Lumentum Warns of an Indium Phosphide Shortage as Nvidia Locks In Optical Capacity

Lumentum has warned that indium phosphide supply cannot match AI demand, despite years of capacity expansion across the optical component industry. The conflict is becoming harder to ignore. Nvidia has moved to secure future production, while customers outside its orbit face tighter access to essential laser components.

The warning entered the Chinese market spotlight on August 6, 2026. A market report connected the shortage narrative with a three-day rally in an indium phosphide concept stock. However, the underlying supply issue predates that trading response.

Lumentum CEO Michael Hurlston publicly described indium phosphide as a looming AI infrastructure bottleneck at the RAISE Summit in late July. Coherent and independent supply-chain analysts had raised similar concerns months earlier.

That timing matters. This is not a shortage revealed by one daily stock move. It is the result of AI clusters adopting faster optical connections while specialized material, wafer, and laser capacity expands much more slowly.

Nvidia’s response provides the clearest signal. The company has committed capital and purchasing support to Lumentum and Coherent, two major producers of the optical devices needed inside high-speed data-center networks.

The primary contest is therefore not Lumentum against another component maker. It is hyperscaler demand against the physical production limits of the optical supply chain.

Lumentum’s Warning Turns a Niche Material Into an AI Constraint

The latest warning moves indium phosphide from the edge of semiconductor coverage to the center of the AI infrastructure debate.

Indium phosphide, usually shortened to InP, is a compound semiconductor used to create lasers and other high-speed photonic devices. These devices convert electrical data into optical signals that can travel through fiber.

AI data centers need those optical links because thousands of accelerators must exchange information with low latency. A cluster can contain exceptionally fast processors and still underperform when its network cannot move data quickly enough.

The shift toward 800-gigabit and 1.6-terabit optical modules increases that pressure. A 1.6T module can carry twice the nominal data rate of an 800G module, but it requires more advanced optical components and manufacturing processes.

Electro-absorption modulated lasers, known as EMLs, are central to many of these systems. An EML combines a laser source with a modulator that encodes data onto the light. Indium phosphide offers properties suited to this high-speed operation.

Lumentum has been expanding EML and continuous-wave laser production for several years. Continuous-wave lasers provide the external light source used by some silicon photonics systems.

Yet the company’s capacity additions have not closed the reported demand gap. Customers are reserving production far in advance, and leading buyers are using long-term agreements to protect their deployment schedules.

The strongest evidence comes from what companies are doing, not from one executive’s warning. Lumentum announced a new 240,000-square-foot manufacturing site in Greensboro, North Carolina, on March 26. The company said the facility would produce InP-based continuous-wave and ultra-high-power lasers.

The Greensboro expansion uses a semiconductor facility acquired from Qorvo. Lumentum plans to retrofit the operational site around six-inch InP wafers.

A larger wafer can yield more devices per manufacturing cycle, assuming production achieves acceptable yields. That makes the six-inch transition an important part of the company’s expansion strategy.

However, a building announcement does not create qualified optical supply overnight. Equipment must be installed, processes transferred, customers qualified, and manufacturing yields stabilized.

That delay creates the article’s central tension. AI infrastructure buyers are planning network deployments years ahead, but the underlying laser supply cannot expand at software speed.

The August 6 market reaction reflects that mismatch. Investors are no longer treating indium phosphide as an obscure material category. They are treating production rights, wafer access, and device yields as strategic assets.

Still, a rising concept stock does not independently verify the size or duration of the shortage. Daily trading can amplify a valid industrial theme far beyond the evidence available for any particular company.

What changed is more fundamental than a share-price rally. Multiple participants now openly acknowledge that optical capacity is limiting deployment plans, while their capital commitments show they expect the constraint to persist.

Nvidia’s Capacity Deals Show Who Is Under Pressure

Nvidia is protecting its AI platform by reserving the optical components that connect its processors, leaving other buyers exposed to the same limited supply.

Modern AI systems depend on more than GPUs. They also require switches, fiber, transceivers, lasers, memory, power systems, cooling equipment, and suitable data-center capacity.

That wider dependency changes Nvidia’s role. The company cannot focus only on accelerator production when missing optical components can delay a complete cluster.

In March, Nvidia moved deeper into the photonics supply chain through strategic arrangements with Lumentum and Coherent. Reuters reported that Nvidia committed $2 billion to each company as part of agreements intended to secure advanced optical production.

