算力緊張？多家初創公司嘗試將 AI 算力金融化

Can AI compute be traded like a commodity?

A handful of startups, including a buzzy company called Ornn, are betting big on the concept as they work to create standardized pricing for GPU usage and connect buyers and sellers with commodities-style futures markets and brokerage services.

Ornn, which helps route GPU capacity from neoclouds and data centers to startups that are willing to pay for short-term contracts, is now in the midst of a new triple-tranche funding round with a top valuation of around $1 billion, with Thrive Capital and General Catalyst in talks to participate, per sources familiar with the matter. Axios had earlier reported the round was in talks. The company raised a $33 million seed round led by a16z crypto back in June.

Another player, Silicon Data, has launched an hourly rental price index for GPUs based on aggregate data from around 100 different sources. It aims to be a pure data play — it doesn’t get involved in brokering access — and raised a $30.5 million Series A in August from Valor Equity Partners and Atreides Management’s joint AI fund.

Former FTX US president Brett Harrison is adding compute to his derivatives exchange, called Architect, and raised $35 million in funding last December. Yet another, Liquid Compute, raised a $15 million seed round in September from FirstMark and Chemistry and has applied to launch a US exchange for compute futures and options.

Despite all the action, the sector faces plenty of challenges. Selling compute futures requires sign-off from financial regulators, which none of these companies have yet attained. CME Group plans to list futures tied to Silicon Data’s H100 and B200 GPU rental indexes on the New York Mercantile Exchange, and ICE plans to list contracts referencing Ornn’s Compute Price index, but they have yet to receive regulatory approval. CME had targeted October 5 as its start date, but that’s since been pushed back.

Some of the newer players want to challenge the existing exchanges and become regulated venues themselves.

The CFTC issued a public request for comment on compute futures that could interrogate the biggest question of all: whether compute is standardized, interchangeable, and liquid enough to support a futures market. Unlike, say, metals, GPUs depreciate rapidly as new chips hit the market, making financialization of them even trickier.

Even if compute is fungible enough to be traded like a commodity and regulatory approvals are forthcoming, it’s not clear whether the shortage of compute that’s creating demand for these kinds of services is a permanent state of affairs or simply a transient phase. Two investors told Newcomer they had passed on deals in the space due to those concerns,

Startup Scramble

AI mania has left many startups in the tricky position of trying to compete with industry giants for access to the compute they need. That’s what’s driving the brokerage side of the emerging compute marketplaces.