CPO Emerges as AI Infrastructure Bottleneck: Citi Names Lumentum and Coherent
The next bottleneck in artificial intelligence infrastructure may be moving away from GPUs and toward the optical systems that connect them.
As AI clusters become larger and data moves between accelerators at increasingly higher speeds, conventional electrical connections face growing bandwidth, power and thermal constraints. That is pushing the industry toward co-packaged optics (CPO) and near-packaged optics (NPO), technologies that place optical components much closer to computing and switching silicon.
Citi's latest research on the optical communications supply chain highlights Lumentum Holdings (NASDAQ: LITE) and Coherent (NASDAQ: COHR) as key beneficiaries of this transition. Citi maintains Buy ratings on both companies, citing tight supplies of indium phosphide (InP) lasers and strong demand from 800G, 1.6T and emerging CPO/NPO architectures.
For investors, the important point is not simply that CPO is another AI theme. The more significant development is that optical connectivity is becoming a potential constraint on how quickly AI computing capacity can scale.
Why CPO Matters for the Next Phase of AI
AI data centers are no longer primarily about installing more GPUs. Thousands or even hundreds of thousands of accelerators need to communicate with one another rapidly.
As bandwidth requirements rise, copper connections become increasingly difficult to scale because of signal loss, power consumption and physical limitations.
CPO attempts to address this by integrating optical engines much closer to switching or computing chips. Instead of sending high-speed electrical signals over relatively long distances before converting them into optical signals, the architecture moves optical conversion closer to the source.
The potential benefit is higher bandwidth density and improved power efficiency.
But CPO also creates a more complicated manufacturing challenge. Lasers, optical engines, photonic integrated circuits, packaging, coupling and testing all have to work reliably at very high volumes.
That is where the investment opportunity identified by Citi becomes particularly relevant.
InP Lasers Could Become the Critical Supply Constraint
Citi's research estimates that industry laser-chip supply could rise from approximately 279 million units in 2025 to around 1.3 billion by 2028, representing roughly a 66% compound annual growth rate. Even with substantial capacity expansion, Citi estimates that EML demand could still exceed supply by about 30% in 2026, 20% in 2027 and 11% in 2028.
EML, or electro-absorption modulated laser, is an important optical component used to transmit high-speed data.
A supply deficit does not automatically mean prices will continue rising indefinitely. New manufacturing capacity, alternative technologies and changing architecture choices can eventually alter the balance.
However, persistent supply tightness can give established suppliers greater pricing power and potentially support margins.
That is one reason Citi's thesis extends beyond simply betting on the growth of AI data centers.
Why Citi Likes Lumentum
Lumentum is positioned particularly close to the laser bottleneck.
Citi identifies the company as a core beneficiary of growing demand for high-speed EML and ultra-high-power lasers used in AI optical infrastructure. The research highlights Lumentum's position in 100G and 200G EML as well as its external-light-source technology for CPO and NPO systems.
The company's own recent commentary also points to increasing CPO and NPO demand.
At Citi's 2026 Global TMT Conference, Lumentum said the timing of CPO deployment had not changed, but the volume and intensity of demand were increasing. Management expects CPO to be used in scale-out switches initially and later in optical scale-up applications.
Lumentum also said CPO and NPO are likely to coexist rather than one immediately replacing the other. It expects CPO to represent roughly half of the optical scale-up market and NPO the other half in the company's view.
That distinction matters for investors. A slower-than-expected CPO rollout does not necessarily eliminate the entire optical opportunity if NPO and conventional high-speed optical modules continue expanding.
Coherent Has a Different Advantage
Coherent offers another way to participate in the optical AI infrastructure cycle.
Citi points to its 6-inch InP wafer platform, vertical integration and expanding internal manufacturing capacity as important advantages. The larger wafer platform has the potential to increase chip output and lower manufacturing costs compared with traditional smaller-wafer production.
Coherent also has exposure across multiple parts of the optical ecosystem, including InP, silicon photonics, lasers and optical modules.
That diversification is important because the industry has not yet settled on one architecture for every AI workload.
CPO, NPO, pluggable optics and other approaches can coexist depending on the distance, bandwidth, thermal requirements and serviceability needs of a particular system.
NVIDIA Is Helping Accelerate the Optical Supply Chain
NVIDIA is an important part of this story.
