Wednesday, Oct 7 | --:--
Back to home

CoreWeave’s $104B Backlog Is Demand — and Delivery Risk

CoreWeave reported ~$2.58B Q2 revenue (+112% YoY) and a $104.2B contracted backlog (+246% YoY), with ~$626M net loss as interest expense climbed. Beside NVIDIA’s $500B financing MOUs, this is the public neocloud print showing how fast rented GPUs convert to backlog — and how hard they are to energize on schedule.

Times of AI Desk 4 min read Livingston, NJ View as Markdown
Cover illustration for CoreWeave’s $104B Backlog Is Demand — and Delivery Risk

A $100B+ backlog is both a demand signal and a delivery risk. If CoreWeave cannot energize sites and rack GPUs on schedule, frontier training slips.

CoreWeave reported Q2 2026 results that double as a demand census for rented AI compute. Revenue reached roughly $2.58 billion (about +112% year over year from $1.21 billion). Contracted revenue backlog hit $104.2 billion as of June 30, 2026 (+246% YoY from ~$30.1 billion), with a larger share of contracts extending beyond 48 months. Net losses remained large (on the order of **$626 million** for the quarter in contemporaneous coverage) as interest expense from debt-funded expansion climbed — even as adjusted EBITDA margins stayed high (~59% on ~$1.51 billion adjusted EBITDA in company materials). Shares jumped after hours on the print.

Key figures

Metric Q2 2026 (company / wire consensus)
Revenue ~$2.58B (+112% YoY; slight beat vs ~$2.56B street)
Revenue backlog $104.2B contracted (+246% YoY)
Backlog mix Longer-dated contracts growing (share >48 months rising into the ~20% range)
Adj. EBITDA ~$1.51B · ~59% margin (company materials)
Net loss ~$626M (widened YoY on interest / scale)
EPS Loss ~$1.14 vs street loss ~$1.41 (beat on loss)
Customer color Period included large Meta / Anthropic / other AI cloud commitments referenced in earnings coverage

CoreWeave is the public neocloud bellwether: GPU clusters as a service for labs that will not wait for hyperscaler lead times. Finances / infra earnings day — not a model GA. Equal-strength beside NVIDIA’s $500B financing platforms (Aug 10): one side sells chips and structures credit; the other reports how fast rented GPUs convert to backlog.

Limits

  • Net-loss and customer-concentration color lean on wire consensus alongside company materials.
  • Backlog is contracted revenue, not cash in hand — delivery and power timing remain the open risk.
  • Competition from hyperscalers and other neoclouds is not priced in this print.

Sources

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading