Optical Communication Becomes New AI Bottleneck! Lumentum: This 'Semiconductor Material' Shortage Could Be Worse Than Memory Crisis

As demand for optical connectivity in data centers grows exponentially, a compound semiconductor material rarely in the spotlight—indium phosphide (InP)—is rapidly emerging as the most vulnerable link in the AI supply chain. Industry insiders warn that the severity of this shortage could surpass the current, widely watched memory crisis.

Michael Hurlston, CEO of optical components giant Lumentum (LITE-US), stated bluntly at the RAISE summit held in Paris this month that the supply tightness of indium phosphide will exceed that of the memory market.

He revealed that Lumentum currently operates five indium phosphide wafer fabs, but its shipment volume is still more than 30% below customer demand, and this gap is even larger than in the previous quarter.

Hurlston candidly admitted: "Even if we combine the production capacity of both our companies (Lumentum and Coherent (COHR-US)), it might still not be enough to meet the demand from NVIDIA (NVDA-US) and other customers."

Market research firm SemiAnalysis has issued similar warnings, pointing out that continuous-wave distributed feedback (CW DFB) indium phosphide lasers used in near-packaged optical interconnects are particularly strained.

Currently, the industry mostly uses two- to four-inch wafers, far behind the twelve-inch processes commonly used in silicon-based CMOS chips, creating structural limitations for capacity expansion.

These warnings have already sent shockwaves through the market. In March, NVIDIA invested $2 billion each in Lumentum and Coherent, attempting to secure supply from these two companies, which together control the majority of global high-speed optical communication laser capacity.

However, related stocks have been heavily hit over the past month. Lumentum plunged 22.3%, Coherent dropped 37.6%, substrate supplier AXT (AXTI-US) fell 39.9%, and transceiver module maker Applied Optoelectronics (AAOI-US) declined 35.7%, with stock movements clearly decoupling from fundamental performance.

Silicon Photonics Cannot Avoid Indium Phosphide

Analysis indicates that indium phosphide is irreplaceable due to its intrinsic physical properties.

Silicon is an indirect bandgap material, capable only of guiding, splitting, and modulating light signals, but unable to emit light itself. In contrast, indium phosphide has a direct bandgap of approximately 1.34 electron volts, enabling efficient conversion of electrical energy into photons.

This means that regardless of whether it’s NVIDIA, Broadcom, Marvell, or Cisco, all their silicon photonics platforms must integrate indium phosphide lasers as light sources within the package.

In other words, as the industry shifts from pluggable transceiver modules to co-packaged optics (CPO) architectures, only the laser’s mounting position changes—indium phosphide lasers remain on the bill of materials.

In fact, CPO architectures demand even more difficult-to-manufacture high-power continuous-wave light sources and external laser modules to simultaneously drive multiple optical channels. Coherent and NVIDIA’s collaboration specifically includes such high-power CW lasers, external laser modules, and fiber array components.

NVIDIA states that its photonic switch, compared to traditional pluggable solutions, reduces the number of lasers per port by four times, improves power efficiency by 3.5 times, and increases network resilience tenfold.

The problem is that these benefits are calculated “per port,” while what truly drives the demand curve is the explosive growth in the total number of transmission ports. For example, the high-end configuration of Spectrum-X Photonics supports 512 800Gb/s transmission ports in a single system, achieving a total switching capacity of 400Tb/s.

From 'Thousands' to 'Millions' in Demand Leap

Hurlston illustrated the force of this demand revolution with a contrasting set of numbers.

Telecom customers used to purchase lasers in “hundreds,” whereas today, demand from NVIDIA and major cloud providers is in the “hundreds of millions.”

He said: "Scaling up from thousands of wafers to millions—on a non-silicon material—is no easy feat."

The numbers also reflect in financial reports. Lumentum’s latest quarterly revenue reached $808.4 million, a 90% year-over-year increase. Component business revenue was $533 million, with pump laser shipments up 80% year-over-year, yet the supply-demand gap continues to widen.

During the earnings call, Hurlston stated that the supply-demand imbalance has expanded from the previously disclosed 25% to 30% to "over 30%," with pump laser supply tightness "even exceeding expectations," and core components "effectively sold out for the foreseeable future."

The company’s revenue guidance for the next quarter is between $960 million and $1.01 billion, with an estimated operating margin of 35% to 36%.

Coherent also posted a record quarterly revenue of $1.8 billion, up 21% year-over-year. The share of data center and communications business in total revenue has surged from about 41% a year ago to 75%, and order visibility extends to 2028.

Coherent CEO Jim Anderson stated that the equipment yield of the company’s six-inch indium phosphide production line has surpassed that of the older three-inch line. Internal capacity is expected to double by the end of the June quarter—earlier than originally planned by one quarter—and will more than double again by the end of 2027.

Key Difference from Memory Crisis: Timeline

Despite continuous warnings, market views on how long this shortage will last remain divided.

Market research firm LightCounting predicted in April that current transceiver demand exceeds supply by about 30%, consistent with Lumentum’s disclosed figures.

However, the firm also forecasts that the shortage will gradually ease by the end of 2026 and has downgraded its growth forecast for Ethernet transceivers this year from 82% in 2025 to 65%.

Coherent achieving its capacity doubling one quarter earlier than planned also points to a similar optimistic outlook.

The biggest difference from the memory crisis is that the solution to the indium phosphide shortage is already underway on production lines—wafer size upgrades with yields exceeding older process nodes—rather than building new fabs from scratch, which typically takes three years. This supports relatively optimistic market expectations regarding the timeline.

Yet Hurlston’s warnings carry real weight, as the market valuation of his company heavily depends on how long this shortage persists.

Analysis indicates that the constraints on indium phosphide supply stem not only from manufacturing capacity but also from inherent limitations in raw material sourcing and geopolitical risks. As a byproduct of zinc smelting, indium supply heavily relies on zinc industry conditions and output, making it impossible to rapidly increase production in response to rising laser demand, resulting in structural bottlenecks across the entire supply chain.