Mount Thor Raises $45 Million to Run Cloud Fleets of Macs for AI Agents

Mount Thor, a San Francisco startup, has raised $45 million across its seed and Series A rounds, with CRV and Conviction co-leading the Series A. The round was announced on November 12, 2026. The company runs managed, elastic cloud fleets of Mac computers and Apple silicon, built for a new kind of customer: AI agents that use computers the way people do.

Mount Thor's infrastructure serves AI computer-use agents, reinforcement learning (RL) environments, Apple app testing, and enterprise inference. In plain terms, it rents out fleets of Macs in the cloud — machines that can be spun up and down on demand — to companies whose software needs to run on Apple hardware rather than the standard Linux servers that dominate cloud computing.

What Are Computer-Use Agents?

A computer-use agent is an AI system that operates a computer by interacting with its interface — clicking buttons, filling in forms, navigating applications — rather than by calling an API. The idea is that much of the world's software has no API, or has an incomplete one, so an agent that can use a graphical interface can automate work that API-based agents cannot reach.

Training and running these agents requires environments that look like real computers. That means actual desktop operating systems, real applications, and realistic screen states — and for the large share of knowledge work that happens on Macs, it means macOS. A cloud provider that offers elastic fleets of Macs gives agent developers a place to train, test, and deploy their systems at scale, without buying and maintaining rooms full of Apple hardware.

Why Reinforcement Learning Needs Mac Fleets

Reinforcement learning is the training method behind many of the most capable AI agents. An RL system learns by trying actions in an environment and receiving rewards for good outcomes — the same way a person learns a video game by playing it repeatedly. For computer-use agents, the "environment" is a computer, and effective RL training requires thousands of parallel environments running simultaneously so the agent can gather experience quickly.

That is what Mount Thor sells: not just individual Macs in the cloud, but managed, elastic fleets — large numbers of machines that can be provisioned and released as training demand rises and falls. Elasticity matters because RL workloads are bursty. A training run might need hundreds of machines for a few hours and none the next day. Owning that hardware would be wasteful; renting it by the hour from a specialist is the economical choice.

Apple App Testing and Enterprise Inference

Mount Thor's other use cases are more familiar. Apple app testing has long been a pain point for developers: Apple's rules require apps to be built and tested on Apple hardware, which has historically meant maintaining a fleet of Mac minis in an office or a data center. A cloud fleet removes that burden, letting developers test across macOS and iOS environments on demand.

Enterprise inference — running trained AI models to serve real users — is the fourth use case, and it points to a broader trend. Apple silicon, the custom chips in modern Macs, has proven to be efficient at running AI workloads, particularly for models that benefit from the chips' unified memory architecture. As more enterprises look to run inference outside the big GPU clouds, Apple silicon fleets offer an alternative worth pricing.

Why CRV and Conviction Co-Led

CRV is a long-established venture firm with a history of backing infrastructure companies, while Conviction is a newer firm focused on AI. Their co-leadership of the Series A is a useful signal: it pairs generalist infrastructure judgment with AI-native conviction. When two firms with different lenses agree to lead the same round, it usually means the company sits at an intersection both recognize — in this case, cloud infrastructure and the AI agent economy.

The $45 million raised across seed and Series A is a substantial early war chest for an infrastructure startup. Infrastructure is capital-intensive: fleets of Macs must be bought, housed, and maintained before the revenue arrives. A large early raise gives Mount Thor the balance sheet to build capacity ahead of demand, which is exactly what customers choosing a cloud provider want to see — assurance that the machines will be there when they need them.

Why It Matters for Founders and Investors

For founders, Mount Thor illustrates a durable pattern in AI: every new class of AI application creates demand for new infrastructure. Computer-use agents need Mac fleets the way chatbots needed GPUs. The founders who spot the infrastructure gap early — before the application layer matures — can build the picks-and-shovels business that every application company ends up renting from.

For investors, the round is a reminder that AI infrastructure is diversifying beyond the GPU cloud. Apple silicon, agent environments, and RL training infrastructure are all plausible growth markets, and they reward operators who understand the specific hardware and software constraints of their niche rather than generalists who offer undifferentiated compute.

The risk, as with any infrastructure bet, is concentration: Mount Thor's fate is tied to the pace at which computer-use agents and RL training move from research demonstrations to production deployments. If that transition is slower than expected, the fleets will be underused. The $45 million raise suggests CRV and Conviction believe the transition is already underway — and that the companies building the agents will need somewhere to run them.

Angeled.com connects founders with investors.