into the switch ASIC or the optical engine, which shifts the value capture toward the packaging and optical components.
This is why the Largan-TSMC partnership is strategically significant. TSMC captures the packaging value, and Largan captures the optical component value. Together, they can offer a complete CPO solution that captures a larger share of the total value than any single company could achieve alone.
In the absence of trust, verify everything twice. The cost projections for CPO are based on assumptions about yield, volume, and learning curves that have not yet been validated at scale. The companies that succeed will be those that can iterate quickly and drive yields up faster than their competitors.
The Contrarian Angle: The Blind Spots in the CPO Narrative
The Apple Dependency Trap
Here's the uncomfortable truth that most analysts are missing. Largan's pivot to CPO is not a sign of strength—it's a sign of desperation. The company has been the dominant player in smartphone camera lenses for over a decade, with Apple as its largest customer, accounting for over 50% of revenue. But the smartphone market is saturated. Global smartphone shipments have been flat or declining for years. The era of multi-camera upgrades that drove Largan's growth is over.
The CPO partnership with TSMC is Largan's attempt to find a second growth curve before the first one flattens completely. This is a classic "pivot before you're forced to" strategy, and it's smart. But it also means that Largan is entering CPO from a position of relative weakness, not strength. The company's core business is mature, its growth is slowing, and its customer concentration is a structural risk.
The question is whether Largan can execute the pivot successfully. The company has the optical design expertise, but CPO requires different manufacturing capabilities, different customer relationships, and different go-to-market strategies than smartphone lenses. The learning curve is real, and the competition is formidable.
The TSMC Strategic Ambiguity
The other blind spot is TSMC's strategic intent. TSMC is the world's largest foundry, with a near-monopoly on advanced logic manufacturing. Its CoWoS packaging business is booming, driven by AI demand. But CPO is a different game. It requires optical expertise that TSMC doesn't have in-house, which is why it's partnering with Largan.
But here's the question: is TSMC fully committed to CPO, or is this a hedge? TSMC's core business is logic manufacturing, and its packaging business is a complement to that. CPO is a smaller market, at least initially, and it's not clear that TSMC sees it as a strategic priority rather than a nice-to-have.
The evidence suggests that TSMC is serious about CPO. The COUPE platform is a significant investment, and the company has been building out its CPO ecosystem with partners like Broadcom and Marvell. But the financial impact of CPO on TSMC's overall business will be minimal—less than 5% of revenue, even in the most optimistic scenarios. This means that TSMC's commitment to CPO could waver if other priorities emerge.
The Competitive Landscape
The third blind spot is the competitive landscape. The conventional wisdom is that Largan-TSMC has a first-mover advantage in CPO. But the reality is more complex. Intel has been developing silicon photonics for over a decade and has a significant head start in the technology. Broadcom has been shipping CPO switch chips for several years. Marvell has CPO DSP solutions. And the traditional optical module vendors—Innolight, Eoptolink, and others—are not sitting still.
The CPO market is expected to grow from $500 million in 2024 to $5 billion by 2028, a CAGR of about 60%. That's a big market, but it's not big enough to accommodate everyone. The companies that succeed will be those that can offer the best combination of performance, cost, and reliability. And the competition will be fierce.
The Largan-TSMC partnership has a unique advantage: the combination of optical design expertise and advanced packaging capability. But this advantage is not insurmountable. Intel has its own optical and packaging capabilities. Broadcom has deep relationships with the hyperscalers. And the traditional optical module vendors have manufacturing scale and customer relationships that Largan lacks.
The Market Opportunity: Sizing the CPO Prize
The AI Datacenter Demand
The demand for CPO is driven by the exponential growth in AI compute. NVIDIA's GB200 platform, which pairs two Blackwell GPUs with a Grace CPU, requires massive optical interconnect density. Each GB200 rack needs hundreds of optical channels, and the power budget for these interconnects is a significant constraint.
The math is compelling. A typical AI datacenter with 10,000 GPUs needs roughly 5,000-10,000 optical transceivers, depending on the architecture. At current power consumption levels, the optical interconnect accounts for 10-15% of total datacenter power consumption. CPO can reduce this by 30-50%, which translates to significant cost savings over the lifetime of the datacenter.
The market size projections are staggering. LightCounting forecasts that the CPO market will grow from $500 million in 2024 to $5 billion by 2028, a CAGR of about 60%. This growth will be driven by the adoption of CPO in AI datacenters, starting with the largest hyperscalers and gradually spreading to smaller operators.
The Value Chain Shift
The shift to CPO will reshape the optical interconnect value chain. Traditional optical module vendors—Innolight, Eoptolink, Accelink—will face significant disruption. Their core products, pluggable optical transceivers, will be gradually replaced by CPO solutions. This is not an immediate threat—CPO won't reach scale until 2026-2027—but the direction is clear.
The winners in this transition will be the companies that can offer complete CPO solutions. This includes the optical engine manufacturers (Largan, Intel, Broadcom), the packaging companies (TSMC, ASE), and the system integrators (NVIDIA, Broadcom, Marvell). The losers will be the companies that are too invested in the traditional pluggable module ecosystem.
