Liquid Compute Raises $15M for an AI Compute Market

Liquid Compute is trying to build the market infrastructure that AI spending has outgrown. The New York company emerged from stealth on September 15 with a $15M Seed round and a plan to create a regulated exchange for compute.

Chemistry and Firstmark co-led the financing. K8 Capital, Night Capital, TrueBridge, Brainchild Holdings, UFO Holdings, Y Combinator, and angel investor Dmitry Balyasny also participated, according to the company's $15M funding announcement. The important part is not another venue where someone can find GPUs. Liquid Compute wants to create verification, price discovery, clearing, and settlement around an asset whose value can disappear when its delivery window passes.

What Liquid Compute Announced

Liquid Compute is led by CEO and co-founder Ronit Jain and co-founder and CTO Aarav Patel. The company is part of Y Combinator's Winter 2025 batch, and Y Combinator's launch page independently corroborates the round, founders, and investor syndicate.

The company says it will use the capital to build a regulated exchange for cash and physically settled contracts on AI infrastructure, starting with compute. It did not disclose a valuation, board changes, revenue, prior funding, customer count, or a detailed allocation of proceeds.

Liquid Compute has filed applications through PMEX Markets and PMEX Clearing. The Commodity Futures Trading Commission lists the Designated Contract Market application and Derivatives Clearing Organization application as pending. That status matters: the company is pursuing a regulated venue, but it is not yet approved as one.

Why Compute Is Difficult to Trade

A barrel of oil can sit in a tank. A GPU-hour scheduled for Tuesday cannot be stored and delivered on Friday after demand improves. Once the window passes, that capacity is gone.

Compute is also not a uniform product. Its value changes with the accelerator, cluster configuration, location, network, power profile, delivery period, contract length, and the party expected to perform. Two contracts that both say “H100” can expose a buyer to meaningfully different infrastructure and counterparty risk.

That makes a simple listing marketplace incomplete. Buyers need to know what is being delivered, sellers need confidence that contracts will settle, lenders need dependable benchmarks, and both sides need a way to transfer exposure without renegotiating an entire infrastructure agreement.

Liquid Compute's product description centers on that connective tissue. The company describes verification for the underlying capacity, an order book for physical transfer, price discovery, clearing, and settlement. In its own market-structure research, it also says it operates a physical brokerage, publishes a daily reference index governed by an independent third-party benchmark administrator, and arranges financing against fleets and contracts. Those are company-reported operations, not independently audited performance results.

The Exposure Behind the Product

AI builders do not necessarily want to speculate on hardware prices. They want dependable access to computation at a cost their products can support. Capacity owners want utilization and financeable cash flow, not an open-ended bet on future spot demand.

Both sides still inherit price risk. A buyer can commit too early and overpay if supply loosens. Waiting can mean missing the capacity required for a launch or training run. A seller can reserve too much inventory for a customer that does not materialize, or lock in terms before demand strengthens.

Liquid Compute's thesis is that standardized contracts and a clearing layer can make some of that risk observable and transferable. A reference price can help participants value agreements. A verified contract can reduce ambiguity around delivery. Clearing can centralize parts of counterparty management. None of those mechanisms creates additional chips, but each can change how existing capacity is financed and allocated.

This is why the company's electricity comparison is useful, within limits. Electricity markets learned to coordinate a non-storable, location-dependent resource through contracts, benchmarks, scheduling, and settlement. Compute has different physical constraints and a much younger market, but it presents a related coordination problem.

What the Investor Syndicate Is Backing

The Seed round is financing more than software. Liquid Compute must earn trust from infrastructure operators, AI companies, traders, lenders, and regulators at the same time. Each group evaluates a different failure mode.

Chemistry and Firstmark are backing that market-building effort alongside funds and operators with exposure to technology, finance, and AI infrastructure. The company also names Susquehanna, BGC, and Wintermute as collaborators in its announcement. That language shows commercial engagement, but it should not be read as proof of exchange approval, trading volume, or long-term customer commitments.

The regulatory path may become a meaningful differentiator if Liquid Compute can complete it. It may also lengthen the time required to launch the full venue. Pending applications are evidence of intent and process, not an endorsement from the CFTC.

What the Round Has to Prove

Liquid Compute now has to show that diverse compute agreements can be standardized without erasing the technical details that determine value. It must verify delivery, produce benchmarks participants trust, attract enough buyers and sellers for useful price discovery, and manage counterparty risk under an unfinished regulatory process.

The company also has to prove that customers want to trade their exposure rather than keep negotiating private contracts. Large buyers may value bespoke terms. Operators may resist transparency that compresses margins. New financial contracts will be useful only if they connect cleanly to real capacity and real delivery obligations.

That is the significance of the $15M round. AI infrastructure is becoming large enough that procurement alone cannot carry every pricing and delivery problem. Liquid Compute is betting that compute needs a market layer where the risk can move, not just a directory where capacity can be found.