Snapshot
- Ticker: CRWV (Nasdaq, IPO March 28, 2025 at $40)
- Bucket: GPU-as-a-Service
- Q4 2025 fund position: **436.7 M common (6.10 M sh) + $774.4 M call notional (10.81 M sh) — 21.96 % of 13F
- Position rank: combined #1 issuer in fund
- HQ: Roseland, NJ
Business Overview
CoreWeave runs a US-based GPU cloud built specifically for AI training and inference. Originally founded in 2017 as Atlantic Crypto, an Ethereum-mining shop, the company pivoted to GPU rental in 2019 and was the first non-hyperscaler cloud to deploy NVIDIA H100s at scale in 2022. As of late 2025 it operates ~32 data-center sites, ~360–470 MW contracted power scaling to >1 GW pipeline, and a fleet that includes H100, H200, B200, GB200 NVL72, and (as deliveries arrive) GB300 systems.
The business model is contracted capacity: customers commit to multi-year reserved GPU-hour blocks against which CoreWeave issues delayed-draw term loan (“DDTL”) debt collateralized by the underlying GPUs. This pattern — large, creditworthy customer contracts financing the GPU purchase — is the source of the company’s distinctive 80%+ debt/EV capital structure.
Financial Trajectory
| Metric (USD M) | FY2022 | FY2023 | FY2024 | FY2025 (guide) |
|---|---|---|---|---|
| Revenue | 15.8 | 228.9 | 1,915 | 5,150–5,350 |
| YoY growth | — | +1,346 % | +737 % | +169 % (mid) |
| Gross margin (revenue level) | — | — | ~74 % | ~74 % |
| Adjusted EBITDA | — | — | ~1,200 (62 % margin) | — |
| Net income (loss) | — | — | −863 | (loss) |
The FY24 net loss of ~$863 M is driven by:
- ~$830 M in depreciation on the GPU fleet (assumed 5-year useful life)
- ~$580 M in interest expense on the asset-backed financing
- Stock-based compensation related to pre-IPO grants
EBITDA is meaningfully positive (~62 % margin) — the loss is entirely below-the-line.
Quarterly Trajectory (FY25)
| Quarter | Revenue (~$M) |
|---|---|
| Q1 2025 | ~981 |
| Q2 2025 | ~1,210 |
| Q3 2025 | ~1,360 |
Sequential growth has been ~20–30 % each quarter as new GPU clusters come online and the OpenAI deal contributes incrementally.
Customer Concentration
| Customer | % of FY24 revenue | Notes |
|---|---|---|
| Microsoft | ~62 % | Largely re-renting capacity to OpenAI |
| Top 2 customers combined | ~77 % | Microsoft + a second hyperscaler-adjacent customer |
| OpenAI direct | growing in 2025 | 4 B in 2025 |
| Smaller customers | balance | Cohere, Mistral, Meta (limited), AI startups |
The OpenAI direct relationship is structurally important — it diversifies revenue beyond the Microsoft channel and creates a long-term anchor that can absorb GB200/GB300 capacity as it deploys.
Balance Sheet (~Q3 2025)
| Item | $M |
|---|---|
| Cash + investments | ~3,000–4,000 |
| Total debt (DDTL + bonds) | ~11,000–13,000 |
| Net debt / TTM Adj. EBITDA | ~5–7× |
The capital structure is unusual: GPUs are pledged as collateral against term loans drawn down as deployments scale. The structure works as long as customer contracts perform; if utilization drops or customer credit deteriorates, the financing model stresses quickly.
Capex & Backlog
- FY24 capex: ~$8.7 B
- FY25 capex guide: $20 B+
- FCF (FY24): approximately −7 B (capex consumed everything)
- Contracted backlog (RPO) post-IPO: ~30 B+** range with OpenAI expansion
The contracted backlog is the key “distance from danger” metric — multi-year contracted revenue underwrites the term-loan stack.
Stock Performance
- IPO March 28, 2025 at 47–55 range)
- Highly volatile post-IPO; traded in the $35–185 range through 2025
- The fund initially held common only (Q1 2025), held puts in Q3 2025, then closed the puts and added a $774 M call notional position in Q4 2025
Why It Fits the Thesis
CoreWeave is the purest publicly-traded expression of “AI compute revenue per unit of NVIDIA shipment”. Whereas Microsoft Azure / AWS / GCP report AI revenue diluted by much-larger non-AI cloud businesses, CoreWeave’s revenue line is approximately a function of:
(NVIDIA accelerators in operation) × (hourly rate) × (utilization)
The fund treats it as the most direct way to ride the NVIDIA shipment ramp without owning NVDA — a thesis that the value capture moves from the chip designer to the operator of the chip when supply tightens. Holding both the common (long delta) and a much larger call-option position (convex, levered upside) is consistent with high conviction in the upside scenario.
Position History in the Fund
| Quarter | Position |
|---|---|
| Q1 2025 | New common, small size (post-IPO) |
| Q2 2025 | Common held flat |
| Q3 2025 | Common + puts (hedge against post-IPO rally) |
| Q4 2025 | Puts closed; calls added at $774 M notional; common increased to 6.1 M sh |
The Q3→Q4 swap from puts to large calls is one of the most decisive position changes in fund history.
Risks
- Customer concentration. Microsoft + OpenAI account for the bulk of revenue; loss or repricing of either is existential.
- GPU obsolescence. 5-year depreciation may prove optimistic if newer NVIDIA generations compress unit economics on the H100/H200 base.
- Debt load. Asset-backed term loans require the underlying contracts to perform; if utilization drops, the structure stresses fast.
- Hyperscaler competition. Microsoft has incentive to migrate OpenAI workloads to its own infrastructure as it builds out.
- NVIDIA allocation risk. CoreWeave’s competitive moat is partly NVIDIA’s preferred-access designation; that could erode if NVIDIA expands allocation to peers.
Sources
- CoreWeave S-1 (filed March 2025)
- CoreWeave Q1 / Q2 / Q3 2025 10-Q filings
- OpenAI deal press releases (September 2024; 2025 expansion)
- CoreWeave investor relations