AI Optical Modules Are Sold Out to 2029: The Real Bottleneck in the AI Build-Out

Everyone talks about GPUs and HBM. But the cables that connect AI clusters are running out too. Lumentum is booked into 2029, and a potential FCC rule could redraw the 3.2T generation.

Published by Chip Seek | Optical communications, AI infrastructure and semiconductor supply-chain research

Executive summary

The optical communications supply chain is being pulled in two directions at once. On one side, AI data centers are driving demand for high-speed optical modules far beyond expectations. On the other, potential U.S. regulatory rules are creating long-term uncertainty about where those modules can be sold and which components they can contain.

Lumentum CEO Michael Hurlston said the company's optical capacity is fully booked by customers, with orders extending into early 2029. Some high-end products still face roughly 70% unmet demand in 2027, and others about 30% in 2028. Capacity expansion, even at 12 times the original level, is still outrunning supply.

Morgan Stanley has argued that a potential FCC rule restricting Chinese optical transceivers is more likely to begin with the 3.2T generation, and that a module could qualify for exemption if at least 65% of its bill-of-materials value comes from U.S. companies. The report is a scenario, not a confirmed regulation.

Our view: the near-term bottleneck is capacity — optical chips, lasers, DSPs, packaging and substrates. The long-term variable is policy. Buyers should plan for tight optics through 2029 and watch the 3.2T transition closely.

1. The breaking headline: optical capacity is sold out to 2029

The story broke with a simple but striking statement: Lumentum's optical device capacity has been fully reserved by customers, with order coverage extending into early 2029. For a component that most people never see, that is an extraordinary level of demand visibility.

The detail matters as much as the headline. Hurlston reportedly said that some high-end products still have about 70% of demand unmet in 2027, and other products about 30% unmet in 2028. In April, his assessment was that full-year 2028 capacity would sell out within two quarters. The visibility has since stretched further.

This is not a consumer cycle or a short-term inventory swing. It is a structural shortage driven by AI infrastructure, and it is manifesting in long-term contracts, prepayments and strategic investments by the largest customers.

For anyone building AI clusters, the implication is uncomfortable: the bottleneck is no longer only the accelerator or the HBM stack. It is also the optical link that connects servers, switches and storage across the data center.

2. Why optics matter so much in AI infrastructure

AI clusters are not a single computer. They are thousands of accelerators, switches, storage nodes and management servers connected by a dense fabric of high-speed links. The faster the accelerators, the more data must move between them, and the more the network becomes a limiting factor.

Optical modules convert electrical signals into light and back again, allowing data to travel over fiber at high speed and low loss. In modern AI data centers, optics are used for server-to-switch links, switch-to-switch links and increasingly for longer intra-cluster connections.

As cluster sizes grow, the number of optical links grows faster than the number of servers. A single rack can contain dozens or hundreds of optical connections, and a large cluster can contain hundreds of thousands. That is why optics demand scales so aggressively with AI build-outs.

The technology is also evolving quickly. 400G gave way to 800G, and 1.6T is now ramping. Each generation increases bandwidth per module and changes the component mix, the packaging requirements and the qualification cycles.

3. Lumentum: the company at the center of the shortage

Lumentum is one of the key upstream suppliers in the optical communications chain. It produces indium phosphide (InP) high-speed optical devices, optical components, cloud optical modules and optical switching systems. Its components sit inside the high-speed links that connect AI data centers.

The company's position is strategic because it supplies both the optical chips and the modules that use them. That vertical integration gives it visibility into demand and influence over allocation, but it also exposes it to the same capacity constraints that affect its customers.

To meet demand, Lumentum has expanded capacity at its core Tokyo-area factory to roughly 12 times its previous level over the past two years. It plans to invest at least $350 million more in that factory and adjacent facilities, and its Caswell, UK facility has also completed an expansion.

Even so, Hurlston emphasized that new capacity typically takes three to five years to come online. Expansion is lagging order growth, which is why the order book extends into 2029.

The table explains why the shortage is so persistent. Even aggressive expansion is not enough to close the gap, because demand is growing faster than new capacity can be qualified and brought online.