The structure represents more than a conventional component order. Large commitments give suppliers greater confidence to fund factories, equipment, and process development before every unit of demand becomes firm revenue.

They also give Nvidia stronger visibility into future supply. When capacity remains scarce, that visibility can become a competitive advantage over cloud providers, networking vendors, and smaller AI infrastructure developers.

A June supply-chain investigation from Reuters reported that Lumentum was sold out through 2028, according to SemiAnalysis. The same account described disruptions involving Chinese export controls and delayed substrate permits.

That reported timeline should be treated carefully. “Sold out” can cover committed capacity, forecasts, reservations, or contractual orders, depending on the speaker and product family.

It does not mean every Lumentum production line has an unchangeable order for every future device. It does indicate that customers are competing for output well beyond the current quarter.

Lumentum’s financial performance offers another signal. The company’s fiscal 2026 releases repeatedly identified cloud and AI data-center demand as a major growth driver.

In its fiscal second-quarter results, Lumentum reported strong demand for cloud networking products and discussed continued investment in optical capacity. The company also highlighted orders associated with co-packaged optics.

Co-packaged optics places optical engines close to a networking chip instead of at the edge of a switch in removable modules. Shorter electrical paths can reduce power use and improve signal integrity at very high bandwidths.

Nvidia has promoted co-packaged optics as part of its networking roadmap. That architecture still needs reliable laser sources, so silicon photonics does not automatically remove the need for indium phosphide.

This distinction is crucial. Silicon photonics can integrate many optical functions using silicon manufacturing techniques, but silicon is a poor light emitter. External InP lasers frequently supply the light.

The pressure therefore falls on several groups at once.

Cloud operators need enough optical modules to bring new accelerator clusters online. Networking companies need lasers for 1.6T products and future co-packaged systems. Module manufacturers need predictable access to wafers, laser dies, and packaging capacity.

Smaller buyers face the sharpest problem. They lack Nvidia’s purchasing scale and cannot easily underwrite a supplier’s multiyear expansion.

Coherent also faces the pressure from the manufacturing side. It must expand output while managing substrate availability, process yields, customer qualifications, and the risk of building too much capacity.

Nvidia’s agreements reduce financing uncertainty for selected suppliers. They do not eliminate the physical work required to produce more qualified lasers.

This is why the capacity race matters beyond Nvidia. If the largest customer reserves a substantial portion of new production, every other customer must respond through longer contracts, alternative designs, or additional suppliers.

Why Lumentum Cannot Expand Indium Phosphide Supply Overnight

The bottleneck persists because indium phosphide production combines scarce substrates, specialized fabrication, demanding yields, and lengthy customer qualification.

Semiconductor shortages are often described as simple capacity problems. Add equipment, process more wafers, and ship more components. Indium phosphide manufacturing is less forgiving.

The first limit appears upstream. Suppliers must produce high-quality InP substrates with tightly controlled crystal properties. Defects introduced at this stage can affect device performance and manufacturing yield later.

The wafer then receives epitaxial layers, meaning thin crystalline layers grown to create the device’s active structure. Precise composition and thickness determine how the laser emits and modulates light.

Fabrication converts that layered material into individual laser structures. Packaging must align the tiny optical device with fiber or another photonic component while controlling heat and maintaining reliability.

Each stage needs specialized equipment and experience. A new participant cannot simply redirect a standard silicon line and become a high-volume optical supplier.

Yield creates another constraint. A factory’s theoretical wafer capacity means little if an unacceptable share of devices fails performance or reliability tests.

That problem becomes harder as data rates rise. Components operating at 200 gigabits per second per lane must meet demanding signal, power, temperature, and lifetime requirements.

Customers also require qualification before using a component in a production network. They test performance across operating conditions and evaluate whether the supplier can produce consistent results at scale.

These qualification cycles protect buyers from costly failures. They also slow the introduction of new capacity and make immediate substitution difficult.

Lumentum’s six-inch wafer strategy addresses part of the scale problem. Larger wafers can increase output and improve equipment economics compared with smaller formats.

The strategy also introduces execution risk. Process recipes, tooling, material uniformity, and yields must work across the larger surface.

The company’s North Carolina facility should expand domestic production once it is fully converted. Its existing operation in San Jose remains important, particularly for advanced laser development and manufacturing.

