The Scoop
Crusoe, the startup developing OpenAI’s massive Stargate data center in Texas, is launching a new cloud computing product that puts it into competition with hyperscalers, and possibly even OpenAI.
The move is part of a trend among neocloud companies to tackle the higher-margin services that can be offered on top of the data centers they’re building. Crusoe’s new service will allow companies like Cursor, the fast-growing startup that is among the company’s first cloud clients, to run their own customized models on their own cloud service.
Erwan Menard, senior vice president of Crusoe Cloud, told Semafor in an interview that his pitch to customers is that they can get better performance and lower costs with Crusoe than with major cloud providers, because it can offer more “technical intimacy” than a hyperscaler — which is typically more focused on the biggest AI models.
Eventually, Menard said, Crusoe envisions having lots of smaller data centers, focused on inference, that will be spread around the country, putting them closer to end users and improving latency.
Some of those data centers would operate entirely on their own power grid, utilizing solar and batteries.
In this article:
Know More
Crusoe got its start in Bitcoin mining. It built crypto mining operations adjacent to oil fields, burning the excess natural gas to power the processors.
When the AI boom began, Crusoe was well positioned to help build even bigger data centers, where it continued to innovate on energy production. For instance, it’s now using repurposed supersonic jets from aerospace company Boom as gas turbines for powering AI data centers.
But as the AI industry matures, it’s starting to look more like the cloud industry, where services offer much higher margins. Companies like Crusoe and CoreWeave are pushing to get into that space, and believe they can find a lane by providing more boutique services and higher performance on AI-specific workloads.
Hyperscalers like Amazon, Google, and Microsoft must offer a wide array of products, including ones that have nothing to do with AI. And they serve mass-market models like Gemini, Claude, and ChatGPT. Neoclouds like Crusoe can focus squarely on AI workloads, seeking out better performance for customized models.
Reed’s view
By looking at companies like Crusoe, you learn things about the broader AI market. Menard told me that the number one reason companies want to run customized AI models is to have more control. These companies can’t afford to rely on the big foundation model firms that might change something without warning and break an AI implementation as a result.
But customized models built on open source don’t perform as well as the big ones made by the frontier labs, so they need to really optimize the customized ones as much as possible. Startups don’t want to spend their time on model optimization, which creates an opening for new players to step in and satisfy that market niche.
A quirk of this change is that Crusoe’s strategy actually puts it in competition with OpenAI, perhaps its most important partner. But the market is so massive — and the size of the pie is increasing so fast — that there’s room for competition and for many new startups to profit.
Notable
- Crusoe raised $1.3 billion in a Series E round last October, and is now valued at $13 billion, The Information reported.