- Vultr says agentic AI is driving fresh CPU demand
- The company is pitching itself as an enterprise AI infrastructure platform, adding identity, access, storage, provisioning and VPN features
- Vultr’s new Chennai AI region underscores its agentic AI bet
AI infrastructure conversations may still revolve around GPUs, but Vultr CMO Kevin Cochrane told Fierce the rise of agentic AI is quietly pulling CPUs back into the spotlight.
“You used to have like eight GPUs and then you’d have one CPU to do orchestration of the GPUs. That whole model has shifted and we’re seeing closer to a one-to-one ratio,” he said.
While CPUs have and will continue to run databases and data pipelines, they are also now key in allowing agents to call various tools they need to complete tasks. “So, at the end of the day, when you’re doing a multi-agent deployment, the bulk of the work is actually happening on the CPU.”
Cochrane’s statements match with similar comments about the importance of CPUs made to Fierce by Google Cloud’s Mark Lohmeyer, as well as executives from Red Hat and, of course, chipmaker AMD.
The discussion came as Vultr launched a new data center region in Chennai, India, its 34th overall and second-largest GPU cluster worldwide with 10,000 AMD Instinct GPUs.
Cochrane said the region – which currently comprises a single data center – will be live and producing tokens in December, adding 100% of the facility’s capacity is already sold out and will begin generating revenue on its first day of operation.
He declined to share any specifics about its customers at the facility.
Asked why it decided to go with AMD chips at the facility, Cochrane told Fierce that Vultr is seeing “an explosion of interest in AMD tied to agentic AI.” That interest is a combination of demand for AMD CPUs to run agent sandboxes and GPUs to provide the reasoning capabilities.
Securing its future with the enterprise
But Vultr hasn’t just been busy hooking up a new data center. This week it also unveiled a new suite of platform capabilities aimed squarely at serving enterprises. These include a range of identity and access management tools, an upgraded cluster provisioning interface, bootable storage options and cloud VPNs.
The release was fortunately timed, coming just after Gartner warned up to 80% of neoclouds (like Vultr, kind of) will likely cease operations by 2030 as pressure increases to turn a profit on pricey infrastructure investments.
Cochrane argued the release makes clear that Vultr is – and has been – more than just a bare metal compute provider. It’s an enterprise partner. And that position couldn’t be a better one to be in as the market turns.
“If you look at the real meat of the release, some of the stuff you’re saying ‘that’s really boring, mundane stuff. Why is that innovation?’ And the answer is it’s innovation in our space,” he explained. “Would we have gotten requests for SCIM support three years ago? The answer is no. But nowadays, that’s front and center.”
SCIM refers to System for Cross-domain Identity Management and is a key pillar of enterprise IT.
Cochrane continued: “The buying requirements have shifted. Early days, three years ago it was just offtake contracts, just give some bare metal to the hyperscalers, they’ll take care of everything else. Enterprise buyers don’t want just raw bare metal. They have to take care of everything themselves and they actually want a real, actual platform.”
To be fair, Cochrane has been making this argument for some time. In an episode of Fierce Network’s The Five Nine from January, Cochrane insisted that platform capabilities are critical for enterprise customers.
Enterprise AI’s home in the Medium Place
If hyperscale platforms are what you might call The Good Place and bare metal is rapidly becoming The Bad Place, Vultr, it seems, is aiming to make a niche for itself in The Medium Place. (Anyone who’s ever watched The Good Place gets it.)
Moor Insights and Strategy VP and Principal Analyst Matt Kimball noted Vultr is in an interesting position, having built itself into something between a neocloud and a hyperscaler. Its latest platform upgrades are a bet that its middling position is a strength in the agentic era, but “winning it means landing on the shortlists of enterprise IT leaders who, for the most part, don’t know the name,” Kimball wrote.
Kimball argued the same things that made Vultr popular among developers – it’s easy interfaces, predictable pricing, diverse regions and chip selection – make it ideal for those looking to run AI agents. But in addition to issues with brand recognition, Kimball pointed out Vultr is also about to face a wave of competition as neoclouds pivot toward the enterprise in a bid to capture inference workloads. See also: CoreWeave’s Forge, launched last month and Crusoe’s efforts to climb up the infrastructure-as-a-service stack.
“Vultr has real advantages in this race… But it also has some work to do,” Kimball concluded. “I think the first thing I’d start with is awareness…The middle can be used to Vultr’s advantage. But I can also see it being used against Vultr if it doesn’t aggressively lean into establishing itself with enterprise buyers.”
Read more about Vultr and CPUs on Fierce Network:
Google Cloud’s Lohmeyer: Agentic AI will kick off a CPU comeback
HPE lands first AMD Helios order in $1.2B Vultr deal
Vultr: AI sovereignty drives enterprises to 'alt-scaler' clouds