4. NVIDIA's $2 billion bet on optical supply

Earlier this year, NVIDIA invested $2 billion each in Lumentum and Coherent to secure high-speed optical device supply. The move is a clear signal that optics are now a strategic input for AI infrastructure, not a commodity accessory.

The logic is straightforward. If you are building the largest AI clusters in the world, you cannot afford to have them delayed because optical modules are unavailable. Investing in suppliers is a way to secure capacity and influence roadmaps.

It also reflects a broader trend: the largest customers are moving upstream. They are no longer content to buy from distributors or module makers alone. They want direct relationships with the companies that make the lasers, optical chips and DSPs.

When NVIDIA invests in optics, it is telling the market that the bottleneck has moved from compute to connectivity.

5. The bottleneck map: where optics actually get stuck

Optical modules are complex assemblies, and the supply chain has several potential pinch points. Understanding them helps explain why capacity cannot be expanded simply by adding assembly lines.

Optical chips and lasers. Indium phosphide lasers, EMLs (electro-absorption modulated lasers) and CW light sources are difficult to manufacture and require specialized fabs. Capacity is concentrated among a few suppliers.

DSPs. Digital signal processors handle equalization, error correction and signal integrity. They are typically designed by companies such as Marvell and Broadcom and manufactured at advanced nodes.

Packaging and assembly. Optical modules require precise alignment, hermetic sealing in some cases, and rigorous testing. Automation helps, but yield and throughput remain challenging.

Substrates and materials. High-speed substrates, optical components and specialty materials are also constrained, and each has its own lead time.

The practical consequence is that optics shortages can persist even when module assembly capacity is available, because the upstream components are the real constraint.

6. The FCC question: what the Morgan Stanley report actually says

The second force shaping the optical supply chain is policy. Morgan Stanley published a report on October 1 arguing that a potential FCC rule on optical transceivers is more likely to arrive with the 3.2T generation than with current products.

The mechanism under discussion is the FCC's "Covered List," which can restrict equipment from companies deemed a national-security risk. The report suggests that Chinese-made optical transceivers could face restrictions through this mechanism, but that implementation would likely be phased and tied to the 3.2T generation.

Critically, the report is a scenario analysis. It uses language such as "potential," "possible" and "may." No formal FCC rule has been published, and the timeline is uncertain. That distinction matters for anyone making procurement decisions today.

Read the report as a map of possible policy paths, not as a confirmed regulation.

7. Why the 3.2T generation is the likely policy line

Policy tends to follow technology transitions. The 3.2T generation is the next major step in optical module bandwidth, and it is expected to require new designs, new DSPs and new optical components. That transition creates a natural point for regulators to define rules.

From a practical standpoint, applying restrictions to current 800G and 1.6T products would be disruptive. These are the workhorses of AI data centers today, and replacing them abruptly would raise costs, extend lead times and slow cluster build-outs. A phased approach starting with 3.2T gives the industry time to adapt.

From a policy standpoint, the 3.2T generation is also where the value chain is most contestable. New designs are being defined now, so rules set at this stage can influence which suppliers and which countries capture the next wave of value.

8. The 65% threshold: a value-based exemption

The Morgan Stanley report outlines a scenario in which an optical module could qualify for exemption if at least 65% of its bill-of-materials value comes from U.S. companies. That is a value-based test rather than a simple country-of-origin rule.

The logic is that the most valuable components in a high-speed optical module are the DSP, the laser and the optical chips. If those come from U.S. suppliers, the module can still be assembled elsewhere while the majority of value remains in the U.S. ecosystem.

The report estimates that in a 3.2T module, DSPs, TIA/driver chips, lasers and CW light sources together account for roughly 60% to 70% of bill-of-materials value. If those components are sourced mainly from U.S. suppliers, the module could approach the 65% threshold.

9. Policy scenarios: from rumor to rule

Policy uncertainty is difficult to plan around, but scenario analysis can help. Three broad paths are worth considering.

Scenario A: No rule. The FCC does not implement restrictions on optical transceivers, or the rules are narrow and do not affect mainstream AI modules. Current supply chains continue, and the main constraint remains capacity.

