The Offtake Is Not the Mandate

The Agent Economy, from someone funding the physical rebuild of the economy.

The offtake is not the mandate.

Jane Street led $1.5 billion into Fluidstack at an $18 billion valuation, reported early this week by Forbes and Crunchbase News. Crunchbase put total funding just over $2.6 billion. The prior disclosed mark, announced in July, was $7.5 billion. Roughly two months. The price more than doubled.

The structural fact underneath the headline is the one that matters, and it is why the check makes sense. Fluidstack builds and operates the buildings, runs the software that stands the clusters up, and leaves the silicon to the customer. Multi-vendor by design. Anthropic's $50 billion American compute program is the anchor: custom sites in Texas and New York. Meta, Mistral, Poolside, Black Forest Labs sit on the customer list. Google has backstopped lease obligations at converted mining campuses. The balance sheet is power, concrete, and contracted cash flow. Not a warehouse of H100s.

Jane Street is not a tourist here. The same desk has been buying compute at multi-billion scale: about $6 billion of CoreWeave cloud plus $1 billion of equity, and a multi-year GPU contract with Crusoe that Bloomberg put near $13 billion. A market maker with visibility into utilization and price just led equity into the company that turns offtake into buildings. That is someone who prices contracted cash flows for a living, deciding Fluidstack's offtake clears.

So in one week: the offtake model got an $18 billion mark, and a trading firm that already rents the GPUs decided the contract was the asset. Fluidstack got that call right.

I should say this plainly. Fluidstack is a portfolio company. Not every week this newsletter draws its news object from something I am in. Most weeks the seam shows up in someone else's filing. This week the named offtake sits in a company I backed, and pretending otherwise would be worse than naming it. Celebrating the round is the honest move. The cut is not that Fluidstack is missing a product. The cut is that the market just priced one instrument cleanly, and still has not named a different one.

The third thing still is not there. That is not Fluidstack's miss.

The read everyone had

The obvious take is that the neocloud stack is maturing. Asset-light wins. Chip neutrality is a feature. Jane Street's check proves compute infrastructure has left venture and entered the desks that underwrite commodities. Crusoe raised over $3 billion the same week, Bloomberg reported. The category is eating capital because the backlog is real.

That read is right about Fluidstack and right about the capital. It is incomplete about the layer next door.

An offtake proves the contract, not the grant

An offtake is a named obligation: capacity, window, counterparty, cash flows a desk can model. Jane Street can underwrite that. Fluidstack can finance against it. Anthropic can plan training runs against it. The contract has a brand, a term sheet, and a price. Fluidstack's whole week is proof that instrument works.

It does not prove, to the party that has to honor a draw against that capacity: that this agent, under this mandate, was cleared for this slice, at this size, in this window.

Those are different artifacts. One is the commercial paper just marked at $18 billion on Fluidstack's book. The other is the authorization record that would have to travel with the agent when it takes the megawatts. The first got a lead investor who prices instruments for a living. The second still has no brand. Confusing them turns a clean offtake into a permission fight after the fact.

Here is where that bites, and notice whose problem it is. A lab's agent, or a reseller's, or a hedge fund's, draws against reserved capacity. The offtake is clean. The building is live. The invoice will clear. Fluidstack did its job. Then the draw sits outside the size the principal meant, or against a blacklisted counterparty, or past the mandate window. The offtake was real. The open question is whether the buyer's agent was allowed to take that slice. When you ask, the answer lives in the buyer's dashboard.

Payment and offtake were captured. Permission never crossed the table. That is a buyer-side hole, not a verdict on the offtake.

This is also why the Jane Street check is the tell, not the trophy. The firm already sits on both sides of the compute book: equity and offtake in the builders, rented GPUs at the clouds, and a stake in the lab that fills the buildings. That triangle prices contracted megawatts, which is exactly what Fluidstack sells. It still does not mint a portable grant for the agent that will draw them. Different product. Different buyer.

Same move as Know Your Agent. Same move as the Ramp line item. Same move as the crawl toll. Markets keep naming the commercial object and calling it the control. Fluidstack named the commercial object correctly. The control for the agent is still unpriced.

The machine-speed problem

Human desks used to reconstruct a sign-off while the market cleared on T+1. An agent drawing capacity does not wait. Megawatt reservations that are still a login expire into nothing, or get consumed by a swarm that already shared the credential. By the time you pull the offtake and argue about intent, the next thousand draws have posted. The offtake was never the weak link. The login was.

Authorization that is not captured at the instant of the draw is authorization nobody has. The offtake was signed last quarter. The agent acted this morning. Those clocks do not meet unless something travels with the agent.

The offtake desk will say the contract was fine. The agent desk will say the reservation was fine. Both can be right and the draw still unauthorized. That is the expensive surprise capacity markets have not priced yet, even as they price everything else, including Fluidstack.

What the right layer looks like

What still needs a name is narrow. A record, at the moment of the draw, of which agent, under whose mandate, cleared for what size, in what window, that can be checked before the next step clears. Portable. Pre-execution. Independent of whose offtake spreadsheet the reservation sits in.

That artifact still has no brand name. The offtake does. Jane Street just paid up for Fluidstack's version of the one with a name, and that was the correct check.

Watch who buys the unnamed one first. My bet is not the desk running the agent. It is the clearing member, the insurer, the principal who cannot reconstruct intent after a swarm has already drawn: the people whose money is at risk when a machine takes a slice nobody on their side can prove was allowed. Fluidstack already sold them the capacity contract. Someone else still has to sell them the grant.

The order is always wrong

Here is the pattern, and this week was the cleanest instance of it in capacity markets. We priced the offtake. We financed the buildings. We let a market maker mark the contracted cash flows. Fluidstack is what that sequence looks like when it works. The one layer the market did not ship is the one a counterparty can hold without trusting the buyer's login.

I've watched that sequence in physical markets. It never runs in the right order. The proof gets built after the first expensive surprise. Capacity just got the offtake priced in equity, correctly. The authorization record is still waiting for its own lead investor.

Close

Jane Street bought a contract, not a warehouse. Fluidstack sold buildings and offtake, not a chip pile. Both are right. That is the win.

Atoms are finally clearing like an instrument. Watch what gets built next. If who authorized this machine keeps collapsing back into the offtake, we will have marked a market we still cannot audit on the agent side. The offtake was never the problem. Fluidstack proved that this week. The grant is the open instrument.

Max

Disclosure: Accords, where I'm Executive Chairman, is a portfolio position I chair. Fluidstack is a portfolio company; this issue draws on their round as the week's news object.

Max Curnin has spent 11 years and 45+ investments on the physical rebuild of the economy: energy, compute, robotics, and the materials underneath. He writes The Agent Economy on the category he invests in.