A new survey from Bloom Energy projects that capacity additions for AI data centers will climb to 23% by 2030. The findings underscore a looming shift in how energy infrastructure must adapt to meet surging demand from artificial intelligence workloads.
What the survey shows
The Bloom Energy survey points to a steady rise in AI data center capacity additions, reaching nearly a quarter of total new capacity by the end of the decade. That growth is expected to put significant pressure on existing power grids, which weren't designed for the high-density, round-the-clock loads that AI processing centers require. The company's analysis highlights that without targeted investments, the mismatch between demand and supply could create bottlenecks.
The push for onsite power
A key finding from the survey emphasizes the growing role of onsite power solutions. Bloom Energy, which specializes in fuel cell technology, argues that relying solely on grid connections will become increasingly difficult. Onsite generation—whether from fuel cells, solar-plus-storage, or other distributed resources—can help data center operators avoid transmission delays and reduce vulnerability to grid instability. The survey suggests that hyperscale operators and colocation providers are already exploring such options to lock in reliable electricity.
Grid challenges remain
For all the talk of onsite power, the survey doesn't downplay the broader grid challenges. Utilities face long lead times for new transmission lines and substation upgrades, and in many regions interconnection queues are backed up for years. The report notes that without coordinated planning between data center developers and grid operators, those delays could slow AI infrastructure buildouts. The 23% figure assumes that some of those hurdles are addressed, but it's far from guaranteed.
The survey didn't name specific data center operators or utility companies. But it did call for clearer policy signals—things like streamlined permitting for onsite generation and faster interconnection processes—to help bridge the gap between ambition and reality.
Bloom Energy's findings arrive as AI training and inference workloads continue to expand, pushing data center operators to lock down power capacity years in advance. The 2030 projection gives a sense of the scale, but the shape of that growth will depend heavily on how quickly the energy side of the equation can adapt.