Scenario B: Phased rule at 3.2T. Restrictions begin with the 3.2T generation and are phased in over several years. 800G and 1.6T remain unaffected in the near term. Module makers adapt by increasing U.S. content or qualifying alternate components.

Scenario C: Broader restrictions. Rules are broader or faster than expected, affecting a wider range of products and suppliers. The industry faces higher costs, longer qualification cycles and potentially slower AI build-outs.

The base case for planning is Scenario B: no near-term impact on 800G and 1.6T, with 3.2T as the point of policy risk.

10. China's dominance in optical modules

While the upstream supply of optical chips and DSPs is concentrated in the U.S. and a few other countries, the module assembly and integration market is dominated by Chinese companies.

According to LightCounting data published in May 2026, seven of the world's top ten optical module suppliers by 2025 sales were Chinese companies, with a combined market share of more than 60%. Innolight and Eoptolink ranked first and second globally, followed by other Chinese suppliers.

The dominance is even stronger in the highest-speed segments. Multiple research estimates suggest that Chinese suppliers account for more than 65% of shipments in high-end 800G and above modules. Innolight's 800G market share is estimated at over 40%, and its 1.6T share at roughly 50% to 70%.

The concentration is a source of strength for Chinese suppliers and a source of concern for policymakers in the U.S. It is also why any regulatory change would have global consequences.

11. Record financials: Innolight and Eoptolink

The financial results of the leading Chinese module makers illustrate the scale of the AI optics boom.

Innolight reported record second-quarter 2026 revenue of about $3.3 billion, up 192% year over year and 16% quarter over quarter. Net profit was about $1.16 billion, up 249% year over year and 40% quarter over quarter. The company attributed margin improvement partly to higher 1.6T shipments.

Eoptolink reported record second-quarter revenue of about $1.8 billion, up 109% year over year and 53% quarter over quarter. Net profit was about $698 million, up 113% year over year and 74% quarter over quarter.

These results are not just about AI demand. They reflect the ability of Chinese suppliers to scale manufacturing, manage supply chains and deliver high-speed modules on aggressive schedules.

12. The manufacturing moat: speed, ecosystem and global capacity

Chinese module makers have built a competitive advantage that goes beyond cost. Their strength rests on three pillars: responsiveness, ecosystem depth and international capacity.

Responsiveness. North American AI cluster builds move in weeks, not quarters. Traditional 12-week delivery cycles have been compressed to 8 weeks or less. Chinese suppliers have the manufacturing flexibility and supply-chain coordination to capture each upgrade window.

Ecosystem depth. The Yangtze River Delta, Pearl River Delta and Wuhan Optics Valley form highly concentrated industrial clusters. Suppliers, component makers, testing houses and logistics providers are close together, enabling cost control and fast ramp.

International capacity. Leading Chinese companies have built factories in multiple countries to address geopolitical risk and tariff pressure. That footprint is becoming a competitive advantage as customers seek supply-chain resilience.

13. Technology paths: silicon photonics, LPO and CPO

The optical module market is not static. Several technology paths are competing to deliver higher bandwidth, lower power and lower cost, and each one changes the component mix and the supply chain.

Silicon photonics. Silicon photonics integrates optical functions onto silicon-based chips, enabling higher integration and potentially lower cost at scale. It is particularly attractive for high-speed modules where traditional discrete optics become complex.

LPO (linear pluggable optics). LPO removes the DSP from the module and relies on the host switch to handle signal processing. This can reduce power and cost, but it places more demands on the switch and on the optical link design.

CPO (co-packaged optics). CPO moves optics closer to the switch ASIC, potentially reducing power and improving bandwidth density. It is a longer-term architecture that could reshape the module market if adopted broadly.

Chinese suppliers are no longer simply following these paths. Innolight's 1.6T silicon photonics modules reportedly achieve yields above 95% with about 40% lower power than traditional designs, while Eoptolink leads in 800G LPO with an estimated 75% share.

14. The upstream weakness: EML, InP and DSP

Despite their strength in module assembly, Chinese suppliers remain dependent on imported components for the most critical parts of the bill of materials.

EML lasers. Electro-absorption modulated lasers are essential for high-speed, long-reach optical links. They are difficult to manufacture and are supplied by a small number of companies.

