Silicon Photonics Bottleneck: Material Shortfall Lags 30%

The race for the next generation of artificial intelligence hinges not just on computing power, but on a crucial bottleneck in the supply chain: the materials that generate light. At the heart of this high-stakes struggle is indium phosphide (InP), the compound semiconductor that powers every laser in an AI data center. Recent statements from industry leaders have illuminated a brewing supply crisis, suggesting that the demand for these components is about to outpace production capacity far more aggressively than we’ve seen with memory chips like DRAM and NAND.

Lumentum CEO Michael Hurlston recently put this concern into sharp focus at the RAISE Summit in Paris. He argued that the supply squeeze facing InP is set to intensify, revealing a profound imbalance between what customers need and what manufacturers can deliver. While telecom clients may have initially purchased lasers in the hundreds, the appetite from giants like Nvidia and hyperscalers is now demanding hundreds of millions.

The situation creates an immediate tension for suppliers. Lumentum operates five indium phosphide fabrication plants, yet despite this capacity, the company is still shipping less than 30% of what customers require. This disparity underscores the difficulty of scaling production when the market demand curve is climbing so steeply.

In response to this mounting pressure, Nvidia made a strategic move in March by committing $2 billion to Lumentum and its main competitor, Coherent. This investment was designed not just for research and development, but also to secure future capacity access and bolster the supply chains necessary to solve the complex resistance problems inherent in data center interconnects.

Hurlston noted that navigating this demand is becoming nearly impossible: “Between the two of us, I don’t think we can service the demand that Nvidia and others are now putting on us to solve this resistance problem in the data center.”

The Physics of Light and Bottlenecks

The need for indium phosphide is rooted in physics. Unlike silicon, which is excellent at guiding and manipulating light but cannot generate it, indium phosphide has a direct bandgap, making it exceptionally efficient at converting electrical current into photons. Consequently, every silicon photonics platform—including those from Nvidia, Broadcom, and Cisco—still requires an InP laser somewhere within the package to provide the necessary light.

As technology shifts toward advanced architectures like co-packaged optics, the location and mounting of this laser become critical. While advancements allow for more efficient switching, they don’t eliminate the fundamental requirement for the indium phosphide component, which remains a non-negotiable part of the system’s bill of materials.

Nvidia’s innovations, such as their silicon photonics switches, boast significant gains in power efficiency and network resiliency. However, these advancements highlight the core challenge: the exploding demand for connections means that every port requires more of this critical light-emitting component.

Scaling Capacity and The Path Forward

The industry’s path to alleviating the crunch involves massive scaling efforts, particularly in moving beyond legacy manufacturing processes. Coherent is advancing rapidly with its 6-inch indium phosphide line, which yields more than four times the number of devices compared to its older 3-inch line at less than half the cost. This internal capacity is poised for explosive growth, with expectations that production will more than double by the end of the June quarter and again by the end of 2027.

This transformation represents a fundamental shift away from traditional chip manufacturing methods, moving toward larger wafers—a transition that parallels the move silicon made in the 1980s. Lumentum’s commitment includes expanding its footprint, with its fifth fab announced for 4-inch and 6-inch compatibility, ramping up around 2028.

Despite these challenges, analysts suggest the supply pressure may ease over time. Market forecasts indicate that current transceiver demand is roughly 30% above supply, but projections point to shortages resolving by the end of 2026 as wafer-size transitions come online with higher yields. This long-term fix—focused on production scale rather than new fab construction—offers a concrete roadmap for navigating the inevitable high-demand environment.