Neoclouds. Neoscalers. GPU clouds. Whatever you call them, they’re having a bit of a rough time of late. Microsoft’s Nscale about-face, Firmus’s IPO cancellation, and Elon Musk’s “guppy-level death” remark regarding the neocloud market paint a less-than-stellar picture of the industry upstarts.

But for all the chatter, neoclouds are busier than ever, fully focused on protecting margins in a bid to move up the value chain. But to realize that dream, neoclouds face having to push beyond catering to one or two major tenants. It’s why everyone is talking about multitenancy of late.

Underpinning that pivot for a sizable number of these players is Rafay Systems. Its orchestration and workflow automation platform serves neoclouds like Firmus, Era4, and Sharon AI, acting as the operational control plane between their vast arrays and compute infrastructure and end-user applications. Its token routing layer, for example, lets these players offer AI models from a single endpoint, shifting between them automatically to help keep capacity busy.

CPO Mohan Atreya told SDxCentral that enterprise customers have gotten used to security, governance, and service capabilities from hyperscalers and don’t just want access to servers. To help neoclouds move beyond simply selling infrastructure, Atreya and the team at Rafay argue that neoclouds need to meet that bar if they want to move up the value chain, providing support for multiple tenants and even launching marketplaces for bare-metal servers, Kubernetes, and token services.

Rafay does cater to customers beyond the neoclouds, but Atreya outlined that these emerging GPU players were “an easier sell for us in the sense they move faster.”

“The decision process is very quick because they can't have these expensive GPUs sitting there idling. The enterprise doesn't have that concern, so the enterprise sales cycle is longer,” Atreya explained. “From a pure go-to-market, the speed of deal, we focused on neocloud, but the same solution is equally applicable in the enterprise.”

Atreya even likened some enterprises to being a “small neocloud” in the sense of multiple business units each having their own inherent complexity and respective regulations to comply with.

But despite macro uncertainties hitting actual neoclouds undeniably hard, Atreya said that of Rafay’s customers in this market, they’re running at what he likened to crazy levels of utilization, noting: “Pretty much every GPU has been spoken for.”

“They're also doing short-term contracts out of choice. Why is that? Because they're able to raise prices versus if they do a one-year commit[ment], they are stuck with that price,” Atreya said. “So they're doing three-month commits, just like hyperscalers. That way they can increase prices if and when needed. So I'm not seeing the pressure on them, and this is also good for them from a margins perspective because they can keep fine-tuning that.”

Rafay’s work in the neocloud space sees it work alongside Nvidia at an engineering level. Its relationship with the GPU giant is “so tight,” in Atreya’s own words, that the pair are “co-designing” offerings supplied to neoclouds. Of course, its relationship with Nvidia is just one part of its broader effort to get its software into neocloud deployments early. The company also works with hardware vendors including Dell, Lenovo, Hewlett Packard Enterprise (HPE), Supermicro, and Cisco, with Atreya stating that Rafay is available through partner price books. Such an approach sees it bundling orchestration with infrastructure rather than leaving customers to discover the software need later.

Atreya pointed to Rafay's work with Australian neocloud darling Sharon AI as an example of a player that started small with the orchestration player, before scaling quickly. Another was Canaidan telecom giant Telus, an altogether different customer experience that the Canadian firm moved about 20% of its token production to a Rafay-powered “token factory,” averaging five to six billion tokens an hour.

“Telus themselves are a large enterprise, so customer No. 1 is their internal teams, and this is just shifting money instead of spending money on an external entity; it becomes like an internal spend, but not by degrading your SLA (service level agreement) and quality of service, you still give the quality of service the internal customer would expect from a hyperscaler or an Anthropic,” Atreya explained.

The Telus example points to the broader ambition behind Rafay’s wider work. Instead of just helping neoclouds to rent out GPUs, giving them the actual software and operating practices to run a cloud service that enterprise customers can get all the bells and whistles they’re used to with hyperscalers, and not just getting hardware online and running.

Atreya sees an opening beyond neoclouds as some enterprises are rebuilding their data-center capabilities for AI rather than relying entirely on hyperscalers.

“We have many enterprises in certain regions, like, for example, in Turkey, Saudi Arabia, and Dubai. They've kind of gone away from hyperscalers because everything is under attack there,” Atreya said in reference to the ongoing conflict in the Middle East. But away from war zones, he pointed to “classified U.S. environments,” as well as data sovereignty and other constraints, making it so that running AI infrastructure closer to home is especially important.

Rafay’s pitch, then, is that the same control plane and orchestration offering should serve those organizations. The challenge, though, is as much operational as technical, with Rafay wanting to make deployments standardized to such a degree that a customer can go from hardware to a working platform in roughly an hour instead of weeks.

“The process, the practice, that’s where the IP is right now,” Atreya added.