Liquid Compute

This update is the first part of an ongoing series on our GPU Offtaker Credit Map. Thanks Stanley Lee for your thoughts. Look in the comments for the rest of our published research.

Announcing the Liquid Compute GPU Offtaker Credit Map We assessed 200 private buyers of compute and ranked 169 on a single, transparent credit framework. Our findings cut against consensus: the strongest private offtakers are trading firms and profitable private corporates, not necessarily best-known AI labs. We released this because the real bottleneck to access compute is capital markets. Capacity gets built when someone will finance it, and financing turns on a question nobody had a shared answer to: can this buyer keep paying for the full term? The framework is published in full, and we intend it to become the stepping stone for providing greater transparency about counterparty credit. If you buy, build or finance compute, we want to hear from you. If you are on the list and think we have you wrong, please reach out. We welcome all comments and direct feedback on the methodology.

Oracle has reportedly sent a force majeure notice to the developer of Project Jupiter, its 2.45GW Stargate campus in New Mexico. It is seeking the right to defer payments if the site misses its 2028 date. Oracle says the project is on schedule and Blue Owl Capital says commitments are unchanged. The market is trading like it matters anyway. Three read-throughs for compute markets: 𝐒𝐮𝐩𝐩𝐥𝐲 𝐢𝐬 𝐭𝐡𝐞 𝐬𝐭𝐨𝐫𝐲 𝐧𝐨𝐰. Power, permits and local opposition are delaying capacity, and data centers are becoming a campaign issue heading into the midterms. Every delayed gigawatt pushes demand onto capacity that already exists. 𝐆𝐏𝐔𝐬 𝐢𝐧 𝐭𝐡𝐞 𝐠𝐫𝐨𝐮𝐧𝐝 𝐠𝐞𝐭 𝐦𝐨𝐫𝐞 𝐯𝐚𝐥𝐮𝐚𝐛𝐥𝐞. That supports near-dated pricing and residual values for operators with live capacity, and widens the gap between built and unbuilt. 𝐂𝐨𝐧𝐭𝐫𝐚𝐜𝐭𝐬 𝐠𝐞𝐭 𝐫𝐞-𝐮𝐧𝐝𝐞𝐫𝐰𝐫𝐢𝐭𝐭𝐞𝐧. A lease or take-or-pay is only as firm as its delay, force majeure and termination language. Lenders will read those clauses first now. Here's the problem. When a contract like this wobbles, nobody can answer the basic questions: what is it worth today, what is the capacity worth if it gets re-let, and how do you hedge the exposure. There is no standard contract, no independent reference price, and no way to transfer the risk. That is the gap Liquid Compute is built to close. We have three pieces: - A price index for GPU compute - Independent marks on contracts and collateral for lenders - Cash-settled instruments that let operators, buyers and lenders hedge against that index Credit markets solved this problem decades ago and compute is next. https://lnkd.in/df7NNjd6

H200 rental rates have climbed back to where they stood at launch, even as Blackwell-generation B200s and B300s come online at scale. This is a clear signal that demand for compute is outpacing supply across every generation, and that prior-generation hardware is holding its value far better than most depreciation models assumed. For anyone buying, selling, or financing GPUs, it's a reminder that compute pricing is less about the chip on the spec sheet and more about who has capacity available when it's needed. https://lnkd.in/dh5QM2Wx

We are pleased to welcome Stanley Lee as our new Chief Product Officer on the Liquid Compute team! He will lead product across the business as we build the venue to buy, sell, trade, and finance compute. Please join us in extending a warm welcome as we take this important step forward together. Stanley brings a career in credit markets to the compute space, most recently as a high yield and distressed trader at Point72. Before that, he was an OTC market maker at Barclays, trading high yield bonds and CDS, and he helped build out the high yield/distressed effort at Laurion Capital Management LP. His background spans distressed and high yield credit, OTC market structure, and derivatives and structured products. He was an early participant in neocloud credit and GPU-backed loans. That work shaped the thesis behind his move: compute today looks like early-day OTC credit, bilateral and opaque, with no standard contract, no reference price, and no way to transfer risk. The sequence that fixed credit markets is the same one that will build this one, starting with standardized documentation and a published price. As CPO, Stanley owns product across all four of our businesses: the brokerage where physical capacity is bought and sold, the LCI index series and forward curve, the derivatives that reference them, and the credit products that finance compute contracts and fleets. He will drive how those pieces fit together into a single venue, from the reference spec and normalization standards that make a GPU-hour comparable, to the contracts, benchmarks, and risk transfer tools that let operators, buyers, and lenders transact with confidence. Welcome aboard!

AI’s next phase will require more than compute capacity. It will require the financial infrastructure to price it, finance it, and manage the risk around it. We’re proud to back Liquid Compute as it builds the capital markets layer for AI compute. “Compute is becoming a strategic commodity, and AI’s continued growth will depend on making its market more transparent, efficient, and financeable. Liquid Compute is building the connective layer between developers, infrastructure providers, and capital that can help the entire ecosystem scale.” - Andre Koo, Founder and GP at K8. The Wall Street Journal article below.

Special thanks go to Luis Garcia for his coverage in the The Wall Street Journal