The Largan Opportunity
For Largan, the CPO opportunity is transformative. The company's current revenue is heavily concentrated in smartphone lenses, with Apple accounting for over 50% of total revenue. The CPO business has the potential to diversify this revenue base and position Largan as a key player in the AI infrastructure ecosystem.
The financial impact could be significant. If Largan captures 20-30% of the CPO optical engine market, that translates to $1-1.5 billion in revenue by 2028, compared to the company's current revenue of approximately $5 billion. More importantly, the CPO business would carry higher margins than the smartphone lens business, potentially boosting Largan's overall profitability.
The valuation impact could be even more significant. Largan currently trades at 20-25x forward earnings, reflecting its status as a mature smartphone component supplier. If the market begins to view Largan as an AI infrastructure play, the multiple could expand to 30-35x, which would represent a significant re-rating of the stock.
The Geopolitical Dimension: Taiwan's Strategic Pivot
The Semiconductor Sovereignty Question
The Largan-TSMC partnership has a geopolitical dimension that is often overlooked. Taiwan is the world's most important semiconductor manufacturing hub, with TSMC producing over 90% of the world's most advanced logic chips. This concentration of manufacturing capacity is a strategic vulnerability, and the Taiwanese government has been actively encouraging its semiconductor companies to diversify into new areas.
CPO is one such area. By extending Taiwan's semiconductor ecosystem into optical interconnect technology, the Largan-TSMC partnership helps to solidify Taiwan's position in the global AI infrastructure supply chain. This is not just a commercial decision—it's a strategic one.
The Export Control Risk
The geopolitical risk is not one-way. The United States has been increasingly aggressive in using export controls to limit China's access to advanced semiconductor technology. So far, CPO technology has not been included in any export control lists, but this could change.
If the US were to impose export controls on CPO technology, it would affect both Largan and TSMC. However, as Taiwanese companies, they would be in a better position than Chinese companies to navigate these restrictions. The US has a strong interest in maintaining Taiwan's semiconductor ecosystem as a counterweight to China, and it's unlikely to take actions that would undermine this.
The China Factor
China is also investing heavily in CPO technology, with companies like Innolight and Accelink developing their own solutions. The Chinese government's "Big Fund" has been providing financial support for semiconductor development, including optical interconnect technology.
However, China faces significant challenges in CPO development. The country lacks access to advanced semiconductor manufacturing equipment, and its domestic foundry ecosystem is not as mature as Taiwan's. This gives the Largan-TSMC partnership a significant advantage in the CPO market.
The Financial Analysis: What the Numbers Say
Largan's Financial Position
Largan's financial position is solid, but not spectacular. The company has a gross margin of approximately 60-65%, which is high for a manufacturing company, but down from the 70%+ levels it achieved in 2019. The decline reflects increased competition in the smartphone lens market and the maturation of the smartphone industry.
The company's operating cash flow is approximately $500-600 million per year, with a free cash flow of $300-400 million. This gives Largan the financial flexibility to invest in CPO development without straining its balance sheet. The company has minimal debt and a strong cash position.
The key financial question is whether Largan can maintain its profitability while investing in CPO. The company's R&D spending is approximately 5-8% of revenue, which is reasonable for a technology company. The CPO investment will require additional R&D spending, but the company's strong cash flow should be able to support this.
TSMC's Financial Position
TSMC's financial position is much stronger. The company has a gross margin of approximately 55-60%, operating cash flow of $300-400 billion per year, and a free cash flow of $100-150 billion. The company's capital expenditure is approximately 35-40% of revenue, reflecting its massive investment in advanced manufacturing capacity.
The CPO investment is a small part of TSMC's overall capital expenditure, but it's strategically important. The company is investing in CPO to extend its advanced packaging leadership and to capture value from the AI infrastructure buildout.
The Valuation Question
The valuation question is whether the market is correctly pricing the CPO opportunity. Largan currently trades at 20-25x forward earnings, which is reasonable for a mature company with moderate growth prospects. But if the market begins to price in the CPO opportunity, the multiple could expand significantly.
The key variable is the timing of CPO revenue. If Largan can generate meaningful CPO revenue by 2026-2027, the market is likely to re-rate the stock. If CPO revenue is delayed, the stock will continue to trade at its current multiple.
The Competitive Landscape: Who Wins and Who Loses
The Incumbents
The traditional optical module vendors—Innolight, Eoptolink, Accelink—are the most exposed to CPO disruption. These companies have built their businesses on pluggable optical transceivers, and they have significant manufacturing scale and customer relationships. However, they lack the advanced packaging capabilities that CPO requires.
The incumbents have a few options. They can partner with packaging companies to develop CPO solutions. They can acquire CPO technology. Or they can focus on the traditional pluggable module market, which will continue to exist for several more years.
The most likely outcome is a mix of these strategies. The incumbents will continue to serve the traditional market while developing CPO capabilities in-house or through partnerships. The companies that can successfully transition to CPO will survive; the ones that can't will be marginalized.
The New Entrants
The new entrants in the CPO market include Intel, Broadcom, Marvell, and the Largan-TSMC partnership. These companies have different strengths and weaknesses.
