Thesis

Inference is becoming the largest commodity in global compute. The markets around it are primitive — no shared spot index, no time gradient, no forward curve. Soma exists to build those markets.

This is the long arc. It will change as we learn, but the direction won’t.

1. A shared spot index

Every market needs a reference price. Inference doesn’t have one yet.

The first step is federated spot: a neutral, continuously updated price for inference, assembled from many independent providers so that no single participant controls it. A trustworthy spot index is the foundation everything else is built on.

2. The time gradient

Spot is the price now. The next step is pricing inference across time — the difference between capacity today and capacity next week.

The time gradient turns inference from a series of disconnected spot trades into something with structure: a way to express that compute next month is worth more, or less, than compute today.

3. The forward curve

With spot and a time gradient in place, a forward curve becomes possible — prices for inference delivered weeks or months ahead.

A forward curve lets buyers lock in cost and lets providers commit capacity with confidence. This is what turns inference into a market that can be planned around, not just bought from.

4. An open market

Each step is built in the open, with rules that anyone can read and verify.

Markets work best when their mechanisms are legible. Soma’s are public by design, not a black box — so the people using the market can trust it.

Principles

• Neutral. No single participant sets the price.
• Legible. The rules are public and verifiable.
• Composable. Each layer is the foundation for the next.

We’ll write up the details as we build them in research.