ML infrastructure powers billions of dollars of commerce and is growing rapidly.

‍Today it is pervasively monolithic - either built by a hardware team for a specific collection of devices or is "research quality" – produced by an industrial research lab. Many systems are collections of incompatible technologies unified by marketing and problematic translators.

The next generation of product breakthroughs will be powered by production quality infrastructure that brings together the best of compilers and runtimes, is designed for heterogeneous compute, edge to datacenter distribution, and is focused on usability. Unifying software and hardware with a "just works" approach will save developers enormous time and increase their velocity.

We believe the fundamental research is done – but we need a first-principles re-architecture of our systems. We need a team motivated to solve the "big problem" in a disciplined way and an architecture that deploys to both large and small systems alike.

Modular AI is that team - solving that problem.

‍We have founded & built much of the world's production ML infrastructure. From TensorFlow, TF Lite, XLA, TPUs, Android ML, Apple ML, MLIR, and more. We are world-leading experts in ML, scaling, and distributed computation and have deployed production workloads to billions of users and devices.

‍Modular, composable & layered architecture is what the world of AI needs, and at Modular AI, we are building it for everyone.