Key Takeaways:
The next bottleneck in AI infrastructure isn't a GPU — it's a compound semiconductor called indium phosphide, and supply already falls 30% short of demand.
Lumentum Holdings Inc. Chief Executive Officer Michael Hurlston warned that indium phosphide, a niche semiconductor material essential for AI data center optics, faces a supply-demand gap exceeding 30% — worse than the memory industry's crunch.
"Between the two of us, I don't think we can service the demand that NVIDIA and others are now putting on us," Hurlston said at the RAISE Summit. "The shortage of indium phosphide will become even more acute than what we see from the memory guys."
Lumentum posted Q3 FY2026 revenue of $808.4 million, up 90.1% year over year, with non-GAAP operating margin expanding 700 basis points sequentially to 32.2%. On the earnings call, Hurlston quantified the supply-demand imbalance at "somewhere greater than 30%," adding that pump laser constraints are "probably the biggest issue and were somewhat unanticipated." Key components are "effectively sold out for the foreseeable future," he said.
The shortage threatens to delay AI infrastructure buildouts at hyperscalers including Microsoft Corp., Amazon.com Inc., and Alphabet Inc., which rely on InP-based lasers inside every data center transceiver, pump laser feeding optical amplifiers, and co-packaged optics engines next to GPUs. Lumentum guided Q4 revenue of $960 million to $1.01 billion with operating margins of 35% to 36%, showing that pricing power persists.
Why InP Matters
Indium phosphide is a compound semiconductor that emits light — a property that makes it indispensable for converting electrical signals into optical ones inside AI data centers. Unlike silicon, which handles computation, InP enables the high-speed data transmission that shuttles information between thousands of GPUs in parallel. Every transceiver, every pump laser for optical amplification, and every co-packaged optics module relies on it.
The scale shift is unprecedented. Telecom customers historically deployed lasers in the hundreds. AI data centers now require hundreds of millions. Ramping from thousands of wafers to millions in a non-silicon material, Hurlston said, is "no small feat." Lumentum operates five InP fabs it is trying to scale simultaneously.
Who Wins, Who Loses
NVIDIA Corp. has invested directly in both Lumentum and Coherent Corp., the two dominant InP producers, showing how critical the material has become to its GPU ecosystem. Coherent received a $2 billion R&D investment from NVIDIA and is on track to double internal InP output by year-end 2026, with plans to more than double again by 2027. Its Datacenter and Communications segment now accounts for 75% of revenue, up from 41% a year ago.
AXT Inc., the InP substrate supplier, raised $632.5 million to fund capacity at its Tongmei facility. Chief Executive Morris Young warned that "capacity will become a critical enabler." Applied Optoelectronics Inc., a transceiver maker, posted Q1 datacenter revenue of $81.4 million, more than double year over year.
Despite the bullish fundamentals, all four stocks have sold off sharply in the past month. Lumentum is down 22.3%, Coherent 37.6%, AXT 39.9%, and Applied Optoelectronics 35.7%. If Hurlston's shortage call is correct, the selloff represents a disconnect between price action and underlying demand.
The InP shortage creates a supply-constrained environment where producers hold pricing power. Lumentum trades with operating margins expanding toward 36%, and its Q4 guidance implies continued acceleration. Coherent's InP capacity doubling timeline positions it to capture incremental demand through 2027. The risk lies in execution: ramping non-silicon fabs to millions of wafers carries technical hurdles that could prolong the shortage and delay AI infrastructure timelines for hyperscalers.
This article is for informational purposes only and does not constitute investment advice.