Lumentum previously announced an expansion focused on ultra-high-power lasers for AI-driven co-packaged optics. The laser expansion shows that management identified this demand before the latest shortage headlines.

Even so, capacity additions must move through a sequence. The company needs equipment, qualified materials, trained employees, stable processes, acceptable yields, and customer approval.

Geopolitics further complicates that sequence. China plays an important role in the production and processing of several compound-semiconductor inputs, including indium-related materials.

Reuters reported in June that Chinese export controls were affecting InP substrate availability. Permits allowed some shipments to resume, but suppliers and customers still faced uncertainty.

The restriction does not affect every company identically. Some producers hold inventory, use different substrate suppliers, or manufacture more of their supply internally.

However, an export system based on individual approvals makes planning harder. A supplier can have strong end demand and adequate device-fabrication space while still lacking enough qualified starting material.

This creates a structural mismatch between the two sides of the market.

AI companies revise product schedules quickly and place large orders when a new architecture proves attractive. Compound-semiconductor supply expands through physical projects measured in quarters or years.

The mismatch also explains why prices alone cannot immediately solve the shortage. Higher prices encourage investment, but they do not shorten crystal-growth cycles or erase qualification requirements.

Alternative technologies can reduce pressure over time. Direct-drive optics, different modulator designs, improved silicon photonics, and new packaging approaches may lower the number or power of external lasers required.

None offers an instant universal replacement. Designs differ by reach, bandwidth, power budget, reliability target, and network architecture.

Lumentum’s position rests on this difficult substitution problem. Its products are valuable because customers need proven performance at scale, not merely because InP is scarce.

That distinction will determine whether the industry is undergoing a durable revaluation or a temporary shortage trade.

The Revaluation Case Still Carries Serious Risks

Scarcity strengthens Lumentum’s negotiating position, but investors can overestimate how much of that advantage will survive new capacity and changing optical designs.

The bullish argument starts with a clear chain. AI clusters require more optical bandwidth. Faster links require advanced lasers. Qualified InP capacity remains concentrated among a limited group of suppliers.

That chain supports higher utilization and longer customer commitments. It can also improve the strategic value of substrate makers, epitaxy providers, wafer fabricators, packaging specialists, and testing companies.

Yet a value-chain revaluation does not mean every company associated with indium phosphide will benefit equally. Some businesses have only indirect exposure, limited production scale, or products that do not meet data-center requirements.

The Chinese concept-stock rally illustrates this risk. A company can be categorized as an InP supplier without having meaningful exposure to the EMLs or continuous-wave lasers now attracting Nvidia’s attention.

Investors must separate material association from qualified production. They should also distinguish potential capacity from products already shipping to major customers.

There is another risk in the shortage estimates. Much of the public discussion relies on executive comments, analyst channel checks, and reported order coverage.

These sources can reveal real pressure, but they do not provide a standardized measure of the supply-demand gap. Product categories, forecast periods, and the definition of capacity can differ.

A reported imbalance above 30 percent sounds precise. Without a common denominator, it should remain an estimate rather than a settled industry statistic.

Customer concentration creates a second risk. Nvidia’s commitments help suppliers finance expansion, but they may also increase dependence on one buyer and its product roadmap.

Long-term agreements can limit downside while demand is strong. Their economics depend on pricing terms, volume obligations, milestones, and what happens when designs change.

Neither public enthusiasm nor a headline commitment reveals every contractual detail. Investors should avoid assuming that reserved capacity automatically produces exceptional margins.

Execution presents the third risk. Lumentum must retrofit its Greensboro site while maintaining quality and serving existing customers.

A delayed equipment installation or slower yield improvement would extend scarcity, but it could also prevent Lumentum from converting demand into revenue. A shortage is not automatically beneficial when the supplier cannot ship enough product.

Coherent provides an important comparison. It also produces advanced optical components and has received Nvidia support.

Competition between these companies should increase as both expand. Customers have strong reasons to qualify multiple sources where technically possible, particularly after recent geopolitical disruptions.

Smaller specialists may also enter selected parts of the market. They can target substrates, epitaxial wafers, laser designs, or photonic integration instead of reproducing the entire Lumentum model.

Technology substitution remains the largest long-term uncertainty. Networking architectures do not stand still while factories expand.

If customers adopt designs that use fewer external lasers, improve laser sharing, or integrate different light sources, today’s demand forecast can change. Better component efficiency could reduce unit requirements even as total bandwidth grows.

