01 Trigger Event

On September 25, 2026, Boom Supersonic CEO Blake Scholl confirmed in a TechCrunch report that Crusoe has removed its $1.25 billion stationary power plant (fixed gas turbine generator set) collaboration from its near-term roadmap.

Crusoe is the non-traditional cloud player that has been repeatedly talked about during the AI infrastructure wave of the past few years. It built its early business on stranded gas (associated gas / idle natural gas) powering GPU clusters, then steadily expanded into modular data centers. Boom Supersonic, the company making supersonic passenger jets, made a pivot around 2024 that I still find fascinating: it repurposed the IP accumulated from supersonic work (high-temperature alloys, axial-flow compressors, combustion chambers) into stationary generator sets targeting AI data centers, betting on the narrative that "distributed gas turbines = local AI power."

A $1.25 billion order book was, at the time, a marquee deal that reinforced the "AI power shortage" story. Specific contract details (number of units, delivery cadence, whether O&M was included) I could not find in public reporting, so I'm extrapolating from total scale.

02 What This Really Means

The issue is not that Boom's turbines underperform, nor that Crusoe suddenly turned conservative.

It's more likely one of two scenarios — possibly both: either Crusoe has already locked in a cheaper power path (grid interconnection rights in mature markets like PJM / ERCOT, long-term pipeline gas contracts, even nuclear PPAs); or Boom's turbine engineering timeline simply failed to keep pace. Hitting 60Hz stable output, grid interconnection compatibility, and long-term load-following — the hard metrics that hyperscalers need to sign a check — takes far longer than the timelines on launch-day PPTs.

Either way, this event draws a line through the distributed turbine path: by mid-2026, it had not matured to the point where mainstream buyers were willing to bet at scale.

What will actually get repriced is this — any infrastructure company still pitching a "proprietary / distributed energy" story will have burn rate running at least an order of magnitude above capacity revenue, all the way through 2027 - 2028 before the picture clarifies.

Crusoe no longer lists Boom's stationary turbines in its near-term plans.
— Blake Scholl, CEO, Boom Supersonic

03 Historical Analogy / Structural Comparison

This is the same script as the early 2020s narrative around SMRs (small modular reactors) powering data centers. NuScale pulling its Utah project, TerraPower still excavating foundations in Wyoming, Bill Gates offering public endorsement — that wave made storytelling easy, but signing real PPAs hard. The ones that broke through were players like Talen and Constellation, who grabbed existing nuclear capacity slots inside PJM. Their play was not building new generation, it was capturing existing interconnection rights.

Crusoe + Boom is the same script replayed on the turbine track. Under conditions of extreme AI compute hunger, the fastest path is never "build a new generator," it is "find existing fuel + existing grid interconnection points."

Another comparison is Boom's own corporate trajectory: pivoting from supersonic passenger jets to stationary power generation is a capital-intensive engineering company's survival pivot when major airline orders fail to materialize. IP can be monetized across industries, but that doesn't mean the product can land a $1.25 billion check from a hyperscaler-grade buyer within 24 months.

This mirrors the 2010s story of small-scale distributed generation companies trying to serve datacom / colo clients directly — the same structure. Engineering demos that miss unit economics always get turned away at the door when facing hyperscalers' demanding SLA and LCOE requirements. Between an engineering demo and a dispatchable, financeable, maintainable generation asset, there is a gap of at least 24 months.

04 What This Means for AI Builders

There is a window over the next month or two worth acting on. The assessments below are working hypotheses extrapolated from a single signal, given that I don't have internal data. Readers should weigh them accordingly.

First, if you or your clients are running GPU clusters, Crusoe's choice is a clear but early signal: 2026 was the year to bet on grid interconnection and pipeline gas, not distributed turbines. For capex planning, the power assumption should shift from "assume we can find unconventional power" to "assume we must squeeze into the existing PJM / ERCOT interconnection queue." This matters especially for teams making build vs. rent decisions — internal IRR models for self-generation need their sensitivity runs re-run over the coming months.

Second, if you're watching AI infrastructure in primary or secondary markets, this gives a simple filter: any company treating "proprietary energy" as its primary moat — whether SMR, turbine, or geothermal — will see burn rate significantly outpace capacity revenue through 2026Q4 - 2028Q2. During this AI capex window, such companies will almost certainly need refinancing to keep going.

Third, opcx.ai sells model access rather than cabinet slots, but a meaningful share of its customers run their own fine-tuning and inference clusters. Their colocation and power decisions directly determine how much they spend on tokens here each month. If signals like Crusoe's keep appearing (Applied Digital, EdgeConneX may make similar moves), then the judgment that "AI application-layer token pricing will continue to decline because of underlying power abundance" gains a concrete supporting point.

05 Counterpoints / Risks

I could be wrong in several ways.

First, this may not be an "AI power inflection" signal at all, just a routine adjustment to Crusoe's business mix. It may have locked in a different natural gas long-term agreement or pipeline gas PPA as early as 2025, and Boom was merely an option on a 2024 slide deck — cutting it now doesn't necessarily signal a shift in industry narrative. I may have over-read "one company's procurement choice" as "a structural inflection in the AI power shortage narrative," a classic over-read I should flag for myself.

Second, I have not run Boom's actual thermal efficiency curve and LCOE internally, so the "distributed turbines won't make it within 24 months" judgment is based on historical analogies from other technology paths (SMR, distributed solar), not on Boom's real performance. Maybe this generation of Boom units is ready, and the Crusoe procurement side simply isn't willing to pay for unit engineering risk — that's a completely different story.

Third, an explanation I don't like but cannot rule out: Boom Supersonic itself may not secure the next round of financing for its turbine business, and Crusoe's exit as anchor customer may stem from supplier credit concerns rather than product problems. If so, my entire previous interpretation is off-target.

Finally, the most hawkish possibility: AI compute demand begins to decelerate in H2 2026, and Crusoe simply doesn't need as much new generation — neither Boom turbines nor anyone else's. This explanation doesn't square with current hyperscaler capex guidance, so I'm not buying it yet, but I'm listing it so readers can weigh it themselves.