Moreh

Moreh builds the MoAI Inference Framework — software that runs NVIDIA, AMD, Tenstorrent, and other accelerators, across vendors and generations, as a single AI cluster.

For companies running AI infrastructure, this changes three things.
1. Cost per token falls: no accelerator sits idle, and each chip handles the part of the workload it is built for, so the same fleet serves more tokens.
2. Procurement is freed from lock-in: with one framework across vendors, the next purchase follows price-performance rather than a single roadmap.
3. And growth stops meaning replacement: a new vendor or a new generation joins the cluster node by node, alongside what is already in service.

Under the hood, every accelerator sits behind one endpoint. The framework sends each request to the chip best suited to it, carries KV cache across vendor boundaries, and tunes itself to the latency target you set. New chips attach to the same gateway.