Co-packaged optics cuts both ways. It increases the need for dependable high-power light sources, supporting Lumentum’s thesis. It can also consolidate optical functions and change where value sits in the system.

The company’s public statements describe a large opportunity, but the technology has not completed a universal transition across data centers. Pluggable transceivers remain widely deployed and continue to improve.

Macroeconomic conditions add another uncertainty. Hyperscalers currently plan extraordinary AI infrastructure spending, yet customers can postpone projects when power, permitting, construction, or model economics fall short.

An optical supplier can hold strong orders and still experience schedule changes. Components must arrive alongside accelerators, switches, power equipment, and finished data halls.

The most defensible conclusion is narrower than the market’s most optimistic version. Indium phosphide is a real constraint in advanced AI networking, and Lumentum owns strategically important manufacturing capability.

It is too early to conclude that every InP-related company deserves a permanent valuation increase. Durable value will depend on qualified output, customer diversity, yields, and relevance to future network designs.

Three Signals Will Test the Lumentum Shortage Thesis

The next three months should reveal whether the shortage is becoming measurable revenue, expanding production, or a narrative running ahead of execution.

The first signal is Lumentum’s fiscal fourth-quarter and full-year 2026 report, scheduled for August 11 after the US market closes. That date provides the nearest formal checkpoint after the July warning and August market reaction.

The company’s earnings schedule confirms the timing. Investors should focus on cloud and networking revenue, order coverage, product mix, and capacity commentary.

Guidance will matter more than a single quarterly result. Strong forward demand accompanied by specific production milestones would reinforce the shortage thesis.

Management should also clarify how much of expected growth depends on existing lines and how much requires the Greensboro conversion. That distinction shows whether expansion is already producing shipments.

Any discussion of customer concentration deserves attention. Nvidia’s commitments reduce uncertainty, but Lumentum still needs a broad market for its expanded output.

The second signal is operational progress at the North Carolina facility. Announced square footage measures ambition, not production.

Useful evidence would include installed equipment, completed process transfers, customer qualifications, and improving six-inch wafer yields. These milestones would show that capacity is moving from a corporate plan into usable supply.

Delays would have mixed implications. They would preserve scarcity but weaken Lumentum’s ability to capture the available demand.

Faster progress would support revenue growth while gradually reducing the imbalance. It would also test whether a six-inch InP process can deliver the economics expected by management.

The third signal is the response from Coherent, substrate suppliers, and optical-module manufacturers. Independent confirmation matters because no single company sees the entire market.

Coherent’s comments on substrate access, lead times, and laser output can strengthen or weaken Lumentum’s account. Module vendors can show whether laser availability is delaying 1.6T shipments or merely complicating procurement.

Export approvals are part of the same signal. More predictable InP substrate shipments from China would ease one upstream constraint, although device fabrication and qualification could remain tight.

Conversely, renewed permit delays would strengthen the case for domestic capacity. They would also increase the strategic value of inventory and alternative substrate sources.

The broader test is whether optical supply remains a limiting factor after other AI constraints change. Memory, power, cooling, and accelerator availability all compete for the title of the next bottleneck.

A shortage can be genuine without becoming the single factor that controls every data-center schedule. Projects usually face several constraints at once.

Developers and enterprise AI buyers should care because networking availability affects deployment timing and system cost. A processor roadmap offers limited value when a complete cluster cannot be connected on schedule.

Infrastructure teams should map optical dependencies alongside compute and power requirements. They should ask suppliers which components are single-sourced, how far orders extend, and whether alternative designs have completed qualification.

They should also preserve the evidence behind these decisions. A searchable knowledge base can keep supplier statements, qualification records, and architecture changes connected as conditions evolve.

For investors, the discipline is similar. Track actual shipments, factory milestones, yields, and customer breadth instead of treating every InP label as equal exposure.

The August 6 rally turned Lumentum’s warning into a visible market story. The underlying event began earlier, as Nvidia and its suppliers committed billions and years of planning to optical capacity.

The next earnings call will show whether that urgency is translating into measurable output. The factory update will show whether supply can respond. Competitor and substrate data will show whether the constraint extends beyond one company’s order book.

Watch those three signals before accepting either extreme. The shortage is more substantial than a one-day headline, but the full-chain revaluation still needs operating evidence.