CoreWeave posted Q2 2026 revenue of $2.58 billion, up 112% year-over-year, edging past LSEG consensus of $2.56 billion. Adjusted EPS came in at a loss of $1.03, beating the expected loss of $1.20. Shares jumped 12% in after-hours trading, adding to a 26% year-to-date gain. The company was simultaneously added to the Nasdaq-100.
The headline growth rate masks the mechanics underneath. Adjusted EBITDA doubled to $1.51 billion at a 59% margin — up from 56% in Q1, which management had called the bottom of the margin cycle. Operating cash flow turned positive. Free cash flow landed at negative $5.74 billion as the company converted borrowed capital into GPU racks. Interest expense hit $640 million in the quarter, more than double the $267 million from a year ago. Total debt: $35 billion.
| Metric | Q2 2026 | Q2 2025 / Prior | Change |
|---|---|---|---|
| Revenue | $2.58B | — | +112% YoY |
| Adj. EBITDA | $1.51B | — | 59% margin (up from 56% in Q1 2026) |
| Adj. EPS | −$1.03 | est. −$1.20 | Beat by $0.17 |
| Free Cash Flow | −$5.74B | — | Capex-driven outflow |
| Interest Expense | $640M | $267M | +140% YoY |
| Net Loss | $626M | $290M | +116% YoY |
| Total Debt | $35B | — | — |
CoreWeave's revenue backlog reached $104 billion as of June 30, with more than $25 billion in additional commitments signed in early Q3 before the report dropped. Anchor customers include OpenAI (up to ~$22.4 billion total committed), Meta (up to ~$35.2 billion across two agreements), Jane Street ($6 billion signed April 2026), and Anthropic on a multi-year deal for Claude model training and inference. New enterprise names this quarter—Bentley Systems, Caterpillar, Grammarly—signal demand is moving beyond hyperscale AI labs into industrial verticals.
| Customer | Committed Amount | Details |
|---|---|---|
| Meta | ~$35.2B | Two separate agreements; anchor customer |
| OpenAI | ~$22.4B | Anchor customer; total committed |
| Jane Street | $6B | Signed April 2026 |
| Anthropic | Multi-year (undisclosed) | Claude model training and inference |
| Bentley Systems | Undisclosed | New enterprise customer, Q2 2026 |
| Caterpillar | Undisclosed | New enterprise customer, Q2 2026 |
| Grammarly | Undisclosed | New enterprise customer, Q2 2026 |
Active power stands at 1.5 gigawatts across purpose-built US and Europe data centers, running ultra-high-density GPU clusters with high-speed interconnects. Total contracted power is ~3.7 GW. CoreWeave set MLPerf records this quarter for training and inference throughput on NVIDIA Grace Blackwell hardware and claims the lowest cost-per-token for inference in its own tests. The platform launched CoreWeave ARIA and CoreWeave Sandboxes—unified tooling for training, inference, observability, and reinforcement learning. Full-year 2026 revenue guidance is $12–$13 billion, exit ARR of $18–$19 billion, and capex of $31–$35 billion—roughly $85–95 million per day in new infrastructure.
| Metric | Value |
|---|---|
| FY2026 Revenue Guidance | $12B – $13B |
| FY2026 Exit ARR | $18B – $19B |
| FY2026 Capex | $31B – $35B (~$85–95M per day) |
| Active Power (as of Q2 2026) | 1.5 GW |
| Total Contracted Power | ~3.7 GW |
| Unactivated (contracted-but-dark) | ~2.2 GW |
The structural risk is a duration mismatch. Customer agreements average roughly three years. CoreWeave's newest financing—a $3.1 billion term loan backed by HPC infrastructure—runs five years. The company carries more than $10 billion in unsecured debt and convertible bonds. Net loss widened to $626 million in Q2, from $290 million a year earlier. Annualized interest expense runs $2.56 billion—a fixed-cost floor the company must clear before backlog converts to margin.
SpaceX has begun selling excess GPU compute capacity from its own Colossus cluster. Meta has publicly considered launching a cloud business, which would complicate the same $35.2 billion relationship currently anchoring CoreWeave's backlog. Microsoft alone accounted for roughly 67% of CoreWeave's full-year 2025 revenue, highlighting customer concentration.
The buy-build calculus is shifting toward specialized GPU clouds in the near term. CoreWeave's viability as a long-term counterparty depends on converting 2.2 GW of contracted-but-dark capacity into revenue before its debt service schedule tightens.