Quake AI neocloud thinks agents, not meatbags, will become its main customers

A cloud built for AI

Quake AI is an Nvidia Cloud Partner with an estate of 22,000 GPUs. It is a neocloud, a specialized infrastructure provider focused on AI workloads. Rather than building a full stack of services around its GPU farms, the company aims to make its infrastructure easier for autonomous agents to discover and use. Those agents could increasingly browse the internet and act for their owners.

Agents as compute customers

Mike Massey, chief financial officer of RUM Group, described conventional cloud services as applications that give people an interface for interacting with machines. He said tools such as dashboards and model-as-a-service may not suit agents’ needs.

Massey said agents will seek standardized computing resources, with clear information about data, performance and uptime, consumed as a utility. He also said agents ultimately serve people.

APIs instead of screen clicks

Massey illustrated the potential with an agent ordering pizza through DoorDash: it searched for pizza, chose a highly ranked option, and added pepperoni based on his preference. Quake AI claims that accomplishing the task this way would use one thousandth of the compute required by the alternative.

He said agent APIs would make sense for services such as DoorDash, and that Quake AI is developing APIs on the same principle. The approach depends on broader acceptance of how agents interact with online services.

A bet against commoditization

McKinsey & Company said last year that neocloud providers should move beyond selling GPU access and offer services higher up the AI stack, or risk competing mainly on price. Quake AI is wagering that autonomous agents will become widely used and that APIs tailored to them will help the company succeed.