Optical supplier booked out to 2028
Lumentum says demand tied to hyperscalers and the AI boom has its production capacity booked through 2028, reflecting accelerating orders for optical components that feed large data centers. That kind of multi‑year backlog suggests supply constraints will persist for parts of the AI hardware stack. (x.com)
Lumentum makes the laser parts that let data move through fiber as pulses of light, and its chief executive said on April 10 that customers are ordering so fast the company could be sold out through all of 2028 within two quarters. Those customers are the giant cloud companies building artificial intelligence data centers at a pace Lumentum says it cannot fully match yet. (bloomberg.com) Those parts sit in the networking gear between racks of processors, where copper cables run out of room and light takes over. Lumentum says its products are used for high-speed fiber links in cloud data centers, artificial intelligence and machine learning systems, and communications networks. (investor.lumentum.com) The reason optics suddenly matter so much is that an artificial intelligence cluster is not one computer but thousands of processors that have to swap data constantly while training a model. If those links slow down, expensive processors sit idle, so buyers now treat optical parts like a bottleneck instead of a commodity. (investor.lumentum.com) Lumentum is talking about indium phosphide devices, which are semiconductor laser components made from a compound that is especially good at generating light for fiber networks. The company said on March 26 that a new 240,000-square-foot factory in Greensboro, North Carolina, will make indium phosphide-based optical devices for the world’s largest artificial intelligence data centers. (investor.lumentum.com) It is also adding capacity in Japan, with Bloomberg reporting plans to invest at least $100 million in a key Tokyo-area factory and a neighboring site. That tells you this is not a software shortage that disappears with an update but a factory problem that needs buildings, tools, yields, and trained workers. (bloomberg.com) Lumentum has already been warning investors that demand is outrunning supply. In its February 3 fiscal second-quarter presentation, the company listed manufacturing expansion and supplier capacity among the main risks to meeting forecast demand. (investor.lumentum.com) The company’s own recent numbers show how fast the mix has changed. When Lumentum reported full-year results for the year ended June 28, 2025, chief executive Michael Hurlston said the company had seen robust demand across its cloud portfolio supporting artificial intelligence data centers and had raised guidance earlier that quarter. (investor.lumentum.com) This is also why Nvidia has been stitching closer ties with optics suppliers instead of relying only on chip output. Lumentum said on March 2 that Nvidia and Lumentum had entered a strategic partnership to develop optics technology for artificial intelligence infrastructure. (investor.lumentum.com) A backlog that stretches into 2028 does not mean every server will be delayed until then. It means the companies spending tens of billions of dollars on data centers are now reserving laser and fiber capacity years ahead, the same way airlines lock in aircraft orders when they think demand will outrun supply. (bloomberg.com) The immediate effect is that the artificial intelligence hardware race is no longer just about who can get the most graphics processors. It is also about who can secure the light engines that connect them, and Lumentum is saying that line is already forming three years out. (investor.lumentum.com)