Six Months Ago, Lumentum's AI Laser Backlog Ran Through 2028. Now It's 2029.

Michael Hurlston runs a company most generalist investors had never heard of before this year. On Friday, in a Bloomberg Television interview from Tokyo, Lumentum's CEO said the company's optical components for AI data centers are completely sold out through early 2029. Six months ago, his own timeline stopped at 2028.

The specifics matter more than the sound bite. For some product lines, Lumentum can currently fill only about 30% of what customers are ordering through next year, leaving 70% of demand with no delivery date. For other products, the company is meeting roughly 70% of demand through 2028, with the remaining 30% still waiting in line. New laser manufacturing capacity takes three to five years to bring online, so relief, if it comes, likely will not arrive until 2029 or 2030.

[[IMG1]]

This is not the first time Nvidia has tried to buy its way around a bottleneck. In March 2026, the company invested $2 billion each in Lumentum and Coherent, pairing the capital with multiyear purchase commitments for the lasers and optical-networking components that move data between AI chips inside a data center. Jensen Huang tied the deals to what he called the largest computing infrastructure buildout in history. Seven months later, the two companies Nvidia bankrolled still cannot keep up.

The underlying numbers explain the urgency. Lumentum's latest quarterly revenue more than doubled year over year to just over $1 billion, its eighth straight quarter of growth, and guidance for the current quarter points to roughly $1.25 billion, up more than 130% from a year earlier. The company is converting a newly acquired fab in Greensboro, North Carolina, into indium phosphide laser production, and it is ramping pump-laser capacity at an existing US site. Lumentum's own executives say the Greensboro plant will not come online until 2028.

Why It Matters

Most of the AI infrastructure story so far has centered on two scarce inputs. Nvidia's chips sell out almost as fast as the company can make them, and finding electricity to run them has turned utilities into AI plays overnight. Optical components are a third, quieter bottleneck, and arguably a more useful one for judging whether AI demand is real. Chip orders can be inflated by customers hedging against future shortages. A hardware shortage that stretches three years out, confirmed by a supplier with no obvious reason to talk up its own backlog, is harder to wave away as hype.

That distinction matters this week because the market has spent the past few days arguing about exactly this question. OpenAI's revenue came in roughly $20 billion short of what investors had been told, and chip stocks tied to its buildout sold off on the news. Lumentum's backlog points the other way: whatever is happening to any single AI company's own revenue, the physical infrastructure underneath the broader buildout is still oversubscribed years in advance. It is one reason Marvell's decision to raise its 2028 AI revenue forecast landed differently than a routine earnings beat would have: the demand side and the supply side are telling a similar story, even as market breadth outside AI names stays historically narrow.

There is a less flattering way to read the same facts. Nvidia's $4 billion across Lumentum and Coherent looks less like a passive investment and more like vendor financing, with the chipmaker bankrolling the suppliers it needs the same way it has backed customers on the other side of the ledger, including SpaceX's plan to borrow $40 billion partly to buy Nvidia chips. When one company's balance sheet underwrites both its suppliers and its customers, a demand signal and a financing arrangement start to look harder to tell apart.

What to Watch Next

Coherent, Applied Optoelectronics, and Corning all report earnings in the coming weeks, and each call is a chance to confirm or complicate Hurlston's timeline. China complicates the picture from the supply side rather than the demand side: the country controls roughly 69% of the world's refined indium, the raw material behind every laser in this story, and Beijing's 2024 export controls have already cut shipments of it out of the country. Capital spending guidance from Microsoft, Alphabet, Amazon, and Meta, due alongside their own earnings later this month, will show whether any of them are quietly slowing deployment because the optical parts are not there, rather than because demand has cooled.

The Pulse24 Take

A shortage this severe cuts in more than one direction. In the near term, it is good news for the small group of companies holding the scarce capacity, and it is independent evidence that AI infrastructure spending is responding to a real constraint rather than momentum alone. Chip supply and power access have driven most of the AI buildout debate so far. Optics is now forcing the same conclusion: hyperscalers are trying to build faster than the physical world allows.

The less comfortable question is what happens once the capacity Nvidia is funding today actually arrives. The last time an entire industry convinced itself that network bandwidth demand would outrun supply forever was the optical buildout of the late 1990s, when names like Nortel and JDS Uniphase rode that logic to record valuations before new capacity, and a demand slowdown, arrived at the same time and crushed both stocks. Lumentum's own history runs through that era: the company was spun out of JDS Uniphase in 2015, long after the bust that nearly erased its predecessor's value. None of this means today's shortage is a bubble. It does mean a three-to-five-year lead time is a long way to extrapolate demand, and the companies cashing in on scarcity now are the same ones whose stock prices will move the most if that scarcity eases faster than expected.