Having already driven massive shortages in memory chips and CPUs, the global artificial intelligence boom is now disrupting a more obscure part of the supply chain: optical communications.
From fibers that transmit data to transceivers that connect servers together, optical technology has long been a crucial part of communications infrastructure. But the sudden surge in AI infrastructure and appetite for ever faster transmission speeds has resulted in shortages and price increases across the entire optical supply chain, from lasers and substrates to optical fibers and connectors.
Shen, an industry veteran who has seen decades of up-and-down business cycles, never thought his field would flourish like it does today.
“I visited many clients in Guangzhou recently and all of us, including myself, have strong mixed feelings,” Shen, who is a vice president of a transceiver component supplier, told Nikkei Asia. “None of us expected this industry, across the supply chain, would have grown this much in such a short period of time.”
His mixed feelings, he added, are not about the state of business. “On the contrary, the overall demand is so far beyond the entire industry’s capacity that everything is in shortage. But I didn’t have the vision to buy a single stock in any of those companies.”
Pluggable optical transceivers are a prime example. These stick-like modules connect servers together to enable high-speed communication between them, and AI data centers need massive amounts of them. The global rush to build AI infrastructure is driving shortages and price increases.
Optical fiber prices have also surged due to worsening supply constraints.
“People only noticed the memory chip shortage, but in fact, fibers are also in serious constraints,” a manager with a leading optical fiber supplier told Nikkei Asia. “All the capacity allocation has been locked by those huge data center clients, there’s nothing we could provide for additional orders.”
Optical fiber maker Corning is set to expand capacity in the US by 10 times after inking a multiyear supply deal with Nvidia, for example, while Fujikura of Japan said in March it will triple its capacity with a JPY 300 billion (USD 1.9 billion) (USD 1.89 billion) investment to address American clients’ needs.
The hunger for faster data transmission is also accelerating advances in optical tech.
Nvidia, Broadcom, Taiwan Semiconductor Manufacturing Company, and Intel, for example, are pushing a next-generation technology known as co-packaged optics (CPOs), in which optical engines are placed closer to the main chips than is possible with external pluggable transceivers, dramatically improving data communication speeds between server racks.
An optical engine is an integrated chip package component that combines electrical computing chips with photonic chips to replace external pluggable transceivers, reduce size and increase transmission speeds.
Even pluggable transceivers, which have existed in the optical industry for more than a decade, are evolving more rapidly than ever. While it took more than five years for 100G transceivers (industry abbreviation for 100 gigabits per second) to gain global widespread use, it took less than a year for the industry to migrate from 400G to 800G and 1.6T (terabit per second) modules because of the AI infrastructure buildout.
“The order forecast our client gave us till 2028 equals the capacity of some 20 new production lines, but we currently only have three … and that is just for one client. We’re on the verge of making a big decision that might affect our company’s future for the next decade or beyond,” said a manager with another pluggable transceiver supplier. “The market demand for optical modules is greater than our capability.”
Another constraint is emerging in mSAP printed circuit boards, on which transceiver modules are built. Several quarters of short supply have pushed up prices, attracting many PCB makers that did not make mSAP products before, such as Dynamic Holding of Taiwan, to foray into the segment.
Lasers are becoming yet another bottleneck, according to Nikkei Asia‘s supply chain checks.
Previously, the driving force for pluggable optical transceivers in which these lasers are used was telecom equipment and traditional, general-purpose servers. Their requirements for lasers were far smaller and demand cycles were relatively manageable. Now, the AI data center boom and rise of next-generation CPOs are driving up demand for laser sources.
Laser-making, in turn, requires indium phosphide (InP) substrates as a key base material. The sudden surge in demand has pushed InP substrate supplies into significant constraint, sources told Nikkei Asia. In addition, China’s export controls on indium and other critical raw materials amid trade tensions with the US have further complicated the supply situation.
The optical boom has pushed related companies into the spotlight. UK-based IQE, which specializes in InP wafer processing, has seen its stock soar more than 800% this year. Shares of US-based market leaders Coherent and Lumentum have climbed more than 100% and 150%, respectively. Meanwhile, major Japanese laser suppliers Sumitomo Electric and Mitsubishi Electric have seen share gains of more than 85% and nearly 40%, respectively, this year.
Critical suppliers in this field are mainly from the US and China. The latter has built a rather complete supply chain ecosystem for optical transceivers, from materials, fibers, components and parts to modules, though it still relies on American chip developers for the main processor chip.
China’s top optical transceiver maker Zhongji Innolight, a major supplier for Google data centers, has seen its revenue surge nearly 300% since 2022, while profit skyrocketed more than 838%. Eoptolink, a key optical transceiver supplier to Nvidia and Amazon, has seen its revenue surge more than 657%, while profit jumped over 957% since 2022, when OpenAI’s ChatGPT first became publicly available.
The global market for AI-related optical transceivers is projected to expand from USD 16.5 billion in 2025 to USD 26 billion in 2026, according to research company Trendforce.
Commercialization of CPOs being pursued by Nvidia and others may also hit a tipping point this year. The key concept of CPO technology—placing optical engines closer to the main chipsets to boost data transmission speed—could eventually mean that fewer pluggable transceiver modules are required for the same performance.
The higher the transmission speed required the greater the demand for transceivers or optical engines and the CPO approach can save a massive space in the server rack while significantly improving transmission speed and efficiency.
Nvidia and Broadcom both introduced new CPO designs in 2025, with production and commercial shipments set to begin this year. Foxconn, a key Nvidia supplier, recently announced it will start shipping a CPO switch tray—a total solution that integrates the CPO module into a system—in the third quarter of this year.
The CPO market is projected to exceed USD 20 billion by 2036, a compound annual growth rate (CAGR) of 37% from 2026 to 2036, according to market research company IDTechEx. Bank of America Research estimates total CPO revenue at just USD 100 million this year, but projects it will surge to USD 15 billion by 2030.
Although increasing adoption of CPOs is expected to gradually reduce the use of external pluggable optical transceivers, most industry executives believe the two optical solutions will coexist for a long time due to the costs and difficulty of aftersales maintenance.
“I know many people worry about the disappearance of transceivers, but don’t forget that one of the perks for the pluggable module is easy aftersales maintenance,” said an executive with a maker of optical transceiver modules. “With the CPO approach, if an optical engine has a problem, then the entire CPO has a problem and it is not feasible to fix it.”
The optical transformation, which is just at its nascent stage, has brought massive opportunities but also numerous challenges.
“Electronic materials have fundamentally different characteristics from optical materials and components, making integration and testing especially complex,” said Collins Sun, technical director at WinWay Technology, a key Nvidia chip testing solution provider. “The integration, design, manufacturing, and testing of CPO systems remain highly complex, with no single company able to solve all the challenges alone, making close collaboration across the technology supply chain essential.”
This article first appeared on Nikkei Asia. It has been republished here as part of 36Kr’s ongoing partnership with Nikkei.
Note: JPY figures are converted to USD at rates of JPY 159.31 = USD 1 based on estimates as of May 29, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.