When more than 6 GW is contracted and approximately 1 GW is operational, the same number that looks like protection is also the obligation that has to be financed.
Crusoe’s newsroom announced the initial closing of its Series F on 17 September 2026 — that round, its co-leads, and the capital stack around it were covered here when announced and are not re-reported. What the announcement added, and what has not been read carefully enough, is the juxtaposition of two capacity figures: more than 6 GW of gross contracted capacity across its data centre and cloud businesses, of which approximately 1 GW has already been delivered and is operational.
Alongside those figures, Crusoe states more than $140 billion in total contracted value — with no disclosed term attached, which means it cannot be read as an annual number and is not treated as one here. Customers span AI-native companies, hyperscalers, frontier model developers, and enterprises. The company also describes a vertical integration thesis: it owns the chain from power to tokens, including modular data centre construction.
The two capacity figures sit quietly in the same press release, a few lines apart. Put them next to each other and a structural question opens that the backlog headline on its own does not answer: roughly five-sixths of what has been contracted still has to be built — which means the same number that is cited as evidence of demand certainty is also a forward commitment to spend capital that has not yet been deployed.
The key insight: A backlog against already-built capacity is a pure asset — cost is sunk, revenue arrives. A backlog against unbuilt capacity is both an asset and an obligation simultaneously, because delivering on it requires capital that still has to be raised and deployed. One number, two signs — and the sign that dominates depends entirely on execution.
The gap between these two bars is the part that still has to be paid for.
The Structural Read
The standard read of a contracted backlog is downside protection: demand is already signed, revenue is already visible, the demand-side risk has been taken off the table. That reading is correct, but only conditionally — the condition being that the capacity required to fulfil the contracts already exists. When it does not, the contract is doing two jobs at once. It is an asset because it represents committed demand. It is also an obligation because it obligates future capital expenditure regardless of what happens to the market environment in the interim.
This is a general property of backlog accounting in capital-intensive industries, not a specific observation about Crusoe. But it matters more when the gap between contracted and operational capacity is wide — and here, the gap is, on Crusoe’s own figures, substantial. The question a backlog at this scale raises is not whether demand exists. The contracts answer that. The question is what the obligation to build costs, and what it earns, across a range of futures — including futures where the rate of demand growth is different from the rate assumed when the contracts were signed.
That last clause is where the most precise thing said about this category this week becomes relevant. Rory O’Driscoll, general partner at Scale Venture Partners, made a distinction in a 20VC investment-committee discussion that is easy to read past.
Rory O’Driscoll — Scale Venture Partners, 20VC
“The data center trade is very levered, so very exposed, not just to AI usage, but to growth in AI usage… if I had a portfolio across the whole kind of AI spectrum, I wouldn’t want it all to be in that AI capex trade where any kind of slowdown when you’ve got like four or five to one leverage can be pretty brutal.”
Two clarifications before following this through. The four-or-five-to-one is O’Driscoll’s characterisation of the sector trade in general — it is not a measured leverage ratio for Crusoe or for any named company, and it is not used as one here. And his framing of the risk is the part that matters analytically: the exposure he names is not to AI usage but to growth in AI usage. The first derivative, not the level.
Follow that distinction through and it reframes what a long backlog actually protects against. A signed contract protects against a decline in demand — a customer walking away, a market reversal. It does not protect against a decline in the rate of growth, because the capital expenditure commitment was sized against an expected rate of increase. If demand goes flat — not down, flat — the obligation to build is unchanged, while what the finished capacity earns relative to expectations is not. The backlog is long in the level and potentially short the growth rate. Those are not the same hedge.
Map of AI — Infrastructure Layer
Where the leverage sits in the AI stack
The infrastructure layer — data centres, power, compute — is the part of the AI stack where contracts are longest, capital commitments are largest, and the gap between signing and delivery is widest. That is structurally different from software layers above it, where marginal delivery cost is near zero. The Map of AI makes this layer distinction explicit: the further down the stack, the more a backlog is simultaneously an asset and a build obligation — and the more sensitive outcomes are to the timing of demand, not just its existence.
Jason Lemkin of SaaStr, arguing for a portfolio approach to data centre investment in the same discussion, named the entry conditions he underwrites to: “invest in all the data centers that rise above a certain level of projected DCF, projected growth, protected margins, and most importantly, guaranteed circular financing.” That last item deserves a full stop after it.
Circularity — vendor financing, equity taken by suppliers, capital that returns to the party that supplied the hardware — is the mechanism that sceptics of the AI build-out cite when arguing the cycle is self-referential. Here, an investor names it as the most important criterion he underwrites to, not as a risk to be managed around. That is reported as what he said; it is not evidence about any company’s financing structure, and nothing here asserts that Crusoe’s financing is or is not circular. The general property is the observation worth sitting with: when a structural risk factor becomes a stated entry requirement rather than a reason for caution, it has stopped functioning as a warning and started functioning as infrastructure — as a load-bearing assumption in the underwriting model rather than a variable to be tested.
