Source

Tags: ai-power, bloom-energy, be, smr, nuclear, oklo, nuscale, terrapower, kairos, onsite-power, fuel-cells, 800v-dc, deliverability, grid-bottleneck

• Source: The Next Bloom Energy?
• Author: Nutty (@NuttyCLD)
• Date published: 2026-05-01
• Date captured: 2026-05-01

TL;DR

Side story (not Part 5) to NuttyCLD's AI Power Crisis series. The investor question changed from "who can provide the biggest power source someday?" to "who can power the GPUs now?" SMRs (Oklo, NuScale, TerraPower, Kairos) answered the first question on paper but stumbled on the second — NuScale's first major project cancelled, Oklo's data center commitments long-dated and largely non-binding, commercial delivery on long timelines. Bloom Energy ($BE) put three words AI wants in the same sentence: onsite power, fast deployment, 800V DC. Not because BE is a perfect answer — because the market is repricing the entire "AI power crisis" sub-thesis around deployment speed and orders/backlog rather than gigawatts-announced.

Key takeaways

• The series argued one thesis: the next decade of AI infrastructure is about who delivers power to the GPU with the least loss. This piece argues the parallel question that AI capex 2026 is asking: what do GPUs do while the grid takes 4 years to fix?
• The split: "people who can fix the grid will keep fixing the grid. People who can't wait will route around it." → onsite gener...