Mount Thor Manifesto

Mount Thor believes that Apple’s foundational technologies are optimal for the era of agentic AI.

Apple’s devices run on chips designed by one of the best semiconductor teams in the world. Industry-leading performance and foundational experiences - from Face ID to Apple Intelligence and more - all extend from innovative silicon, deployed on-device and in Apple’s Private Cloud Compute (PCC). Commitment to energy-efficient architectures - including co-founding Arm in 1990 and releasing the Apple Neural Engine in 2017 - enables best-in-class compute per watt from increasingly powerful chips, at a time when power is the bottleneck to scaling inference. Apple’s obsession with the craft of human-computer interaction has created operating systems with powerful primitives for fast, accurate, agentic workflows. The relentless focus on silicon performance and human-machine interactions has primed Apple’s product ecosystem for much broader adoption at a much larger scale.

Agentic AI is transforming the scale and shape of demand for Apple’s operating systems, particularly macOS. Research labs require massive amounts of data to train models that use computers well. Human computer-use involves complex interactions across many applications, and that complexity is increasing, as coding models enable everyone to create software personalized to their needs and preferences. This more than offsets algorithmic improvements in data efficiency. The only way to generate this data is to run many parallel instances of macOS, which requires many Apple devices operating in parallel, at a scale never before seen or needed. Deploying agentic products to billions of people - for enterprise and consumer applications - requires even greater scale. People want their agent to understand their full Apple identity, use Apple tools like iMessage, and be available anywhere, anytime. Some days people need one agent; other days they need many. The future of AI is not only local compute on iPhone or only a country of geniuses in a data center; it’s both. It’s also elastic OS instances, managed for and linked to humans.

The future of enterprise inference is optimized for Apple silicon. Companies will run many models, most of them small, serving a team or individual, and trained on proprietary data, workflows, preferences, and priorities. This model zoo - with low concurrency for any one model - will only partially utilize the FLOPs available on chips from NVIDIA, AMD, and others. Most tasks will remain bottlenecked on human verification, limiting the returns on fast token generation. Financial priorities will drive volume to the lowest-cost provider. Apple silicon provides significant memory capacity in a highly power-efficient footprint, accelerates CPU / GPU coordination with a unified memory architecture, and is rapidly increasing its compute power. While other companies compete to serve frontier models, Apple silicon is optimal for workloads that represent the majority of inference demand over time: non-frontier tasks that increase human productivity at home and work. Apple silicon is the largest platform - and the only custom silicon among the Mag 7 - not yet deployed at rack scale for AI.

Mount Thor builds infrastructure that transforms Apple hardware from a discrete device to an elastic consumable compute resource. Reliability at cloud scale demands excellence in fleet management, networking, scheduling, virtualization, and enterprise GTM. Cloud hardware and software ensure redundancy, reliability, and remote management. Customers expect configurability, stability, and availability with very low tolerance for service disruption. Anyone can put a few computers on shelves and expose IP addresses; few can maintain performant, secure infrastructure at scale, particularly for AI workloads. Even for companies that have sufficient resources to build cloud infrastructure internally, the opportunity cost of human and financial capital often outweighs the benefits. This is true for most companies that need access to fleets of Apple hardware. macOS and Apple silicon will remain underutilized assets until they are accessible both on desks and in data centers. This is the problem Mount Thor solves.

As the cloud infrastructure provider for Apple hardware, Mount Thor eliminates undifferentiated heavy lifting for builders and expands what can be built on macOS and Apple silicon - from virtualization to networking to inference engines. We maintain the largest fleets of the most advanced hardware. We innovate at every layer - from kernels to applications - to materialize the performance and capabilities of macOS and Apple silicon at scale. We strongly support and contribute to open-source projects around Apple’s ecosystem. Computer-use RL, managed agent environments, software testing and validation, secure enterprise inference, and much more all run best on Mount Thor.

Apple is led by hardware visionaries who want to demonstrate what their technology can do without running wide as a company. They have a world-class supply chain capable of supporting new growth vectors and the intellectual honesty to recognize and support new use cases for their hardware. The world has a few too many clouds these days, but it still needs one for Apple hardware. Come help us build it.