Two announcements on the same day describe the same machine from opposite ends. Boost Run (Nasdaq: BRUN), a GPU cloud provider, signed a five-year AI cloud services contract worth $525.6 million with what it calls a leading sovereign AI company. Liquid Compute, a New York startup building trading infrastructure for compute, set up a $250 million facility with K8 Capital so buyers can finance the deposits operators demand before they deliver capacity. One sells GPU capacity years ahead. The other lends against it.
Boost Run’s Contracts Run Far Ahead of Revenue
The new Boost Run contract covers capacity on NVIDIA GB300 NVL72 systems, plus network storage and CPU nodes. The customer isn’t named. Acceptance of the first infrastructure is expected to start in early Q2 2027, so revenue from it is still months away.
The deal takes Boost Run’s total contract value past $2.6 billion, up from $1.9 billion when it reported second-quarter results in August, its first as a public company. That quarter’s revenue was $31.1 million, up 270% from a year earlier. Annualize it and Boost Run is running at about $124 million a year against $2.6 billion of commitments. The company says its signed contracts carry upfront payments and describes its financing as non-dilutive. That’s the model: the customer’s commitment, and some of its cash, comes before the hardware.
Liquid Compute Finances the Deposit
On the buyer side, capacity has to be paid for before it exists too. Operators usually want a deposit before they deliver, and buyers have mostly paid it out of equity. Liquid Compute’s facility, arranged with K8 Capital, lets qualifying buyers that contract capacity through its platform draw funds as they sign. Each draw is secured by the prepaid capacity and checked by Liquid Compute before money moves. No rates or terms were disclosed, and no buyers or operators were named.
The logic is a commodity trader’s. Ronit Jain, Liquid Compute’s chief executive, argues that once a forward contract for compute is standardized and priced in the open, it can serve as collateral. Transparent pricing lets lenders mark the position, and the ability to re-let capacity gives them a way out if a buyer fails. The company, a Y Combinator alum, is building what it calls a physical grid for compute, underneath a cash-settled futures exchange that is still waiting on CFTC approval. It raised a $15 million seed round three weeks ago, co-led by FirstMark and Chemistry, with K8 Capital among the investors.
The Rest of the Chain
Penguin Solutions filled in the middle the same day. It reported four new neocloud customers in its fourth quarter, including one with $10 billion of contracted compute from a leading AI lab that hired Penguin to build and run a 36,000-GPU site in Norway, and a publicly traded neocloud with more than $3 billion in signed multi-year contracts. Put the pieces together and the chain looks like this: AI labs and sovereign buyers sign long contracts; neoclouds raise money against them; builders like Penguin put up the racks; and now lenders finance the deposits in between.
The risks are the usual ones when contracted value runs this far ahead of revenue. Counterparties go unnamed and delivery dates can slip. Five years is also a long time in GPU generations, and hardware contracted today will be one or two generations old before these deals end.
The financing is ready. The hardware is still on its way.