-
Crusoe is a startup valued at over $1 billion that is taking a new approach to powering data centers
-
It is focused on tapping into trapped, wasted or clean energy sources
-
AI demand is driving a rapid expansion of its infrastructure
We said it before but we’ll say it again artificial intelligence (AI) workloads will consume a lot of energy. Particularly in the U.S., that’s a problem given how strained the electrical grids in many areas already are. But Crusoe has an interesting solution. Instead of bringing the power to the data center to run its cloud, it’s building modular facilities in the field to capture energy sources that would otherwise be wasted.
Patrick McGregor is Crusoe’s chief product officer, having joined the company earlier this year after spending eight years of Google, six of which were spent as its head of AI. He told Silverlinings AI is one of the driving forces behind Crusoe’s work.
“By the end of the decade, the majority of all computation running anywhere – that means on phones, on laptops, in the cloud – will be in the service of neural networks to power artificial intelligence,” he said. “Artificial intelligence is going to be the reason computers exist not a reason, so there’s a tremendous amount of capacity and infrastructure that has to be built out in this period of time in order to satisfy the demand.”
A fresh forecast from IDC predicted spending on generative AI solutions – including software, infrastructure hardware and IT services – will jump from nearly $16 billion in 2023 to $143 billion in 2027. By the end of its forecast period, spending on generative AI will account for just over 28% of overall AI spending compared to 9% this year.
Crusoe’s mission, McGregor said, is to find a way to build the underlying infrastructure for AI as fast as possible using sustainable energy sources. Specifically, McGregor said it is working to tap into “stranded, wasted or clean” energy sources. That includes sources like natural gas remnants that typically come up with oil deposits as well as wind energy.
These sources – and natural gas in particular – are typically hard to leverage because it is cost prohibitive to build the pipelines necessary to transport the gas to an area where it can be used. So, oil wells generally burn off the natural gas in giant flares to dispose of it.
To capture rather than waste this resource, Crusoe has modular data centers roughly the size of semi-trucks that can be deployed on-site in oil fields, next to windmills or in other locations. McGregor said each boasts “multiple supercomputers” which are supplied by HPE and based on NVIDIA’s H100 GPU chips. These supercomputers can be used by customers looking to tap into extra capacity for their AI engineering teams or for scientific research focused on things like weather forecasting or medicine development.
By the end of this year, Crusoe will have around 200 data centers, most of which are in the U.S. But it’s also got its eye on Argentina and recently secured financial support from Middle Eastern investors, making that region another area of focus. McGregor said looking ahead Crusoe expects to exponentially grow its number of data center facilities, more than doubling them in a short period of time. It is also looking at building bigger data centers that are about the size of a warehouse, he said.
“The growth is insane,” he said. “We can’t build this stuff fast enough. We’re sold out the day that we announce that we’re even planning on deploying stuff.”
Connectivity infrastructure and software are two other areas of focus. McGregor said the company has been working to build robust fiber infrastructure to connect its data centers and is also innovating with software defined networking that allows it to dynamically reconfigure which data centers are connected to each other.
Thus far, Crusoe has raised more than half a billion dollars to fuel its mission. In its most recent funding round in April 2022, the company bagged $350 million in equity funding and $155 million in debt capital.
Rumor has it some more news from the company is imminent, but we’ll have to sit tight to wait and see.
Want to discuss AI workloads, automation and data center physical infrastructure challenges with us? Meet us in Sonoma, Calif., from Dec. 6-7 for our Cloud Executive Summit. You won't be sorry.