Lumentum CEO Michael Hurlston told an audience at the RAISE Summit in Paris that the compound semiconductor indium phosphide, used in every laser inside AI data centers, is heading into a supply squeeze more severe than what the industry saw with DRAM and NAND. His warning came as AI-driven demand for co-packaged optics has pushed customer requests into the hundreds of millions of lasers, a leap from the hundreds that telecom buyers once ordered.

What You Need to Know

Indium phosphide lasers are essential for silicon photonics, which allows AI data centers to move data using light instead of electricity. The material's unique ability to emit light efficiently is not easily replaced by silicon, and production capacity has not kept pace with the explosion in demand from Nvidia and other hyperscalers. The shortfall, now exceeding 30%, threatens to slow the expansion of AI infrastructure unless new fab capacity comes online quickly.

The Core Problem: Indium Phosphide Supply

Silicon can guide, split and modulate light, but it cannot generate it. Every silicon photonics platform in production from Nvidia, Broadcom, Marvell and Cisco relies on an indium phosphide laser somewhere in the package to supply the light. Moving to co-packaged optics changes where that laser sits but does not remove it from the bill of materials.

Hurlston noted that while Lumentum runs five indium phosphide fabs, it is still shipping more than 30% below what customers want. The supply-demand imbalance on electro-absorption modulated lasers (EMLs) has widened to above 30%, and pump lasers are even tighter. Lumentum posted record revenue of $808.4 million in its fiscal third quarter, up 90% year over year, yet the gap continues to grow.

  • Indium recovery: Indium is a byproduct of zinc refining, so supply is tied to zinc economics and cannot be scaled independently.
  • Fab transition: The industry is moving from 3-inch to 6-inch wafers, a transition silicon completed in the 1980s. Lumentum's fifth fab in Greensboro, North Carolina, is expected to ramp around 2028.
  • Demand explosion: Nvidia and hyperscalers now order lasers in the hundreds of millions, not hundreds, overwhelming existing production lines.

Nvidia’s Multibillion-Dollar Response

Nvidia’s answer, in March, was to write $2 billion checks to Lumentum and Coherent, the two suppliers that between them make most of the world’s high-speed datacom lasers. The deals included purchase commitments and future capacity access. “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.

Coherent, Lumentum’s biggest competitor, is also racing to expand. Its 6-inch indium phosphide line yields more than four times as many devices as its 3-inch line at less than half the cost, and CEO Jim Anderson told investors that internal capacity would double by the end of June and more than double again by the end of 2027. Coherent’s revenue hit a record $1.8 billion, with data center and communications now 75% of the total.

Why This Matters

This supply bottleneck has direct implications for the pace of AI infrastructure deployment. Every new data center cluster depends on indium phosphide lasers for the photonic interconnects that move data between GPUs at high speed. If wafer production cannot keep up, delays in server shipments and higher costs will ripple across the industry. Lumentum and Coherent are investing heavily, but the material itself is constrained by indium availability, which is concentrated in China. The USGS reports that China produced roughly 69% of global primary refined indium in 2025, and a 72% year-over-year drop in unwrought indium exports between September 2024 and September 2025 underscores the geopolitical risk.

The transition to co-packaged optics, which Nvidia is pushing for next-generation AI clusters, will only increase the demand for high-power continuous-wave lasers and external laser modules. Both Lumentum and Coherent are ramping, but the gap between supply and demand is widening. For hyperscalers and AI companies, the laser shortage may become the next critical bottleneck after GPUs and memory.

What Comes Next

Nvidia’s funding ensures that Lumentum and Coherent have capital to expand, but new fabs take years to come online. Lumentum’s Greensboro facility will not ramp until around 2028, and Coherent’s doubling plans still leave a significant gap against demand that is growing faster than capacity. The industry’s reliance on indium, a byproduct metal, adds a further layer of vulnerability that cannot be solved by fabs alone.