Mount Thor Raises $45 Million to Put Apple Hardware in AI Data Centers

Mount Thor has raised $45 million in equity funding to run Apple hardware as cloud infrastructure for AI research and deployment. The San Francisco startup is chasing a specific and fast-growing need: computer-use agents require complete macOS environments where they can open applications, learn, and carry out work over long stretches of time.

The bet matters because agentic AI is shifting infrastructure demand. Agents don't just need a model endpoint — they need a working operating system. If autonomous software becomes routine, someone has to supply, isolate, and manage millions of machine environments. Mount Thor wants that someone to be it, with Apple's platform as the product.

Where the $45 million came from

The financing spans two rounds, Seed and Series A. CRV and Conviction co-led the Series A, while Chad Byers and Offline Ventures co-led the Seed. CSC Leasing, Enea Capital, Long Journey Ventures, and Upper90 also participated. So did strategic angels affiliated with Cognition, Core Automation, Daytona, Meta, OpenAI, Together AI, and Vercel — a roster that reads like a map of the agent and developer-tooling ecosystem.

Mount Thor separately raised non-dilutive capital to fund hardware purchases. It hasn't disclosed how the $45 million equity total splits between the two rounds, or the size of that additional financing.

Why agents need a full operating system

An agent that navigates software needs somewhere to open applications, inspect their state, manipulate files, and resume a task when the next instruction arrives. For workloads built around Apple applications or Apple development tools, that environment has to be macOS. Linux containers and API endpoints don't cover this case.

Mount Thor's customer list reflects three distinct uses:

• Research labs building reinforcement learning environments for computer-use models
• Providers offering macOS sandboxes
• Agent products that need persistent machines

The company also names Apple software development and enterprise inference as target workloads.

Each use brings different operational demands. A training experiment might reset an environment repeatedly to evaluate a model under controlled conditions. A deployed assistant might need its environment to persist for weeks. A development platform might spin up machines in parallel to build and test apps. Managing those lifecycles — provisioning, resetting, sustaining, tearing down — is a substantial part of the infrastructure problem Mount Thor is selling against.

"Making Apple's operating system and silicon available at datacenter-scale is a massive untapped opportunity in AI infrastructure," said Peter Voell, Mount Thor's founder and CEO. Voell points to Apple's long investment in accessibility and human-computer interaction as part of the platform's appeal — for agents and for the people supervising them.

Dedicated Macs or macOS virtual machines?

Mount Thor's compute documentation describes two deployment options. Bare metal gives a customer an entire physical Mac. A virtual machine provides an isolated macOS environment on a Mac reserved for that customer, with two VM slots available per host.

The split changes what you can do. The documented bare-metal account is a permissive non-root user; VM users get administrator access. That makes VMs the documented choice for workloads that need to change system settings. Dedicated hardware fits when performance measurements need a whole machine, or when software simply won't run in a virtual environment.

The VM service is currently in preview. Customers pick an image, create an instance, and connect over SSH or a remote desktop. Connections run through private authenticated tunnels, with guest addresses kept off the public internet.

The practical payoff for agent developers: a Mac becomes a provisioned execution environment rather than a workstation someone sets up by hand. Isolation and access controls still have to match the workload — especially once an agent touches accounts, proprietary files, or external services.

A Kubernetes control layer for Apple compute

Management happens through a portal, the company's mthr command-line tool, and APIs. The Kubernetes interface exposes VMs and bare-metal Macs as custom resources inside a customer's tenant namespace, with actions governed by assigned permissions.

Infrastructure teams can therefore describe a desired machine configuration and manage its lifecycle through orchestration workflows they already know. One caveat worth stating plainly: this doesn't turn macOS into a Linux container. Kubernetes acts purely as the management layer for allocating Apple compute underneath.

Mount Thor also splits its offering into managed and customer-managed fleets. In the managed version, Mount Thor prepares, monitors, and wipes machines. Customer-managed deployments let organizations run their own Macs at a Mount Thor site, under their own management tools and security model.

The inference angle on Apple silicon

Beyond environments where agents act, Mount Thor wants to supply hardware where models run. The company plans to build an inference lab focused on open-source models optimized for Apple silicon, giving enterprises another serving option.

That's a different proposition from renting a remote desktop. Inference infrastructure has to balance model capacity, response latency, throughput, and operating cost. Apple silicon's integrated architecture makes it an alternative worth evaluating, but the announcement offers no comparative serving benchmarks or prices that would establish a universal advantage over other accelerators.

For buyers, the sensible move is to match infrastructure to an actual workload — a macOS-dependent agent, an Apple software pipeline, or a model-serving deployment. Strong results in one area don't guarantee the best economics in the others.

What the money buys next

Mount Thor says it deployed its first data center in under six months. Its team includes people who previously built infrastructure at Cerebras, CoreWeave, Fluidstack, Nebius, OpenAI, and xAI — a hiring pedigree aimed squarely at hyperscale-style operations. The new financing will fund capacity expansion and hiring.

The company's website lists colocation relationships in major U.S. metropolitan areas, with Europe and Asia planned. That draws a clear line between the current footprint and the expansion roadmap.

The larger wager is that autonomous software will create demand for operating environments, not just raw model compute. Mount Thor is positioning Apple infrastructure as a service developers can allocate and manage at scale — without building and maintaining the physical fleet themselves.