Nikkei flags optical supply strain
What happened
- Nikkei Asia reported on May 27 that AI data-center demand is tightening supplies of lasers, optical fiber, connectors and substrates used in high-speed networks. - LightCounting said demand for optical connectivity is running about 30% above supply, with indium-phosphide EML and laser-chip capacity among the limits. - LightCounting expects shortages to ease by the end of 2026; Coherent and Lumentum are expanding capacity for AI-networking demand.
Why it matters
Nikkei Asia reported on May 27 that AI data-center demand is straining supplies of lasers, optical fiber, connectors and substrates used in high-speed networking gear, extending bottlenecks beyond the accelerators that have dominated the AI buildout story. The report said shortages are lifting prices and giving a boost to parts of China’s supply chain as cloud and AI operators race to add capacity. Industry researchers and suppliers have been describing the same pressure points in optical links and photonics used to connect GPU clusters. The result is that networking hardware, not only chips, is emerging as a constraint on how fast new AI capacity can be deployed. ### Which parts of the AI stack are getting squeezed? Lasers, optical fiber, connectors and substrates sit inside the high-speed links that move data between servers, switches and accelerator clusters. Those links are carried by optical transceivers and related photonic components that are essential for scale-out and scale-up AI networks. LightCounting said in an April 2026 market forecast that demand for optical connectivity was exceeding supply by 30%. The research firm said growth was being limited by production capacity for indium-phosphide electro-absorption modulated lasers, or EMLs, and other laser chips, even as cloud spending on AI clusters remained strong. ### Why are optics suddenly as important as chips? AI clusters use far more high-bandwidth interconnects than traditional server fleets because training and inference workloads require large numbers of accelerators to exchange data at high speed. LightCounting said its January 2026 report modeled optics demand not only for compute nodes but also for the back-end networks inside AI clusters at Amazon, Google, Meta, Microsoft and Oracle. (lightcounting.com) Nikkei Asia said the bottleneck is widening into industrial components and power equipment supply chains as data-center construction accelerates. That matches a broader pattern in vendor disclosures showing AI-networking demand spilling into lasers, modules and switching architectures rather than remaining confined to GPUs. ### What are suppliers saying about demand and capacity? Coherent said in its fiscal third-quarter 2026 results that it was ramping capital investment to increase capacity because of “strong visibility” into ongoing demand. (lightcounting.com) The company has also highlighted next-generation VCSEL array products for AI scale-up networks and near-packaged or co-packaged optics, which are intended to handle dense, short-reach interconnects in AI systems. (asia.nikkei.com) Lumentum said in its fiscal third-quarter 2026 results that AI and cloud data-center demand was supporting growth and that co-packaged optics and optical circuit switches were expected to become larger drivers over time. The company reported quarterly revenue of $808.4 million and pointed investors to opportunities tied to AI-networking architectures. ### Does this mean prices are rising across the network layer? (coherent.com) Nikkei Asia said shortages were pushing up prices for some optical and networking inputs as buyers competed for limited supply. LightCounting’s forecast also said the market remained supply-constrained, though it added that shortages should ease by the end of 2026 if capacity expansions continue and GPU supply does not suffer fresh disruptions. (investor.lumentum.com) A separate implication is that any delay in accelerator shipments can ripple through the rest of the AI hardware chain. LightCounting said even a minor glitch in GPU supply would reduce demand for transceivers and other products used in AI clusters, underscoring how tightly linked the component stack has become. ### Where does this leave the next phase of data-center expansion? (asia.nikkei.com) May 27 is the date Nikkei Asia’s report put fresh attention on optics as a live supply constraint in AI infrastructure. LightCounting said the current shortages should fade by the end of 2026, while Coherent and Lumentum have both told investors they are adding capacity and positioning products for AI-networking deployments. (asia.nikkei.com) (lightcounting.com)
Key numbers
- Nikkei Asia reported on May 27 that AI data-center demand is tightening supplies of lasers, optical fiber, connectors and substrates used in high-speed networks.
- LightCounting said demand for optical connectivity is running about 30% above supply, with indium-phosphide EML and laser-chip capacity among the limits.
- LightCounting expects shortages to ease by the end of 2026; Coherent and Lumentum are expanding capacity for AI-networking demand.
- Nikkei Asia reported on May 27 that AI data-center demand is straining supplies of lasers, optical fiber, connectors and substrates used in high-speed networking gear, extending bottlenecks beyond the accelerators that have dominated the AI buildout story.
