If Take-or-Pay Bills Idle Hours, Where Does Cash Stick?

ANALYSIS · MEMBERS CONTINUE · 3 min free / 12 min full · Data as at 29 Sep 2026

When committed take-or-pay is about 98% of neocloud revenue and contracted power sits about 2.5× active power, the money question is not whether offtake was signed — it is what stuck step the map already assumes: the delivery gate that starts the bill, offtake softness before halls catch, or concentration and credit as the long clock.

In short

• The question: if take-or-pay bills idle hours after delivery, where does cash stick on the neocloud map?
• The answer: Scenario B — delivery remains the bind. Energised megawatts and accepted GPUs are the stuck step that starts the bill; idle hours after start still bill the customer; undelivered take-or-pay is a promise (plus optional prepay/deferred), not recognised cash-from-hours
• Why: CoreWeave’s Q2 2026 10-Q prints 98% committed/take-or-pay revenue, US$103.7B RPO, and deferred revenue of US$9.7B; the same quarter’s IR print shows 1.5 GW active against ~3.7 GW contracted — about 2.5×. Microsoft’s FY2026 10-K puts US$169B of US$194B purchase commitments (including take-or-pay) due in the next fiscal year
  Behind the wall: three scenarios with the arithmetic · RPO band splits + hall-gap maths · six practical takeaways · Members' FAQ
• What would break it: committed mix shifting toward on-demand before halls catch — or a large offtaker that does not perform while contracted GW stays dark
• Nothing here is a recommendation. This is a map of what today’s prints already assume, with the working shown

What's in this piece: the numbers that matter · why take-or-pay sits at the stuck step · what the map assumes in one line · members from here · who gets paid · three scenarios, arithmetic shown · our read · what would change the view · practical takeaways · FAQ · sources

The numbers that matter

!Neoclouds — take-or-pay committed mix vs hall gap

Chart 1 — Committed offtake versus the hall gap. Sources: CoreWeave Form 10-Q period ended 30 Jun 2026; Q2 IR 11 Aug 2026. Education only.

Why does take-or-pay sit at the stuck step?

The neoclouds layer sits between scarce accelerators and builders who need clusters now. Specialist fleets finance GPUs, place them in power-ready halls, and sell capacity by the hour or by contract — different from a hyperscaler platform (see Neocloud explained, What does CoreWeave do?, GPU-as-a-service explained, Take-or-pay GPU contracts explained, Remaining performance obligation explained).

Who gets paid in this layer is whoever keeps a delivered take-or-pay block running: after the operator energises the megawatts and the customer accepts the GPUs, the reserved fee bills even if utilisation is low. Who waits is the operator with a signature against a dark hall — or the customer who prepaid into deferred revenue before access. Today’s Explained teaches that mechanism. This Analysis asks the money so-what: what does the map already assume when 98% committed revenue sits next to a ~2.5× hall gap?

CoreWeave’s Q2 2026 10-Q states the split in one sentence: it sells access “either through committed contracts, which are take-or-pay, or on-demand, which are pay-as-you-go,” and committed contracts “generally have a predetermined term and start either on a fixed date or when we deliver the capacity specified in the contract.” Revenue from customer commitments — including capacity delivered before a commitment start date — was 98% of total revenue for the three months and six months ended 30 June 2026. The same filing prints US$103.7B of remaining performance obligations and US$9.7B of deferred revenue. The IR release the same week shows 1.5 GW active power against approximately 3.7 GW contracted (11 Aug 2026).

Two clocks sit on the same book. The offtake clock is the take-or-pay clause and the 98% committed mix. The delivery clock is energised megawatts and accepted capacity — the on-switch the 10-Q names. Collapse those clocks into one backlog headline and you miss the bind. A later contracted-power print of ~4.2 GW as of 11 August (company press, 17 Sep 2026; short-contracts Wire) widens the offtake paper further; it does not silently replace the 30 June 3.7 GW figure, and we do not invent an updated active-power denominator for that later date.

Hyperscaler contrast, not confusion: Microsoft’s FY2026 10-K shows purchase commitments of US$194.060B, of which US$169.008B is due in fiscal 2027, and the footnote says those commitments “primarily relate to datacenters and include open purchase orders and take-or-pay contracts.” Oracle’s +850 MW / 97.9% util print and Project Jupiter force majeure sit on the same scarcity weather as a delivery/power contrast — already worked as a company piece (Oracle power Analysis; Oracle Wire). This piece stays on the neocloud map: what take-or-pay plus the hall gap already assume.

That is not a results recap. It is a map question. Where the stuck step moves tracked packaging → memory → power clocks at the layer level. Short contracts near $40m/MW asked what the map assumes when short-price and GW prints arrive together. Yesterday’s Oracle company piece asked what one hyperscaler price assumes about power. Today asks what the neocloud map already assumes when take-or-pay is nearly the whole book and contracted power still dwarfs active power.

What does the map already assume?

Three candidate binds compete for the label “stuck step”:

1. Offtake softness / mix shift — committed take-or-pay fades toward on-demand before halls catch; the bill starts later.
2. Delivery gate — energised MW + accepted GPUs remain scarce; idle hours after delivery still bill; undelivered take-or-pay is a promise plus optional prepay.
3. Concentration / credit — a handful of offtakers (Microsoft, OpenAI, Jane Street ~US$6.0B) decide whether delivered blocks keep paying even when the clause looks firm.

A tip sheet would pick a winner and a ticker. Analysis works the other way: ask which bind today’s dated prints already require for the Stack’s neocloud read to keep holding — then show the arithmetic members can check.

Our read, in one line: the map already assumes Scenario B — delivery (energised MW + accepted GPUs) is the stuck step that starts the bill; idle hours after delivery still bill; undelivered take-or-pay is a promise (plus optional prepay/deferred), not recognised rent; concentration and credit remain the long risk clock. The three scenarios and the arithmetic are below.

The free preview ends here.

—— Still to read · ~8 of 12 minutes ——

MEMBERS CONTINUE HERE

• Three scenarios with the arithmetic
• RPO band splits + hall-gap maths
• Six practical takeaways
• Members' FAQ