Citi's analysis highlights NVIDIA's strategic investments in both Lumentum and Coherent, designed in part to secure access to critical laser capacity and support the development of next-generation optical infrastructure.
The connection is strategically significant because NVIDIA's AI systems increasingly require networking technologies capable of moving enormous amounts of data between accelerators.
Citi expects NVIDIA's CPO roadmap to continue progressing, with qualification and risk-production activity in 2026 followed by increasing laser volumes and broader deployment in 2027–2028.
That timeline is still an expectation rather than a guaranteed industry outcome.
The pace of adoption will depend on manufacturing yields, packaging capacity, customer qualification and the economics of deploying CPO at scale.
CPO Is Not the Only Optical Technology Investors Should Watch
One potential mistake is treating CPO as an immediate replacement for every existing optical technology.
Industry participants increasingly describe the transition as layered.
TrendForce expects AI data-center optical interconnects to move progressively from pluggable optics toward LPO/LRO, NPO and CPO architectures. It also identifies yield, thermal management, optical coupling, testing and reliable field-serviceable connections as important challenges for mass production.
Marvell CEO Matt Murphy recently described NPO and CPO as technologies that are likely to coexist for years rather than as mutually exclusive alternatives. Marvell expects a substantial NPO ramp while CPO also expands.
This creates a broader investment theme: AI networking may expand faster than any single optical architecture.
The Supply Chain Could Extend Beyond LITE and COHR
Citi's research also identifies upstream companies that could benefit from tight InP supply.
AXT (NASDAQ: AXTI) is an important InP substrate supplier, although its manufacturing exposure to China creates additional export-control and geopolitical risks.
Sumitomo Electric Industries has significant InP wafer and laser capacity and is another important supplier in the chain.
JX Advanced Metals is investing heavily to expand InP substrate capacity, although much of the new capacity is expected to become available later rather than immediately.
These companies offer different risk profiles from Lumentum and Coherent. Upstream suppliers can benefit from constrained materials, but their earnings are also more exposed to capacity additions, customer concentration, export restrictions and technology shifts.
What Could Go Wrong With the CPO Investment Thesis?
The CPO opportunity is significant, but several risks could change the economics.
First, adoption could take longer than expected. CPO is technically complex, and customers may continue using pluggable or NPO solutions for longer.
Second, alternative technologies could take market share. Silicon photonics, LPO/LRO and other optical architectures could reduce demand for particular laser or packaging technologies.
Third, manufacturing capacity could catch up with demand. Today's shortage-driven pricing environment may not persist once new fabs and wafer capacity come online.
Fourth, both Lumentum and Coherent have significant exposure to a relatively small number of large technology customers. Changes in customer architecture or supplier allocation could therefore have an outsized effect on revenue.
Finally, valuation is important. A strong AI infrastructure market does not automatically mean every optical stock will generate attractive shareholder returns from current prices.
What Investors Should Watch Next
The most useful signals will come from actual orders and production rather than headlines about CPO.
Investors should watch:
Lumentum's CPO and NPO order growth and laser shipments.
Coherent's 6-inch InP manufacturing ramp and customer qualification.
NVIDIA's optical networking roadmap and deployment volumes.
Demand for 1.6T and eventually higher-speed optical connectivity.
InP wafer availability and pricing.
CPO versus NPO adoption among hyperscalers and AI-chip suppliers.
Manufacturing yields and packaging capacity.
Whether supply shortages persist into 2027 and 2028.
Citi's research currently sees the supply-demand imbalance as favorable for Lumentum and Coherent, with its report assigning Buy ratings to both.
That is Citi's view, not a guarantee of future stock performance.
CPO and the Next AI Infrastructure Cycle
The AI infrastructure investment story is gradually expanding beyond GPUs. As computing clusters grow, moving data between chips becomes almost as important as processing the data itself.
That makes optical connectivity an increasingly strategic part of the AI supply chain.
Lumentum's laser technology and Coherent's vertically integrated optical manufacturing give both companies exposure to this transition, while upstream InP suppliers provide another potential bottleneck in the ecosystem.
The key question for investors is now whether CPO and NPO adoption moves from qualification and early production into sustained high-volume deployment. Actual order growth, capacity utilization and customer adoption will be more important than the narrative alone.
Follow the blog for more AI infrastructure, semiconductor and technology-stock updates.
This article is for informational and educational purposes only and should not be considered investment advice.