What happens next
- Nikkei Asia reported on May 27 that AI data-center demand is straining supplies of lasers, optical fiber, connectors and substrates used in high-speed networking gear, extending bottlenecks beyond the accelerators that have dominated the AI buildout story.
- (lightcounting.com) The company has also highlighted next-generation VCSEL array products for AI scale-up networks and near-packaged or co-packaged optics, which are intended to handle dense, short-reach interconnects in AI systems.
- (asia.nikkei.com) Lumentum said in its fiscal third-quarter 2026 results that AI and cloud data-center demand was supporting growth and that co-packaged optics and optical circuit switches were expected to become larger drivers over time.
Quick answers
What happened in Nikkei flags optical supply strain?
Nikkei Asia reported on May 27 that AI data-center demand is tightening supplies of lasers, optical fiber, connectors and substrates used in high-speed networks. LightCounting said demand for optical connectivity is running about 30% above supply, with indium-phosphide EML and laser-chip capacity among the limits. LightCounting expects shortages to ease by the end of 2026; Coherent and Lumentum are expanding capacity for AI-networking demand.
Why does Nikkei flags optical supply strain matter?
Nikkei Asia reported on May 27 that AI data-center demand is straining supplies of lasers, optical fiber, connectors and substrates used in high-speed networking gear, extending bottlenecks beyond the accelerators that have dominated the AI buildout story. The report said shortages are lifting prices and giving a boost to parts of China’s supply chain as cloud and AI operators race to add capacity. Industry researchers and suppliers have been describing the same pressure points in optical links and photonics used to connect GPU clusters. The result is that networking hardware, not only chips, is emerging as a constraint on how fast new AI capacity can be deployed. Which parts of the AI stack are getting squeezed? Lasers, optical fiber, connectors and substrates sit inside the high-speed links that move data between servers, switches and accelerator clusters. Those links are carried by optical transceivers and related photonic components that are essential for scale-out and scale-up AI networks. LightCounting said in an April 2026 market forecast that demand for optical connectivity was exceeding supply by 30%. The research firm said growth was being limited by production capacity for indium-phosphide electro-absorption modulated lasers, or EMLs, and other laser chips, even as cloud spending on AI clusters remained strong. Why are optics suddenly as important as chips? AI clusters use far more high-bandwidth interconnects than traditional server fleets because training and inference workloads require large numbers of accelerators to exchange data at high speed. LightCounting said its January 2026 report modeled optics demand not only for compute nodes but also for the back-end networks inside AI clusters at Amazon, Google, Meta, Microsoft and Oracle. (lightcounting.com) Nikkei Asia said the bottleneck is widening into industrial components and power equipment supply chains as data-center construction accelerates. That matches a broader pattern in vendor disclosures showing AI-networking demand spilling into lasers, modules and switching architectures rather than remaining confined to GPUs. What are suppliers saying about demand and capacity? Coherent said in its fiscal third-quarter 2026 results that it was ramping capital investment to increase capacity because of “strong visibility” into ongoing demand. (lightcounting.com) The company has also highlighted next-generation VCSEL array products for AI scale-up networks and near-packaged or co-packaged optics, which are intended to handle dense, short-reach interconnects in AI systems. (asia.nikkei.com) Lumentum said in its fiscal third-quarter 2026 results that AI and cloud data-center demand was supporting growth and that co-packaged optics and optical circuit switches were expected to become larger drivers over time. The company reported quarterly revenue of $808.4 million and pointed investors to opportunities tied to AI-networking architectures. Does this mean prices are rising across the network layer? (coherent.com) Nikkei Asia said shortages were pushing up prices for some optical and networking inputs as buyers competed for limited supply. LightCounting’s forecast also said the market remained supply-constrained, though it added that shortages should ease by the end of 2026 if capacity expansions continue and GPU supply does not suffer fresh disruptions. (investor.lumentum.com) A separate implication is that any delay in accelerator shipments can ripple through the rest of the AI hardware chain. LightCounting said even a minor glitch in GPU supply would reduce demand for transceivers and other products used in AI clusters, underscoring how tightly linked the component stack has become. Where does this leave the next phase of data-center expansion? (asia.nikkei.com) May 27 is the date Nikkei Asia’s report put fresh attention on optics as a live supply constraint in AI infrastructure. LightCounting said the current shortages should fade by the end of 2026, while Coherent and Lumentum have both told investors they are adding capacity and positioning products for AI-networking deployments. (asia.nikkei.com) (lightcounting.com)