Intel has been developing silicon photonics for over a decade and has a significant technology lead. However, the company has struggled to commercialize its optical technology, and its foundry business is still developing.
Broadcom has deep relationships with the hyperscalers and has been shipping CPO switch chips for several years. The company's CPO solutions are already in production, giving it a first-mover advantage.
Marvell has CPO DSP solutions and has been working with TSMC on advanced packaging. The company is well-positioned in the CPO market, but it lacks the optical component expertise that Largan brings.
The Largan-TSMC partnership is the most interesting new entrant. The combination of Largan's optical design expertise and TSMC's advanced packaging capability is unique. If the partnership can execute on its technology roadmap, it has the potential to be a major player in the CPO market.
The Competitive Dynamics
The competitive dynamics in the CPO market are still being defined. The technology is in its early stages, and the market is not yet large enough to support all the players. The companies that succeed will be those that can offer the best combination of performance, cost, and reliability.
The Largan-TSMC partnership has a unique advantage: the combination of optical design expertise and advanced packaging capability. This is a difficult combination to replicate, and it gives the partnership a significant competitive moat.
However, the partnership also has weaknesses. Largan is a newcomer to the AI infrastructure market, and it lacks the customer relationships that companies like Broadcom and Marvell have. TSMC's commitment to CPO is not as strong as its commitment to logic manufacturing, and the company could shift its priorities if other opportunities arise.
The Roadmap: What to Watch
Near-Term Signals (1-3 Months)
The first signal to watch is whether Largan announces any CPO business development progress. The company has been relatively quiet about its CPO plans, and any announcement would be a positive signal.
The second signal is whether TSMC updates its CPO technology roadmap at its next technology symposium. The company has been developing the COUPE platform, and any updates on the platform's progress would be significant.
The third signal is whether NVIDIA or AMD announces adoption of CPO technology. These companies are the primary drivers of AI infrastructure demand, and their adoption of CPO would be a major validation of the technology.
Medium-Term Signals (3-12 Months)
The medium-term signals are about execution. The first is whether Largan's CPO products pass customer validation. This is a critical milestone, and any delays would be a negative signal.
The second is whether TSMC's CPO production capacity is on track. The company has been building out its advanced packaging capacity, and any delays in CPO capacity would be a negative signal.
The third is whether the CPO market size data is updated. LightCounting and other market research firms provide regular updates on the CPO market, and any significant revisions would be important.
Long-Term Signals (12+ Months)
The long-term signals are about market adoption. The first is the penetration rate of CPO in AI datacenters. This will be the ultimate test of whether CPO delivers on its promise.
The second is the revenue contribution of CPO to Largan's overall business. If CPO becomes a significant revenue driver, it will validate the company's strategic pivot.
The third is the evolution of the CPO technology roadmap. The technology is still in its early stages, and the roadmap will evolve as the industry learns more about what works and what doesn't.
The Takeaway: The Invariant Holds
The Largan-TSMC partnership is a bet on the future of AI infrastructure. It's a bet that optical interconnect will become the dominant technology for AI datacenter networking, and that the companies that control the optical component and packaging layers will capture significant value.
The bet is not without risk. CPO technology is still in its early stages, and the yield, cost, and reliability challenges are significant. The competitive landscape is crowded, and the market is not yet large enough to support all the players. And the geopolitical environment is uncertain, with the potential for export controls and other disruptions.
But the invariant holds. AI compute density will continue to outpace the ability of electrical interconnects to keep up. The physics of the problem is clear, and the direction of travel is inevitable. The companies that position themselves at the intersection of optics and semiconductors will be the winners in the next phase of the AI infrastructure buildout.
Code is law until the reentrancy attack. In the semiconductor industry, the code is the technology roadmap, and the reentrancy attack is the unexpected disruption that changes everything. The Largan-TSMC partnership is a bet that the roadmap will hold, and that the disruption will come from the companies that are best positioned to execute on it.
The question is not whether CPO will happen—it's who will capture the value. And the Largan-TSMC partnership has as good a chance as anyone. The combination of optical design expertise and advanced packaging capability is unique, and the timing is right. The next 24 months will tell us whether this bet pays off.
Optimism is a feature, not a bug, until it fails. The CPO market is projected to grow at 60% CAGR, and the Largan-TSMC partnership is positioned to capture 20-30% of that market. But the projections are based on assumptions that have not yet been validated. The yield challenges, the competitive dynamics, and the geopolitical risks could all derail the optimistic scenario.
The smart money is watching the signals. The near-term signals—business development announcements, technology roadmap updates, customer adoption—will tell us whether the optimistic scenario is playing out. The medium-term signals—customer validation, production capacity, market data—will tell us whether the technology is delivering on its promise. And the long-term signals—market penetration, revenue contribution, technology evolution—will tell us whether the bet has paid off.
In the meantime, the invariant holds. Entropy increases, but the physics of AI compute demand will drive the industry toward optical interconnect. The companies that are best positioned to capture this value will be the winners. And the Largan-TSMC partnership is one of the best-positioned players in the game.