Newly formed Quake AI is a cloud provider that expects its users will be mostly agents rather than humans.
It was created after US-based Rumble Inc acquired German datacenter operator Northern Data in June. The parent then rebranded as RUM Group, a holding company for Quake AI (formerly Northern Data) and the Rumble video platform.
Quake AI is an Nvidia Cloud Partner (NCP) operating a 22,000 GPUs estate, making it a neocloud: a specialized infrastructure provider dedicated to AI workloads.
Yet where other neoclouds are building a full stack of services around their GPU farms, Quake AI wants to make its infrastructure discoverable to AI agents, which may soon be traversing the internet in hordes on behalf of their owners.
RUM Group's chief financial officer Mike Massey told The Register that "Cloud 1.0 - let's call it applications - is like a user interface of how you interact with a machine of some kind."
He added: "Now, if you think about agents, they don't use vision. That's not how they interact with systems. They interact with APIs and with information effectively, and that means they want highly discoverable, highly data-driven platforms..
“So, to us, the things that make GPUs and AI compute easier to consume for humans, things like model-as-a-service, dashboards and one-click settings, are actually orthogonal to, if not detrimental to what agents are going to want, agents are going to want homogeneous compute… to understand and know what your data and your expectations are in terms of performance and uptimes, and then they're going to want to consume that compute as a utility,” he claimed.
In other words, Quake AI believes that, at some point, the customers for its compute services are going to be AI agents rather than people, but carrying out tasks on their behalf.
“I think that is the case, and it's the vision that we are driving towards, and you know ultimately agents are in service of people, right?” Massey said.
In its view of things, agents won't go through an intermediary, they'll go to the compute to retrieve the answers they need. "So all of these tool frameworks and all of these things that we have set up, I think, are a hindrance or an encumbrance to what the agents are going to want to consume, which is a utility," he explained.
"Yes, we're a little crazy in this regard," Massey added. "It's different. But I think our worldview is very grounded in the data and the realities. Like, we're a technical team. We understand the engineering, and at the end of the day, I think that most would validate our hypothesis," he continued.
John Hoffman, CEO of Northern Data and acting CEO of Quake AI, compared this with a solution he had been shown of an AI that was developed to navigate the internet by itself.
"I told it I wanted pizza. It went on to DoorDash's website. Typed pizza in the search box, clicked on the pizza with the highest ranking, went to add to order, knows I like pepperoni, and it clicked through all of this," he said.
"Don't get me wrong, this was a good solution that solves the problem. But the amount of compute that you are using to try to get an AI agent to act like a human that has eyes that can see web links is insanity compared to if DoorDash just created an API that is queryable," Hoffman claimed.
Done that way, the compute usage would be one thousandth of the way that the problem was actually accomplished, Quake AI claims, but it will require that everybody embraces the idea of how agents interact.
"And they don't interact with interfaces or applications. They interact with data and APIs. Just give me the raw data. That's what it wants," Hoffman told us.
"I would imagine DoorDash is developing an API for agents," he added. "It just makes sense, and that’s why we're doing the same exact thing."
Earlier this year, TollBit, an outfit that tracks AI bot traffic, said that agents were gradually taking over the internet. Its State of the Bots report found that by the end of 2025, there was roughly one AI bot visit for every 31 human visits to a site, up from just one for every 200 human visits at the start of the year.
Last year, consultants McKinsey & Company produced a report saying that the neocloud operators had to evolve beyond being just a provider of GPUs-as-a-service and move up the AI stack to sell services on top of bare metal, or risk being commoditized into competing on price.
Analyst firm Gartner was also reading the last rites for much of the neocloud industry recently, forecasting that AI services are unlikely to deliver enough revenue to cover the expected capital expenditure on IT infrastructure between now and 2030, and that consolidation in the provider market was inevitable.
Quake AI's gamble is that AI agents will be a reality, and that by tailoring its APIs to make it easier for these autonomous software bots to access its resources as needed, it will be one of the winners. ®