InP substrates and optical chips. Indium phosphide is the material platform for many high-speed lasers and detectors. Substrate quality and process maturity are critical, and supply is concentrated.

DSPs. High-speed digital signal processors are designed by companies such as Marvell and Broadcom and manufactured at advanced nodes. They are a major share of module value.

Chinese optical chip companies are investing heavily and making progress, but high-end products still require customer qualification and time. The gap is not only technical; it is also about yield, reliability and ecosystem maturity.

15. China's optical chips: progress and limits

China's optical chip industry has been growing quickly. Companies are investing in InP lasers, silicon photonics, EMLs and high-speed detectors, supported by demand in China and policy incentives.

Progress is real but uneven. Some products are competitive at mainstream speeds, while the highest-performance devices still lag. Customer qualification cycles in AI data centers are long and demanding, which slows adoption even when a product is technically ready.

The result is a dual-track market: Chinese suppliers gain share in modules and in some components, while the most critical upstream parts remain concentrated among a few international suppliers.

The module business is Chinese-led; the component business is still largely Western-led. That asymmetry is the core of the policy debate.

16. Supply and demand: the 2027–2029 balance

Putting the pieces together, the optical module market is likely to remain tight through 2029. Demand from AI data centers is strong, capacity additions take years and the upstream component supply is constrained.

The balance will vary by product. 800G is mainstream and relatively mature, but demand remains high. 1.6T is ramping and is likely to be the tightest segment in the near term. 3.2T is still developing and will become the focus of policy and technology competition.

New capacity will arrive gradually. Lumentum's expansion, Coherent's investments and Chinese module capacity additions will all contribute, but the lead times are long and the demand curve is steep.

The table is a planning framework, not a forecast. Actual outcomes will depend on AI capex, technology transitions and policy decisions.

17. Pricing, margins and the value shift

In a tight market, pricing power shifts toward suppliers with scarce capacity. Optical module makers with strong positions in 800G and 1.6T have seen margins expand, and component suppliers with scarce lasers and DSPs have gained leverage.

The value shift is important. Module assembly is competitive and increasingly concentrated in China, while the highest-value components — DSPs and lasers — are concentrated among a few suppliers. That asymmetry explains why U.S. policy focuses on value content rather than assembly location.

For buyers, the implication is that cost pressure may come from components as much as from modules. A module price increase can reflect DSP or laser scarcity, not just assembly costs.

Understanding the value chain helps buyers negotiate more effectively and anticipate where shortages will appear.

18. Impact on AI cluster build-outs

Optical modules are a small share of total AI infrastructure spending, but they can be a large share of the schedule risk. A cluster cannot be completed without the optics that connect its servers and switches.

That is why hyperscalers are signing long-term agreements and making strategic investments. They are not just buying components; they are securing the ability to complete their build-outs on schedule.

The risk is asymmetric. A shortage of optics can delay a multi-billion-dollar cluster, while the cost of securing optical supply is relatively small. That imbalance encourages customers to pay premiums and lock capacity early.

In AI infrastructure, optics are a small line item with outsized schedule leverage.

19. What it means for distributors and component buyers

For distributors, the optical shortage creates both opportunity and risk. Demand is strong, but supply is constrained and allocation is common. Success depends on supplier relationships, inventory visibility and the ability to help customers plan.

For buyers, the lessons are familiar from other allocation markets: secure critical volumes early, qualify alternates and maintain visibility into lead times. Optical modules are not a commodity that can be sourced at the last minute.

Traceability is also important. Counterfeit and out-of-spec optical modules can appear in tight markets, and the consequences of a failure in a data center can be severe.

At Chip Seek, we help customers navigate component availability across categories. Our component catalog, manufacturer directory and BOM RFQ tools are designed to support that planning.

20. The three variables to watch

The outlook for optical modules depends on three variables. Watching them will provide early signals about whether the market tightens further or begins to ease.

Capacity expansion progress. How quickly can Lumentum, Coherent and other suppliers bring new capacity online? Are yields improving? Are new fabs and packaging lines ramping as planned?

