@Oracle has sent a force majeure notice on Project Jupiter, the 2.45GW Stargate campus in New Mexico, per Bloomberg. Oracle says it's on schedule. Its credit is trading like it isn't.
Read-through for compute:
1. The binding constraint is increasingly power, permits and politics, not chips.
2. Capacity already built and energized gets scarcer. GPUs in the ground are worth more.
3. A take-or-pay is only as firm as its delay and termination clauses. Every one gets re-underwritten.
What's missing: a standard contract, an independent price for compute, and a way to hedge it.
That's exactly what we're building at Liquid Compute.
bloomberg.com/news/articles/…
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.
lnkd.in/dh5QM2Wx
I’m excited to share that today @liquidcompute emerges from stealth with a $15 million seed round. We are announcing pending applications before the CFTC for Designated Contract Market (DCM) and Derivatives Clearing Organization (DCO) status to build the first regulated
Our latest piece at @liquidcompute discusses the term structures that impact how GPU capacity trades today. Please feel free to reach out or leave a comment if you would like to discuss! x.com/i/article/2094…
We spent the last few months talking to credit investors about financing GPU fleets. Four questions came up in nearly every conversation:
1. What does the market actually price?
2. Is there a secondary market for the hardware?
3. How do you tell good offtake from bad?
4. What does the depreciation curve look like?
No document answered them, so we wrote one.
Investing in the Middle Market for Compute covers the segment below the hyperscale tier, where the borrower is a real operating business and the offtaker is not investment grade. The core finding is that capital is not the scarce input here; bankable offtake is. Operators routinely hold term sheets they cannot draw because the layer beneath the debt is unfilled.
It also covers the down payment gap that kills deals, how the market views obsolescence, and why recovery in this asset is a distribution problem. Plus observable pricing from our own book and the security package items operators will actually give you.
If you'd like a copy, shoot us a message and we'll send it over.
Great working with @wintermute_t to advance the commodification of compute. If you’re an AI lab, neocloud, or speculator looking to trade OTC, shoot me a dm.
Wintermute executed its first compute forward referencing Nvidia H100 pricing
Compute is becoming a market of its own, with the tools to price and hedge it now starting to emerge
The same H100 trades at different prices across three markets.
Guaranteed on-demand, August 24, 2026:
- Hyperscaler $10.53 / GPU-hr
- Neocloud $3.73 / GPU-hr
- Marketplace $3.01 / GPU-hr
That is not a wide bid-ask. It is three products sharing a chip name.
Price, tenor, prepayment, fabric, tier and delivery still move together. Two offers at the same headline rate can differ by twenty percent in economic terms. The hyperscaler to marketplace gap on this print is more than 3x before those fields are even negotiated.
Until the description is standard, buyers cannot rank quotes, sellers cannot lay off the book, and lenders have nothing to mark.
Source: CCIR
Startups buying long-term compute get asked for 20–30% down. On a 16-node B300 cluster, ~$13.5M over 3 years, that's a $4.5M down payment.
We're partnering with lenders to finance that prepayment.
Buying compute or looking to originate? lnkd.in/emggdy3g
This week, Nvidia CEO Jensen Huang decided to change tack: He went public this week with the effort, saying a group of firms including Goldman and Blackstone are aiming to collectively finance AI computing deals totaling $500 billion. bloomberg.com/news/articles/…
May: $3.1B GPU-backed loan, 6x book, priced tighter.
July: same issuer, same structure, flexed 150bps wider with covenants attached.
Not a repricing of risk. A change in who sets terms, inside one quarter.
ronitjain2.substack.com/p/public-ai-cr…
The race for compute is driving the biggest data center buildout in history, and it's all still done by hand. For the past few months as a part of the YC S26 cohort, @apai253 and I have been tackling this problem.
Today, we are excited to launch @ProprioRobotics. At Proprio, we are building the physical automation layer for data centers.
Our robots assemble, maintain, and tear down server infrastructure, harvesting reusable components along the way to tackle material shortage.
Welcome to the future!
2K Followers 2K FollowingPrivate investor. Former PM. Former Head of FX Vol at CS, GS. Mathematician “Science is the belief in the ignorance of experts" - Feynman
270 Followers 2K Following🛠 Building in public from UA
⚡ Vibe coder | AI-first stack
💸 Indie SaaS | B2C tools | Medtech
I build, break, and ship — usually with AI.
18K Followers 776 FollowingCo-founder & GP @chemistry. Before, @Indexventures, builder @dropbox. Invested in Plaid, Persona, Bridge, many more. AI, Fintech, SaaS 🎯 Dad x3 boys.
9K Followers 2K FollowingInvesting in startups with @natfriedman + @danielgross; run @aigrant and the Andromeda Cluster; stan dogs and warm carbs & cheese; tech/AI for human flourishing