Crusoe tackles AI's data center problem

Infrastructure is perhaps the biggest dilemma facing the AI market today. 

Data centers weren’t built with AI in mind. They’re not optimized for high-performance compute and AI workloads. This reckoning has caused many in the industry to turn the hardware stack on its head. Enter neoclouds. 

The rapid spike in high-performance compute demand over the past year has fed into the rise of neoclouds, or clouds purpose-built for AI. 

This is the market that Crusoe, a self-described “AI factory company,” is squarely targeting. Founded in 2018, Crusoe takes an “AI and AI only approach,” building modular data centers designed specifically to pack as many GPUs as it can per square foot, Erwan Menard, SVP of Product at Crusoe, told The Deep View. 

“You can look at AWS, Azure, or Google Cloud [and] you will not find an AI purpose-built form factor,” said Menard. “These are general-purpose, edge data centers that are very CPU rich.” 

Though this focus means that Crusoe offers fewer products than a traditional hyperscaler, Menard said that gives the company a “leaner cost structure.” Additionally, Crusoe’s popularity is a sign that its AI niche is a stark differentiator to conventional cloud providers, and one that its customers are asking for, with a roster including Cursor, Fireworks, Odyssey and Together AI. 

“When you call us, everybody does AI and AI only,” said Menard. “You don't need to hop two or three or five people to get somebody who understands what it takes to improve your inference.” 

This focus has also drawn attention from investors and some of AI’s biggest names. The company closed a nearly $1.4 billion funding round in October at a $10 billion valuation and completed a partnership with OpenAI in November to build a 1.2 gigawatt data center in Abilene, Texas.

Why Crusoe hits every piece of the stack

Crusoe’s success is a sign that the vertically integrated AI factory model is a key differentiator for them.. 

According to Synergy Research Group, revenues from neoclouds are estimated to exceed $180 billion by 2030. However, Crusoe offers something that sets itself apart: The ability to “connect the dots between energy sourcing, data center development and cloud services,” said Menard. 

The issue is that the different pieces of the AI puzzle all work on different timelines. While the energy field speaks about projects in terms of years and decades, data center developers think about projects in terms of quarters. Software folks move the fastest of all, considering projects in terms of weeks and sometimes days. “Coordinating those three processes is not trivial,” said Menard. 

So rather than just coordinating one piece of that stack, Crusoe helps from “electrons to tokens,” Menard said. Its full-stack approach includes: 

• Energy sourcing, feeding into its data centers from a 45 gigawatt energy pipeline, said Menard

• Facility development, with modular data centers that already have a capacity footprint of more than 946,000 GPUs as it stands and turnaround times of less than a year in the best cases

• And cloud and scalable managed inference services for AI developers

“I'm absolutely convinced that having those three [areas of] expertise in-house… we can take all these parameters, put them in a room and make a decision,” said Menard. “It is translating into speed from a customer point of view.”

Crusoe lets you bring your own model

The streamlined infrastructure has given Crusoe an advantage in aiding some of the biggest pain points for enterprises and startups today: Cost and speed. 

The company’s AI-optimized infrastructure allows its customers to run model inference with fast time-to-first-token, low latency, limitless throughput, and resilient scaling, while also handling the hassle of maintenance and security on customers’ behalf. To put it simply, the only thing users need to do is build. 

It’s a service that Menard calls “managed intelligence,” he said. Whether it’s an off-the-shelf model, an open-source system, a fine-tuned model or something that a customer has built entirely themselves, Crusoe aims to handle it. 

The goal, said Menard, is to be the go-to place that developers trust to build and run their agents, even though it doesn’t offer agent development frameworks itself. “We're really focusing on enabling these agents today,” said Menard.

This strategy allows Crusoe to stay open to any potential customer, something that’s vital as the model market becomes more fragmented. Whether a customer comes with a proprietary model or an open-source one, customers are broadly looking for solutions to the same problems: cost savings and ROI. 

“We want to operate in a neutral way,” said Menard. “We let you choose, as a developer – choose your model, choose your development framework – and we'll simplify infrastructure for you, and we'll give you speed and cost efficiency.”

Start building