CRWV small group meeting notes (Part 1)

CFO

GPU

• Contracts extend 3–5 yrs (rev-wtd ~4.5 yrs); GB200/NVL72 deal = 5 yrs + 2 yr extension.

• Mid-cycle re-contracts (A100, Hopper) at 1–3 yrs, despite Blackwell rollout.

• Tesla/Volta GPUs remain in market >8 yrs with strong economics.

• Useful life ~6 yrs, often longer.

• Submitted ~2 500 GB200 cabinets to MLPerf; AWS/GCP/MSFT did not participate.

Capacity

• Incremental deals will fully commit 1.6 GW with GPU compute by end-2026.

• 1.6 GW covers ~$29 B backlog + $4 B incremental → capacity > backlog.

• Sold out next ~2 qtrs; any extra MW would sell “in hours or days” — capacity is the bottleneck.

• Clients typically sign capacity ~2 qtrs ahead—rarely > 6 m before go-live.

• 420 MW deployed today; plan >2× (≥420 MW) over next 9 mos to meet visible demand.

Rev mix

• 2024 mix: 96 % contracted, 4 % on-demand; similar split expected near term.

• Target 2.5 yr payback (±0.25 yr) on each architecture, within the first committed contract.

Debt

• Debt tied to RPO/backlog via signed off-take contracts—no speculative CapEx.

• Multi-yr, take-or-pay, fixed-price, non-cancellable contracts → self-amortizing project-finance debt.

• At maturity, debt is paid off; cash flows return to equity.

• Operating cash flow is positive on existing contracts.

• Structure ensures natural deleveraging as backlog converts to revenue—no perpetual debt buildup.

Cost

• Cluster cost breakdown: 85 % GPU nodes, 10 % networking, 5 % storage/ancillary.

CORZ

• No plan to exit Core Scientific leases—large-scale capacity now vs. waiting until ~2028 is a peer advantage.

• New leases only on client-demand consensus (e.g. 5 clients for 100 MW → secure mid-2026 delivery).

Main contents:

Q: What is the strategy for diversifying the existing client base over the next 5–10 years, and how will offerings apply to enterprises and smaller businesses?