Cloud capex. Are hyperscalers continuing to increase AI infrastructure spending? Are new clusters being announced and built? A slowdown in capex would ease demand pressure.

Policy. Does the FCC move from scenario to formal rule? What is the timing, scope and threshold? How do module makers and component suppliers respond?

These three variables are interconnected. A policy change could affect capex; a capex slowdown could ease capacity; a capacity expansion could reduce the impact of policy.

21. A practical playbook for buyers

Optical modules are now a strategic component, and buyers need a plan that reflects that reality. The following steps are a starting point.

1. Map optical exposure. Identify every product and platform that depends on 800G, 1.6T or 3.2T optics, and estimate the number of links per system.
2. Quantify schedule risk. Estimate the impact of a four-week, eight-week or twelve-week delay in optical supply on your build-out plans.
3. Secure critical volumes. Negotiate long-term agreements or buffer inventory for the platforms where optics are on the critical path.
4. Qualify alternates. Identify second sources for modules and for key components, and begin qualification before a shortage forces the issue.
5. Monitor policy. Track FCC activity and the 3.2T transition, and assess exposure to potential U.S. content rules.
6. Verify traceability. Confirm supplier credentials, date codes and test reports, especially in a tight market.

The best time to secure optics is before the shortage becomes obvious. In an allocation market, planning is a competitive advantage.

22. Risks to the outlook

The base case is tight optics through 2029, but several risks could change the picture.

Demand risk. If AI capex slows, optical demand could weaken faster than expected. A pause in hyperscale spending would ripple through modules and components.

Supply risk. Faster-than-expected capacity additions or yield improvements could ease the shortage. Conversely, delays in new fabs or packaging lines could make it worse.

Policy risk. A broader or faster FCC rule could disrupt supply chains and raise costs. A rule that never materializes would remove a source of uncertainty.

Technology risk. A shift toward CPO or another architecture could change the module market and the component mix. A slower-than-expected 3.2T transition could delay policy decisions.

23. Three scenarios for 2027–2029

Scenario planning helps teams prepare for uncertainty. The three scenarios below are structured ways to think about the range of outcomes.

The base case is the most useful planning assumption, with explicit contingency plans for the bull and bear scenarios.

24. Frequently asked questions

Is Lumentum really sold out to 2029?

Reports indicate that Lumentum's optical capacity is fully booked by customers, with order coverage into early 2029. Specific product availability varies, and some segments face larger gaps than others.

Does the FCC rule already exist?

No. The Morgan Stanley report describes a potential rule, not a confirmed regulation. No formal FCC rule on optical transceivers has been published.

Will 800G and 1.6T be affected?

In the base scenario, near-term impact on 800G and 1.6T is limited. Policy risk is more likely to begin with the 3.2T generation.

Why is China so strong in optical modules?

Chinese suppliers have built advantages in manufacturing speed, ecosystem depth and cost, and they have scaled aggressively to meet AI demand.

What is the biggest weakness in the Chinese supply chain?

High-end optical chips, EML lasers, InP substrates and DSPs remain dependent on international suppliers.

What should buyers do now?

Map exposure, secure critical volumes, qualify alternates and track policy developments.

25. Glossary: optical module terms

• Transceiver: a module that converts electrical signals to optical signals and back.
• 800G / 1.6T / 3.2T: aggregate bandwidth per module, in gigabits per second.
• DSP: digital signal processor used for equalization and error correction.
• EML: electro-absorption modulated laser, used in high-speed optical links.
• InP: indium phosphide, a material platform for high-speed lasers and detectors.
• LPO: linear pluggable optics, a design that removes the DSP from the module.
• CPO: co-packaged optics, integrating optics closer to the switch ASIC.
• Covered List: an FCC mechanism for restricting equipment deemed a national-security risk.

Understanding these terms helps buyers interpret supplier updates and policy news.

26. The optical module value chain and BOM breakdown

To understand where value and risk sit, it helps to break an optical module into its main components. The exact mix varies by speed and architecture, but the broad structure is consistent.

Optical components. Lasers, detectors, modulators and optical subassemblies convert electrical signals to light and back. These are among the most technically demanding parts of the module.

