A Market for Intelligence: Why We Co-Led Liquid Compute’s $15M Seed

The buildout of compute is shaping up to be one of the largest infrastructure investments in history. Hyperscalers alone are committing well over half a trillion dollars in capex this year. Yet much of the capacity created by that investment is still bought and sold like this:

This isn't an outlier. Significant volumes of compute trade through informal, opaque channels. Compute has plenty of prices, but no real reference price. Ask five vendors for an H100 GPU-hour today and you'll get five different quotes. The GPU may be the same, but its value changes dramatically based on location, network fabric, cluster size, contract length, and counterparty credit.

Oil took over a century to scale and build standard derivative markets around physical supply. Compute reached that scale in under four years (though to anyone procuring clusters, it probably felt like a century). Part of the chaos stems from sheer speed, but most of it comes from the nature of the asset itself.

People love comparing compute to crude oil. But when we first sat down with Ronit, Liquid Compute's co-founder and CEO, he made a far more compelling case for electricity.

You can store oil in a tank. An unused GPU hour expires. Its value depends on where the machine sits, how it is interconnected, and what workloads it can run. Compute combines the perishability and regional characteristics of power with something electricity markets don't have to contend with: rapid technological depreciation. A power plant doesn't become obsolete overnight because Nvidia drops a new architecture, but in compute, performance benchmarks shift every 18 months.

That makes standardization unusually difficult.

A unit of compute cannot simply be "an hour of GPU time." A useful contract has to normalize across chip generations, network topology, cluster size, inter-datacenter latency, geography, and other variables that affect what the underlying capacity can actually do. Get that abstraction wrong and the hedge fails to protect you. Get it right, and thousands of bespoke bilateral contracts can begin to collapse into a standardized financial market.

The need for that market is becoming increasingly visible.

Beneath the physical compute market sits an interconnected web of financial obligations. Neoclouds take on long-duration financing to acquire GPUs, then sell that capacity through customer contracts that may expire well before the underlying debt matures. Labs think in terms of procurement. Lenders underwrite contracted capacity as collateral. Datacenters and power providers require their own multi-year commitments underneath it all.

Across the stack, the same risks keep appearing: duration, residual value, utilization, basis, and counterparty exposure.

While researching the space, a CFO at a leading neocloud told us that hedging can feel unnecessary while demand vastly outstrips supply. That changes as the market normalizes and debt begins to mature against previous-generation hardware. Scarcity can mask bad pricing, mismatched duration, and unhedged exposure because nearly everything sells. A more mature market makes those risks harder to ignore.

The physical market for compute has developed extraordinarily quickly. We believe the financial infrastructure around it will need to develop quickly as well.

That is why we co-led Liquid Compute's $15 million seed round.

Liquid Compute is building CFTC-regulated market infrastructure designed to standardize these risks and make them tradable.

Compute will continue to be bought and sold through many venues: directly from providers, through marketplaces, through brokers, and through models that have yet to emerge. There are excellent companies being built across all of them. Liquid Compute does not need to replace those markets; it can become the financial layer connecting them: a common language for price, risk, and settlement across an increasingly fragmented physical market.

The more places compute trades, the more valuable that shared layer becomes.

Legacy exchanges like CME know how to scale existing, well-defined asset classes, but Liquid Compute is building the foundation to discover the compute benchmark itself. Liquid is tackling the hardest structural challenges head-on: standardizing trade units, establishing reliable reference prices for swaps, and linking financial contracts directly back to physical hardware. Liquid Compute is the only player in the compute space following in the footsteps of full-stack startup exchanges and clearinghouses who challenged the incumbents and won, just as ICE did for energy, CBOE did for VIX, and Kalshi did for event contracts.

That is the opportunity Liquid is pursuing. In other markets, purpose-built exchanges have helped establish new financial primitives, from energy contracts to volatility products to event contracts. Compute has the potential to become another major financial market, but its infrastructure has to be built around the unique characteristics of the asset from day one.

Ronit grew up in a family of commodity traders. Markets aren't an analogy he applies to compute, they're the native lens through which he views the world. Where others see simple GPU contracts, he sees tenor, basis risk, residual value, and counterparty exposure. What has impressed us most is his slope: every interaction with the physical market sharpens his conviction of what the financial market must become. Around that vision, Ronit and Aarav have assembled a rare combination of market structure veterans, regulatory experts, and bare-metal systems engineers.

We're thrilled to co-lead Liquid Compute's seed round with our close friends at Chemistry, alongside K8 Capital, Night Capital, TrueBridge, Brainchild Holdings, UFO Holdings, and Dmitry Balyasny.

We're excited to welcome Ronit, Aarav, and the entire Liquid Compute team to the FirstMark family as they build the core financial infrastructure for the intelligence economy.

Read more on WSJ here.