Architect Financial Technologies' Liquid Inference platform lets providers bid for model-serving jobs in real time, with pricing tied to marginal costs

Brett Harrison thinks AI inference should be priced like a trade, not a menu item.

The CEO of Architect Financial Technologies is making the case for an auction-based marketplace for AI inference. His argument is that letting providers set prices based on their marginal costs could push the cost of running AI models down.

Architect launched a platform built on that idea, called Liquid Inference , on or around October 7, 2026.

Inference is the part of AI most users actually touch. It is the moment a trained model takes your prompt and produces an answer, and every one of those answers costs compute.

Liquid Inference turns that compute into something providers compete for. Inference providers bid in real time for model-serving jobs. Buyers then choose the offer that best fits their priorities.

Those priorities are not limited to price. Buyers can weigh cost, latency, and how their data gets handled, so a team that cares about speed can choose differently from a team that cares about the bill.

The economic logic rests on marginal cost. That is the cost of serving one more request. A provider with idle GPUs at 3 a.m. pays very little to take on another job, so it can afford to quote low.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

Harrison argues the format can meaningfully cut inference costs. His case is that competitive bidding improves price discovery, much like it does on a traditional exchange.

As of mid-October 2026, the platform had 28 verified providers serving over 1,000 models. Quotes update in real time.

Architect has also tried to keep the barrier to entry low. Providers can reportedly onboard and integrate their models in an average of about 3 hours.

The catalog covers hundreds of models, both open-weight and closed-weight. The platform also supports multi-modal models, meaning ones that handle inputs beyond plain text.

On the buyer side, new users get a starting balance of inference credits. There is also a referral program. Rewards are linked to transaction fees and paid back as credits, which users can spend on free inference.

Liquid Inference supports the OpenAI and Anthropic APIs. Developers who already build on those interfaces should not need to rewrite their apps to try a new supplier.

Architect did not start from zero in AI. The company already offers financial instruments tied to compute, including compute forwards and options.

A forward locks in a price today for something delivered later. An option gives you the right, but not the obligation, to make that trade. Applied to compute, these tools let buyers and sellers hedge against swings in what processing power costs. Liquid Inference adds a spot-style marketplace next to those longer-dated instruments.

The company has previously raised a $5 million seed round. Liquid Inference expands its lineup of AI-related financial products.