Crusoe is abandoning a massive $1.25 billion agreement with Boom Supersonic that was supposed to bring aviation-derived natural gas turbines to the rapidly expanding world of AI data centers, marking a significant change in how the infrastructure company plans to power its next generation of computing facilities.
Image Courtesy : energiesmedia.com
The agreement originally called for Crusoe to purchase 29 of Boom's Superpower stationary turbines, each capable of producing 42 megawatts of electricity. Together, the systems were expected to provide roughly 1.21 gigawatts of generating capacity for Crusoe's data-center operations, with the first deliveries originally scheduled to begin in 2027.
That plan is now off the table. Boom Supersonic CEO Blake Scholl confirmed on September 25 that the two companies are no longer moving forward with the launch partnership, explaining that turbines are no longer part of Crusoe's near-term primary power mix for its Abilene, Texas, operations and other sites. Crusoe separately confirmed that it is no longer doing business with Boom under the arrangement.
The cancellation represents a notable reversal because the original deal had been presented as a major bridge between the aviation and AI industries. Boom had developed Superpower as a stationary version of the technology being developed for its supersonic Overture aircraft. The company said the stationary turbine shares approximately 80% of its components with its Symphony aircraft engine, allowing Boom to pursue an additional market for the technology while continuing development of its supersonic airplane.
For Crusoe, the attraction was straightforward. AI data centers need enormous amounts of reliable electricity, and the ability to install dedicated natural-gas generation could potentially allow a facility to obtain power without waiting years for a utility connection or major transmission upgrades.
That need has become one of the biggest constraints facing the AI infrastructure industry. Technology companies are racing to construct data centers capable of housing increasingly powerful GPU clusters, but obtaining enough electricity has become a major challenge. In some parts of the country, data-center projects are competing for grid capacity with industrial facilities, manufacturers and existing communities.
Texas provides an especially dramatic example. The state has attracted an enormous pipeline of proposed data-center projects, but concerns over their impact on electricity demand, water use and infrastructure have grown alongside the AI boom. Texas Gov. Greg Abbott recently ordered a halt on new state-issued data-center permits while officials conduct an audit of their impact on the electrical grid. Reuters reported that more than 470 gigawatts of data-center projects were seeking grid connections in Texas, an amount several times greater than the state's peak demand.
That environment helps explain why on-site power has become increasingly attractive to data-center developers.
Rather than waiting for a utility to construct new generation or transmission capacity, an operator can potentially install its own power plant next to a computing campus. Natural-gas turbines can provide electricity continuously and can be deployed independently of some of the constraints associated with the traditional grid.
Crusoe has extensive experience with unconventional energy strategies. The company was originally founded around the idea of using excess natural gas from oil production sites to generate electricity for computing rather than allowing that gas to be flared. As the company's focus shifted from cryptocurrency mining toward AI infrastructure, that experience with energy became one of its defining advantages.
But Crusoe's latest decision shows that having access to a particular power technology does not necessarily mean that technology will become part of the company's long-term infrastructure portfolio.
Crusoe spokesperson Andrew Schmitt said the company builds AI factories "from the power up" and wants to remain flexible in selecting energy solutions for each site. The company said its future portfolio can include turbines alongside wind, solar, batteries and conventional grid power, depending on the requirements of individual locations.
That flexibility is becoming increasingly important as Crusoe expands its business.
The company recently raised $3.9 billion in new financing, giving it a valuation of approximately $30.9 billion, according to reporting from The Wall Street Journal. The financing comes as Crusoe expands beyond its reputation as a builder of enormous AI computing facilities and moves toward a broader infrastructure model involving data centers, GPUs and managed AI computing services.
One of the most significant changes is Crusoe's move toward smaller, modular data centers.
The company is developing what it calls Spark, a smaller data-center format that can be manufactured in factories, transported by truck and deployed in locations where power is available. The approach represents a contrast with the massive campuses that have become synonymous with the AI infrastructure boom.
The strategy is partly a response to the difficulty and expense of building huge data centers. Large projects can require enormous amounts of land, power infrastructure and construction capacity, while obtaining grid connections can take years in some markets. Smaller modular facilities can potentially be deployed more quickly if developers can identify locations with suitable power.
That shift helps put the abandoned Boom agreement into perspective.
The original $1.25 billion order was designed around a very large centralized power requirement. Crusoe's evolving strategy gives the company more flexibility to match the energy source to the individual data center instead of relying on one major turbine program across its portfolio.
Crusoe is not abandoning natural-gas power altogether.
The company's existing and planned Texas infrastructure demonstrates that gas turbines remain an important part of its strategy. Its initial 1.2-gigawatt Abilene data center, developed for Oracle and OpenAI, is powered by the grid, while a gas-turbine plant at the site serves as backup power. Crusoe is also building a 900-megawatt data center in Abilene for Microsoft that is expected to use on-site gas turbines as its primary power source.
That distinction is important. The cancellation of the Boom deal does not mean Crusoe has decided that gas turbines are no longer useful for AI infrastructure. Instead, the company appears to be choosing power technologies on a site-by-site basis rather than committing to Boom's particular turbine platform as a major part of its near-term portfolio.
For Boom Supersonic, the consequences are more direct.
