Ren Valley
Every market, on your silicon.
We think the compute economy is too inefficient. Compute is underleveraged across its potentialities and thus stranded in its holder’s niche; for research, hand-writing and tuning kernels can create hardware-lottery effects that limit progress. When both the hardware and model layers shift rapidly, software becomes stale by the cycle. Your leverage should not be owned by the ever-changing market; it should adapt to your ideal workload, or arbitrage over the market’s demand for compute and all of its derivatives.
If we mapped the space of efficiency for all (architecture, hardware) tuples, the frontier would currently be extremely spiky, and not very far. We are building the software to both balance and push this progress.
Concretely, we have three goals:
- Expand the frontier of hardware usage for more efficient model serving, and ease bottlenecks for AI research
- Allow anyone to use compute to its fullest capacity, both for individual purposes and for the market
- Build an open platform on which compute exchanges can occur over all of its derivatives, decentralizing progress and power in AI
We’re starting with clouds and labs. If you are interested in engineering a more intelligent world, write to us.