Overview

Check out our Manifesto for Our Vision of the Future of Inference Financialization.

The Grid is a market for AI inference. It treats inference as a standardized commodity and lets suppliers and buyers meet in a single real-time market. Instead of buying access to a specific model from a specific provider, you buy Units of standardized Instruments. The specs for a given Instrument guarantee a minimum level of quality and reliability. Your purchased units can then be consumed via the standard completions API.

Why inference needs a market

Today, inference capacity is fragmented.

• Each provider publishes its own pricing menus, with opaque backdoor negotiations with enterprises for custom pricing.

• Every model has its own API, quotas, and fine print.

• Capacity is over-provisioned because demand is hard to predict.

• Unused GPU time disappears forever.

Inference cannot be stored. Every idle GPU minute is lost production. This makes inference more perishable than electricity and more like a throughput constrained commodity.

The Grid addresses that by:

• Standardizing what is being sold through Instruments and Units.

• Creating a single market where prices emerge from bids and asks.

• Tying every trade to a near term delivery window.

• Making delivery auditable so developers and suppliers can trust the system.

What you can do with The Grid

As a developer building AI apps, or as researcher generating synthetic data using AI inference, you can:

• Choose an Instrument for your use case (for example - Chat Fast - if you’re building an AI chat assistant) instead of choosing specific models.

• Buy Units at real-time market prices, where multiple Suppliers compete to serve you inference

• Run workloads against your balance, with clear rate limits and predictable performance using the Consumption API (which uses the Chat Completions API standard)

Suppliers will be able to:

• Issue Units of 1M tokens, serving models that conform to the Instrument Benchmarks.

• Sell Units on the exchange and see the market consuming inference within tight windows, instead of vague usage based commitments. Suppliers can use our UI or Trading API to set their own programmatic sell orders, or contract with us to aggregate capacity with other suppliers.

• Compete on pricing power with other suppliers in a real-time market.

Currently, The Grid is serving as a supplier on the platform, and several other suppliers will be onboarded in due course. If youre interested in becoming a supplier, please reach out to us at [email protected]

How the docs are organized

Dive into the following focussed pages to learn more about The Grid:

1. Core Concepts - The mental model for Instruments, Units, and Trading and Consumption Accounts.

2. Instruments and Benchmarks - How our current Instruments are defined, how we measure quality, and how benchmarks evolve.

3. Market Mechanics - How the order book works, how prices are formed, and how trading by consumers results in predictable demand for suppliers.

4. Consumption, Delivery and Metering - How tokens are consumed from the Units you hold in your Consumption Account, how the API queues Lots for consumption and how tokens and metrics are metered.