Investment flows into artificial intelligence (AI) are expanding from graphics processing units (GPUs), memory and networking equipment into power infrastructure. No matter how advanced the semiconductors secured, servers cannot operate if data centers do not receive electricity on time. As competitiveness in the AI data center market shifts from power unit costs to the ability to secure electricity when needed, Bloom Energy's business model is drawing attention.
Bloom Energy is a U.S. company that provides on-site power solutions based on solid oxide fuel cells (SOFC). SOFCs are devices that generate electricity through a chemical reaction between oxygen and fuel. They can be installed on or around data center sites, allowing power to be supplied without waiting for large-scale transmission grid expansion.
Currently, transformers, turbines and transmission grid connections are all cited as bottlenecks in the U.S. data center market. Orders for large gas turbines have piled up, extending delivery times, and grid interconnection has been slowed by permitting and local community concerns. While existing data centers have relied mainly on large power grids and gas turbine generation, Bloom Energy is focusing on reducing the time needed to secure power by installing generation sources at required locations.
The Oracle project is a symbolic case of this shift. Bloom Energy explained that it will replace up to 2.45 GW of power blocks with its own solution in Oracle's large-scale AI data center project. It is significant in that the power structure will be centered on Bloom Energy servers, rather than the existing approach combining diesel generators, batteries, engines and turbines. Fuel cells have begun to be evaluated as a primary power source candidate for AI data centers, beyond simply serving as auxiliary or emergency power.
The key point to watch going forward is capacity expansion. Bloom Energy is responding by adding hundreds of megawatts of capacity each quarter, rather than through one-time large-scale expansions. The company has mentioned that its annual production capacity could expand from the existing 2 GW to up to 5 GW. With AI data center power demand rising rapidly, the key lies not in demand itself but in how quickly supply can match it.
Connectivity with next-generation AI data center architectures is also important. Nvidia's next-generation rack-scale systems require higher power density than before. The power supply method is also expected to shift from the existing 415V alternating current (AC) to 800V direct current (DC). In this process, power components such as transformers, switchgear and rectifiers could emerge as new bottlenecks.
There are risks, however. The share price has already reflected much of the expectations for AI power infrastructure expansion. Heavy reliance on major customers and execution risk during capacity expansion are also variables. Fuel cells are unlikely to completely replace gas turbines or power grids. Still, considering that the bottleneck in AI infrastructure investment is moving from semiconductors to power, Bloom Energy has room to be re-evaluated as an AI data center power platform company, beyond a simple fuel cell maker.