Electrical components. DSPs, transimpedance amplifiers, driver chips and clocking devices handle signal integrity, equalization and error correction.

Packaging and assembly. Precision alignment, hermetic sealing, thermal management and testing are labor- and equipment-intensive.

Firmware and test. Module firmware, calibration and test procedures are essential for performance and reliability.

The value distribution explains the policy debate. If DSPs and lasers are the largest value items and are supplied mainly by U.S. companies, then a value-based rule can favor U.S. content even when modules are assembled elsewhere.

27. How hyperscalers qualify and procure optics

Hyperscalers do not buy optical modules the way consumers buy accessories. They qualify designs, test them in real workloads and deploy them at massive scale. That process shapes the supply chain in important ways.

Qualification. A new module must pass electrical, optical, thermal and reliability tests. It must interoperate with switches and network operating systems. Qualification can take months and often involves multiple iterations.

Volume commitments. Once qualified, modules are deployed in large volumes. Hyperscalers want consistent quality and predictable supply, which is why they favor long-term relationships and agreements.

Dual sourcing. Hyperscalers typically qualify more than one supplier to reduce risk. That creates opportunities for module makers but also makes qualification a competitive barrier.

Roadmap alignment. Hyperscalers work with suppliers on future speeds and architectures, which gives them influence over the supply chain and early access to new products.

The qualification process is a major reason why capacity cannot be switched quickly. A new supplier cannot simply appear; it must earn qualification.

28. Lessons from the HBM shortage

The optical module shortage has parallels with the HBM shortage that has dominated memory headlines. Both are driven by AI demand, both are constrained by advanced packaging and qualification, and both have led to long-term contracts and strategic investments.

There are also differences. HBM is a memory product with a relatively concentrated supplier base, while optical modules involve a broader ecosystem of components and assembly. HBM is sold primarily to accelerator vendors, while optics are sold to hyperscalers and network equipment makers.

But the core lesson is the same: in an AI-driven cycle, the bottleneck moves to whichever component is hardest to scale. Today it is HBM and optics; tomorrow it could be power delivery, substrates or thermal materials.

The comparison suggests that AI supply chains will continue to face sequential bottlenecks. Planning for one component is not enough; the full dependency chain must be mapped.

29. Key takeaways for executives

If you need a short summary for leadership, the following points capture the essentials.

Capacity is the near-term constraint. Lumentum is sold out into 2029, and 800G and 1.6T modules are tight. Expansion takes years, and upstream components are hard to scale.

China dominates module assembly. Seven of the top ten module suppliers are Chinese, with more than 60% combined share and even higher share in high-speed segments.

The upstream is still Western-led. DSPs, EML lasers and InP substrates remain concentrated among a few suppliers, which is the basis for value-based policy proposals.

Policy is a long-term variable. A potential FCC rule is more likely to begin with 3.2T and may include a 65% U.S. value threshold. No formal rule exists yet.

Planning beats prediction. Map exposure, secure critical volumes, qualify alternates and track policy. In an allocation market, preparation is a competitive advantage.

Bottom line: plan for tight optics through 2029 and treat 3.2T as the policy inflection point.

Conclusion: capacity now, policy later

The optical module supply chain is being shaped by two forces with different timelines. Capacity is the immediate constraint: Lumentum is booked into 2029, 800G and 1.6T are tight, and upstream components are hard to scale. Policy is the long-term variable: a potential FCC rule could reshape the 3.2T generation, but it has not yet become formal regulation.

For buyers, the practical response is to plan for tight optics through 2029, secure critical volumes and qualify alternates. For suppliers, the opportunity is to expand capacity and move up the value chain. For policymakers, the challenge is to balance security concerns with the cost and complexity of reworking a global supply chain.

The AI build-out will not be limited only by GPUs and HBM. It will also be limited by the light that connects them.

Editorial note and official references

This article is an independent industry analysis prepared by the Chip Seek editorial team, based on public disclosures, company statements, research reports and industry sources. Policy details may change as official rules are published.

• Lumentum — company and investor information
• FCC — official rules and proceedings

Chip Seek supplies and sources electronic components for global customers. For availability, traceability and sourcing support, contact the Chip Seek team.