AI is centralised. A handful of companies control the models, the compute, and — by extension — the world’s data. Unlike past technologies, AI is trained on the sum of human knowledge and decisions. Its centralisation is more consequential than anything we’ve seen before — reducing sovereignty, eroding privacy, and limiting accessibility.
At EXO, we’re building the alternative. We make it possible to run AI locally — from a single MacBook to clusters of thousands — giving people and organisations back sovereignty, privacy, and control.
We are building an AI stack from the ground up, optimised for consumer hardware — from the models to the distributed algorithms, scheduling, and orchestration.
On consumer hardware, we face higher memory‑to‑FLOPS ratios, unreliable connections as devices go offline and move networks, and higher network bandwidth‑to‑FLOPS ratios than datacenter GPUs.
As a Founding Researcher, you will drive research that unlocks performance on Apple Silicon: bs=1 attention paths and kernels, MoE routing for unified memory, quantization strategies, topology‑aware scheduling, and KV transport — turning promising ideas into prototypes that deliver measurable product wins.
We’re in‑person in London. The culture is intense, collaborative and there is constant learning.
You don’t need to know all of these upfront. The ability to learn quickly and reason from first principles matters more.