- Crusoe is moving beyond offering raw GPU power to abstract more AI infrastructure decisions
- It recently launched Managed Inference to help developers sidestep infrastructure questions and focus on deployment
- Both Crusoe and Vultr have indicated neoclouds are exploring ways to make AI deployments easier for customers
Neocloud player Crusoe is perhaps best known as a physical AI infrastructure provider – after all, the company is helping OpenAI and Oracle build the flagship Stargate data center campus in Abilene, Texas. But Crusoe is also building a cloud platform designed to make life easy for AI infrastructure engineers and developers.
Up until four months ago, Crusoe’s bread and butter was providing access to GPU clusters. That was great for AI infrastructure engineers who wanted to hand pick best of breed silicon for their AI workloads, and it helped Crusoe gain plenty of share and revenue in the cloud market.
But in November, Crusoe started its climb up the infrastructure-as-a-service stack with the launch of Managed Inference. In a nutshell, Managed Inference lets developers outsource infrastructure decisions to Crusoe so they can just focus on deploying their workloads and models.
“If you don’t have infinite engineering resources and can rely on us to do inferencing with great performance with low cost, that allows you as an agent building organization to focus on your agent more,” Erwan Menard, SVP Product Management for Crusoe, told Fierce.
Menard said Managed Inference was just the start, hinting more managed infrastructure and intelligence offerings will be announced in mid-March around Nvidia's GTC conference. Think offerings that will deliver more granular infrastructure control to developers and engineers so they can better manage model lifecycles, and services that let Crusoe serve models to customers as SLAs.
“We really want to be the cloud where easiest to build and run AI agents,” he said. “If what they want is infrastructure, we’ll make clusters that are self-healing, that correct problems by themselves. If they’re higher up in the stack and they want to consume a model as an API, not even being involved in selecting a GPU, we will give them managed inference and more services.”
Crusoe isn’t the only neocloud to highlight the importance of making AI deployments easier for enterprises and other cloud customers.
“I think the world needs to move beyond just like ‘hey, we're renting GPUs,’” Vultr CMO Kevin Cochrane told Fierce on a recent episode of The Five Nine. “We need to think about like how are we actually getting work done and making people more productive and delivering real outcomes.”
Menard noted Crusoe’s new services address a key hurdle to delivering outcomes: cost. The idea is that rather than treating every AI nail with a huge compute hammer, offerings like Managed Inference can make it easier to right-size resources around AI deployments.
“If you have an agent that’s getting good reception by the business, you are going to get concerned about your inference cost pretty quickly,” he said. “All that nuance around optimizing cost-performance and inference is front and center right now. I think the next type of topics will be around agent memory and the agent preserving context.”
According to Gartner, total spending on AI is expected to jump from $1.76 trillion in 2025 to $3.34 trillion in 2027. AI Infrastructure will account for $1.75 trillion of the 2027 tally, with AI Services contributing $761 billion, AI Software $636 billion and AI Application Development Platforms another $10.9 billion.