Demand Gauges: CoreWeave Sees "Absolutely Not Any Reduction in Orders" — and Margins "Expanding Faster Than the Rates Are Going Up"

Updated

Demand Gauges: CoreWeave Sees "Absolutely Not Any Reduction in Orders" — and Margins "Expanding Faster Than the Rates Are Going Up"

The demand-side rebuttal in the overbuild debate (this note's bear chain: Zitron's $50–100B warehoused-GPU estimate, Burry's depreciation thesis) got its clearest articulation yet from the largest pure-play buyer of Nvidia systems. CoreWeave CEO Mike Intrator, at the company's inaugural Fully Connected conference (Sept 30): "There has absolutely not been any reduction in orders1... They are not going to in any way reduce the demand. The market is looking for more infrastructure... and it's just a question of where you put it" (BigGo Finance). On the rate overhang — the most sensitive point for a debt-funded neocloud: "The margins that we are earning are expanding faster than the rates are going up," a trend he projected to persist through 2027 into early 2028.

Product-side confirmation: Vera Rubin NVL72 reached general availability and is running production workloads — Cognition (the Devin developer) is the first customer. The Vera CPU enters select-customer trials within weeks, with agentic AI generating "substantial demand for CPUs across the board."

The equity market's answer to the same facts: CRWV trades near $87 — up ~22% YTD but still ~53% below its June 2025 record close. Sell-side keeps bulking bullish (JPMorgan upgrade to Overweight, William Blair Outperform, Truist Buy/$165 — roughly double spot). The divergence is itself a datum: public equity investors are pricing neocloud financing-cost and dilution risk, not demand weakness2. Meanwhile the rental-price tripwire stays unfired — Nebius's 17–21% pay-as-you-go hikes took effect Oct 1 (second hike in three months) and CoreWeave is signing contracts at higher prices.

Every observable rental/demand gauge still points up; the bear chain still lacks its confirming disclosure (a write-down, a deployment-percentage disclosure, or a depreciation-schedule reversal — the watch remains open).


  1. An instance of Near-term GPU capacity now sells at a premium to future supply. — The largest pure-play buyer reports unbroken order flow while rival operators raise pay-as-you-go rents for a second time in three months — the observable signature of persistent scarcity pricing, not overbuild. ↩︎

  2. An instance of The cost of money, not customer demand, now gates the AI capex boom. — Order books print strength and rental rates rise while the equity discount sits entirely on financing costs and dilution — the gate on the buildout is the price of money, not demand. ↩︎

Part of

This finding is an example of a pattern recurring across your work:

Revision history

  • Demand-side update: CoreWeave CEO's no-slowdown quotes, Vera Rubin NVL72 in production, and the CRWV equity-vs-fundamentals divergence.
    · by the agent
  • Update bear-case note: Burry's 1960s computer-leasing escalation against Nvidia's Sept 27 depreciation slide, plus the Oct 1 Nebius price card live and CoreWeave step-change quote — no price cuts, falsifier not fired.
    · by the agent
  • Add Burry's 2027 timeline shift and 1968 Money Game parallel, plus the AWS 15% Capacity Block hike as the newest demand gauge.
    · by the agent
  • Add Burry's 2027 timeline shift and 1968 Money Game parallel, plus the AWS 15% Capacity Block hike as the newest demand gauge.
    · by the agent
  • Add Burry's 2027 timeline shift and 1968 Money Game parallel, plus the AWS 15% Capacity Block hike as the newest demand gauge.
    · by the agent
  • Add Zitron's Sept 29 "Dead Money" full arithmetic, the Anthropic S-1 leak, and the demand-side price-hike counters (Amazon +15%).
    · by the agent
  • Zitron's "Dead Money" essay brings the overbuild critique into the mainstream (Fidelity's Timmer, FT/Morgan Stanley unplugged-GPU estimate); Anthropic S-1 leak ($4.6B rev, $8B loss) and the ARR-methodology controversy quantify the demand base.
    · by the agent
  • Zitron's "Dead Money" essay brings the overbuild critique into the mainstream (Fidelity's Timmer, FT/Morgan Stanley unplugged-GPU estimate); Anthropic S-1 leak ($4.6B rev, $8B loss) and the ARR-methodology controversy quantify the demand base.
    · by the agent
  • Burry's Oct 1 Substack attacks Nvidia's residual-value slide (Silicon Data DCF vs resale prices), invokes the 1968 leasing collapse, and swaps the NVDA short for Sept 2027 mid-$100s puts.
    · by the agent
  • Add SemiAnalysis ClusterMAX 3.0 field evidence (backwardation, 4-year H100 contracts, 19/77 Medallion ratings) as counter-evidence to Burry's depreciation case; write-down watch still unfired.
    · by the agent
  • Added the first hard write-down number in the overbuild debate: Nvidia's $0.4B H1 charge for excess H200 inventory — generational churn at the vendor level, while hyperscaler/neocloud disclosures remain absent.
    · by the agent
  • Revision: added Burry's write-off warning (Sept 25) and the ClusterMAX four-year-contract counterpoint; confirmed no write-down disclosure yet
    · by the agent
  • Updated without a stated reason.
    · by the agent