The AI arms race is expanding beyond GPUs, creating critical shortages in the optical and power components essential for data center growth.
The AI arms race is expanding beyond GPUs, creating critical shortages in the optical and power components essential for data center growth.

The insatiable demand for artificial intelligence is exposing critical new bottlenecks in the global supply chain, threatening to slow the multi-trillion dollar data center buildout. While investors focus on GPU and HBM memory availability, severe shortages are emerging in more foundational components, specifically the indium phosphide (InP) required for high-speed optical lasers and the power-grid hardware needed to manage massive, fluctuating electricity loads.
"The indium phosphide situation is very, very, very bad," technology analyst @bubbleboi said in a recent research interview. "It's a complete disaster. A lot of people don't realize how bad it is yet."
The problem stems from the move to co-packaged optics (CPO) in AI clusters, which require higher-power, continuous-wave lasers with better noise performance. This drives up demand for the underlying InP wafers, a market already in a state of "disaster" from mining to fabrication. The shortage is so acute it is forcing a market inversion, with the next-generation 1.6T transceivers being dominated by silicon photonics from day one, as manufacturers divert scarce InP capacity to more profitable CW lasers. At the same time, the volatile power draw of GPU clusters—swinging by hundreds of megawatts as chips switch between computation and communication—is creating what the analyst called a "satanic nightmare" for grid operators, stalling permits for new data centers.
These bottlenecks, while a risk to hyperscalers like Amazon and GPU makers like Nvidia, present a significant opportunity for a handful of specialized component suppliers. Companies that can solve the optical and power challenges, such as Lumentum Holdings Inc. in photonics and Wolfspeed Inc. in power semiconductors, are becoming as critical to the AI buildout as the chip designers themselves. Nvidia's own multi-billion dollar partnership and equity investment in Lumentum, noted in a May 2026 Seeking Alpha report, validates the strategic importance of securing this part of the supply chain.
The scarcity of indium phosphide is a multi-layered crisis. The entire production chain, from processing the raw material into crystalline form, manufacturing wafers, performing epitaxy, and printing the final lasers, is severely supply-constrained. This has created a massive opportunity for the few companies that dominate this space, including Lumentum, Coherent Corp., and substrate suppliers like AXT Inc. and IQE PLC.
Geopolitical tensions are adding another layer of risk. As the US and China clash over critical materials, Beijing's restrictions on indium phosphide exports are further tightening the market, according to industry sources. This makes domestic and allied production from companies like Lumentum and AXT even more critical for US-based AI development. The situation is so severe that it is fundamentally altering product roadmaps, forcing a premature shift away from traditional EML-based transceivers, which are made from a single InP chip, toward silicon photonics solutions that use a smaller, separate laser component.
Beyond the data center walls, a new battle is being fought for stable electricity. The sheer power volatility of AI training runs has forced software developers to implement crude workarounds. A now-infamous patch introduced in the PyTorch coding framework was explicitly named "Power Plant No Blow Up" to prevent grid instability. Its function: force GPUs to perform "junk calculations" during communication phases, keeping power consumption level at a staggering 500 watts per chip to avoid angering utility providers.
The long-term solution is the adoption of solid-state transformers. Unlike traditional, passive transformers with 12- to 18-month lead times, these advanced systems use wide-bandgap semiconductors like silicon carbide (SiC) and gallium nitride (GaN) to dynamically regulate power loads. This allows data centers to get power permits and ensures grid stability. Analyst @bubbleboi predicts the technology will "take off" within 36 months, driving a significant re-evaluation for power semiconductor firms like Wolfspeed, Onsemi, and Infineon, many of which are currently in a cyclical downturn due to weakness in the electric vehicle market.
For investors, the message is clear: the AI revolution is powered by more than just the headline-grabbing GPUs. The underlying infrastructure of light and electricity is facing a supply crisis that creates significant risk for unprepared players but offers a generational opportunity for the specialized companies that provide the picks and shovels. While Wolfspeed's financials currently show negative gross margins, the analyst noted the stock "could 5x" on a market turn, highlighting the extreme leverage to the AI power thesis.
This article is for informational purposes only and does not constitute investment advice.