And then Lemkin, having described the asset admiringly — “they’re able to build their own modular data centers, they own the chain from power to tokens” — declined it on price. His words: “this is all a spreadsheet investment. I’ve done the math, and I believe it’s just below the fold. It’s just a hint too expensive to hit our margin despite the growth.” That is a specific investor, applying his own fund’s margin test, declining this particular price. His math is not shown. Nothing here calls Crusoe expensive or cheap. What the episode is is a data point about where the marginal buyer currently sits — someone bullish on the category who still finds the current price one tick past his threshold.
Three Implications
IMPLICATION 1 — BACKLOG ACCOUNTING HAS TWO SIGNS
The same contracted figure that demonstrates demand certainty also represents a forward build obligation. Reading only one sign — citing the backlog as pure protection without noting that roughly five-sixths of it remains unbuilt — gives an incomplete picture of the capital commitment it implies. This is a general property of capital-intensive backlog accounting. It applies here with unusual clarity because the gap between contracted and operational capacity is, by Crusoe’s own figures, large.
IMPLICATION 2 — DEMAND LEVEL AND DEMAND GROWTH RATE ARE DIFFERENT RISKS
A signed contract hedges against the level of demand going to zero — a customer walking away. It does not hedge against the growth rate of demand decelerating, because the capital expenditure schedule was set against an expected rate of increase. O’Driscoll’s framing — exposure to growth in AI usage, not usage itself — is precise. Any analysis of infrastructure backlog that conflates the two risks will systematically overstate what the contract actually protects against.
IMPLICATION 3 — WHEN RISK FACTORS BECOME ENTRY REQUIREMENTS
Lemkin’s list — DCF, growth, protected margins, and “most importantly, guaranteed circular financing” — is a window into how the current underwriting model for AI infrastructure is constructed. When a mechanism that analysts flag as a structural vulnerability is named by an investor as the most important criterion for entry, it signals that the mechanism has been priced in as a feature rather than treated as a risk. That is a statement about the current state of capital allocation in the sector — not a prediction about how the cycle resolves.
The Map of AI maps more than 200 companies across nine layers of the AI stack — from raw compute and power infrastructure at the base to applications at the surface. The infrastructure layer is where capital commitments are largest, delivery timelines are longest, and the gap between a signed contract and an operational facility is widest. Understanding which layer a company occupies — and what the structural properties of that layer are — is how you read a backlog figure correctly rather than in isolation.
Crusoe’s more than $140 billion in total contracted value — carrying no disclosed term, and therefore not an annual figure — is real evidence of demand at scale. But set next to the two capacity numbers Crusoe itself publishes, it is also evidence of the scale of what still has to be financed and built: more than 6 GW contracted, approximately 1 GW operational, roughly five-sixths of the obligation still ahead. A backlog is only as clean as the capacity behind it. Where the capacity is unbuilt, the same figure that looks like safety is also a commitment — and what that commitment earns depends not on whether demand exists, which the contracts already answer, but on whether the growth rate that justified the capital expenditure schedule actually arrives.
Sources: Crusoe Newsroom — Series F Announcement, September 2026; Rory O’Driscoll remarks, 20VC investment-committee discussion (via analysis); Jason Lemkin remarks, SaaStr / 20VC investment-committee discussion (via analysis). All Crusoe figures are Crusoe’s own with their original qualifiers. Not investment advice.
The capacity and value figures above are Crusoe’s own, from its newsroom announcement of the Series F initial close dated 17 September 2026, and its qualifiers are retained: more than 6 GW of gross contracted capacity, of which approximately 1 GW is already delivered and operational. The more than $140 billion of total contracted value carries no disclosed term. It is therefore not an annual figure and is not annualised anywhere above.Rory O’Driscoll’s four-or-five-to-one is his characterisation of the data-centre sector trade in general. It is not a measured leverage ratio for Crusoe or for any other named company, and it is not used as one above.Jason Lemkin’s arithmetic is not shown, and his conclusion is a test against his own fund’s margin requirement rather than a valuation finding. Nothing above calls Crusoe expensive, cheap, overvalued, or a good or bad investment.The phrase “guaranteed circular financing” is Lemkin’s own underwriting criterion, quoted as such. Nothing above asserts that Crusoe’s financing, or any named company’s, is circular. The round, its co-leads and the capital stack around it were covered here when announced and are not re-reported above as news. Crusoe’s revenue, margins, debt, cost of capital and capital-expenditure schedule, the term of the contracted value, any customer or contract name, any company-level leverage ratio, and any Crusoe or investor response to these remarks are not established and do not appear. Nothing above predicts a slowdown, a bubble or a correction, and nothing above is investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