The Crusoe agreement had been the launch customer for Superpower and represented more than $1.25 billion in expected business. Boom had raised $300 million in 2025 to commercialize the stationary power business, with investors including Darsana Capital Partners, Altimeter Capital, ARK Invest, Bessemer Venture Partners, Robinhood Ventures and Y Combinator.
Boom had viewed the stationary-power business as a way to generate revenue that could help finance the development of its Overture supersonic passenger aircraft. Scholl previously compared the strategy to SpaceX's use of Starlink as a business that can support the development of another ambitious technology program.
The loss of Crusoe therefore removes a major early customer from the business, but Boom says it still has other customers in its pipeline. Scholl said the company expects to deliver approximately 250 megawatts of Superpower capacity to other sites in 2027 and is targeting 1 gigawatt in 2028.
The cancellation also highlights the difficulty of introducing new energy technologies into AI infrastructure.
Data centers cannot simply tolerate experimental power systems in the same way that a technology company might experiment with a new software platform. The computing infrastructure inside an AI data center can represent enormous amounts of capital, and interruptions in electricity can potentially affect expensive GPU systems and the services they support.
Reliability therefore becomes just as important as efficiency.
A power system has to operate continuously, respond to changing loads and integrate with the rest of the electrical infrastructure. It also has to satisfy permitting and environmental requirements while making economic sense over the lifetime of the facility.
The economics can be particularly complicated for AI infrastructure because the power requirement is enormous. A 42-megawatt turbine is substantial by conventional standards, but a large AI campus may require hundreds of megawatts or even gigawatts of power.
That means developers often need multiple layers of generation, grid connections, batteries and backup systems rather than a single technology.
Crusoe's approach increasingly appears to reflect that reality.
The company says its campuses can use a combination of turbines, renewable energy, batteries and grid power. Such a diversified approach can potentially give developers more flexibility as energy availability changes from one region to another.
The decision is also occurring as AI data-center economics continue to evolve.
The first generation of AI infrastructure was heavily focused on training enormous models, which required massive clusters of GPUs operating together. Increasingly, companies are also investing in inference—the computing required to actually serve AI models to users.
Inference workloads can have different infrastructure requirements from training. Some can be distributed closer to users and may not require the same enormous centralized computing campuses used to train frontier models.
That is one reason Crusoe's smaller Spark facilities have become an important part of the company's strategy. The Wall Street Journal reported that Crusoe expects a majority of its cloud inference computing to eventually run from these smaller data centers, with the company planning to scale Spark production substantially.
Crusoe's business has also expanded beyond simply building buildings.
The company now operates across several layers of AI infrastructure, including data-center development, GPU leasing and managed inference services. CEO Chase Lochmiller has described the business as selling data centers, GPUs and tokens, reflecting an attempt to capture value at multiple stages of the AI computing chain.
That diversification could make the company's energy strategy more complicated but also potentially more adaptable.
A developer operating one enormous campus might optimize its power infrastructure around that single facility. A company operating multiple data-center formats across different locations has more reason to select whatever combination of power sources works best for each site.
The abandoned Boom partnership therefore appears less like a retreat from AI infrastructure and more like a sign that Crusoe is becoming less committed to a single power solution.
The broader industry is facing the same question.
AI developers need power immediately, while conventional energy infrastructure can take years to plan and build. This has created a market for alternatives ranging from natural-gas turbines and battery storage to nuclear power, renewable generation and other emerging technologies.
Crusoe itself has already demonstrated another unconventional approach through its partnership with Redwood Materials. In Nevada, Redwood built a 12-megawatt microgrid with 63 megawatt-hours of capacity using repurposed electric-vehicle batteries to provide power to a Crusoe modular AI data center.
That project illustrates the variety of energy technologies now being evaluated for AI infrastructure.
The data-center industry is effectively becoming a laboratory for new combinations of power generation and storage because the traditional grid is struggling to keep pace with demand.
But the cancellation of the Boom deal also demonstrates that not every promising partnership will survive the transition from announcement to deployment.
Crusoe and Boom originally announced a highly ambitious plan built around the idea that aviation-derived turbine technology could become a major new source of electricity for AI computing. Less than a year later, the companies have decided that the arrangement no longer fits Crusoe's immediate requirements.
For Boom, the company will now have to demonstrate that Superpower can succeed with customers other than Crusoe. The company says additional projects are moving forward, with hundreds of megawatts targeted for deployment in the coming years.
For Crusoe, the priority appears to be flexibility.
The company is continuing to build enormous AI facilities, including projects serving Microsoft, while simultaneously developing smaller modular systems and maintaining the ability to use multiple energy sources.
That strategy could become increasingly valuable as the AI infrastructure market enters a more complicated phase. The industry is no longer simply asking how quickly it can build more GPUs. It is asking where those GPUs can be located, how they can be powered, how quickly facilities can be deployed and how much electricity each computing workload actually requires.
Those questions are forcing infrastructure companies to rethink the architecture of AI data centers themselves.
Crusoe's decision to walk away from the $1.25 billion Boom turbine agreement is therefore about more than one failed partnership. It reflects the rapidly changing economics of AI infrastructure and the realization that the best power solution for one data center may not be the best solution for another.
The AI industry still needs enormous amounts of electricity, and Crusoe remains firmly positioned within that market. But rather than locking itself into one turbine technology, the company is increasingly emphasizing a mix of power sources and data-center designs.
As AI computing continues to expand, that flexibility may become one of the most important assets an infrastructure company can have.