CRWV CoreWeave

Research status: Primary-source refreshed through the Q2 2026 10-Q and September 2026 capital raises. Valuation reference: $81.36 close on 18 Sep 2026.

The operating-demand evidence remains exceptional, but the common-equity case has become even more explicitly a capital-structure underwriting. At 30 Jun 2026 CoreWeave reported $35.6B of total indebtedness and $10.0B of undrawn availability across revolving/DDTL facilities. Several secured facilities carried effective rates around 9–15%, including DDTL 1.0 at 15%, DDTL 2.0 at 11%, and multiple facilities around 9%.

On 17 Sep CoreWeave launched another convertible financing and an ATM equity programme. The convertible was subsequently priced at $3.7B of 2.875% notes due 2033, upsized from $3.0B, with an additional $500M purchaser option; the initial conversion price is about $97.85/share. The same capital-markets package established capacity to sell up to 35M Class A shares through an ATM programme. The lower convertible coupon versus secured debt is directionally positive for funding cost, but it introduces additional potential dilution and confirms that enormous external capital needs remain integral to the model.

The thesis is therefore unchanged in direction but strengthened in its main caution: CoreWeave can have extraordinary revenue/backlog growth while still producing mediocre common-equity returns if capex, financing cost, dilution and hardware depreciation consume the economics. At $81.36, CRWV remains investigate. A buy requires project-level evidence that contracted GPU capacity earns returns comfortably above the fully loaded cost of capital under conservative residual-value assumptions.

Investment case

CoreWeave is proving real demand for purpose-built AI cloud capacity, but the common stock is economically a levered residual claim on a rapidly expanding portfolio of GPU/data-center projects.

> The PowerFund thesis is that CoreWeave creates attractive per-share value only if contracted capacity fully amortizes hardware/facility investment, mature GPUs recontract at useful prices, and revenue/operating leverage compounds faster than interest, depreciation, lease obligations and dilution. The September financing reinforces that funding economics—not headline demand—are the decisive variable.

Capital-structure baseline

• Total indebtedness at 30 Jun 2026: $35.6B.
• Undrawn availability at 30 Jun: $10.0B across revolving/DDTL facilities.
• Representative effective secured-debt rates disclosed in the Q2 10-Q: ~9–15%.
• Sep 2026 convertible: $3.7B, 2.875%, due 2033, plus up to $500M additional option; initial conversion price about $97.85/share.
• Sep 2026 ATM programme: capacity for up to 35M Class A shares.
• 2026 capex framework remains extremely large relative to the current equity base.

What improved

The new convertible is much cheaper in cash coupon than much of the secured debt stack and extends maturity to 2033. If proceeds refinance expensive secured obligations, the financing mix can improve.

What did not improve

External funding remains structurally necessary. The convertible and ATM transfer some financing risk into potential dilution rather than eliminating it. Common-equity underwriting therefore requires fully diluted, project-level returns rather than EV/revenue comparisons.

KPI ladder

Existing 24/60-month equity scenarios remain wide and provisional; the September financing does not justify narrowing the range.

• Backlog conversion without material contract cancellations.
• Customer diversification beyond the largest hyperscaler/AI-lab customers.
• Falling interest expense as a percentage of revenue.
• Older Ampere/Hopper fleets recontracting at attractive utilization and pricing.
• Capacity delivered on schedule against contracted demand.
• Positive adjusted operating leverage translating into GAAP/FCF economics.
• Refinancing at materially lower spreads.
• Managed inference/software revenue becoming meaningful relative to raw infrastructure rental.

• Capital structure remains the dominant risk: $35.6B debt at Q2 plus huge future capex requirements.
• High secured funding costs: several disclosed facilities carried effective rates around 9–15%.
• Dilution: the Sep 2033 convertible has an initial conversion price around $97.85 and the ATM permits up to 35M additional Class A shares; capped calls may mitigate but do not remove per-share financing risk.
• GPU/debt-duration mismatch: hardware can depreciate economically faster than debt and leases mature.
• Customer concentration/backlog quality: contracted revenue can carry delivery/performance conditions and is not equivalent to risk-free cash.
• Refinancing/capital-market dependence: access and pricing can change quickly.
• Residual-value uncertainty: lower mature-GPU values can destroy equity IRR even if headline utilization remains high.
• Supplier and power concentration: NVIDIA availability, facility delivery and power access remain critical.
• AI-capex factor concentration: CRWV would add directly to PowerFund's dominant portfolio factor.

Warning — no capital

• Fully diluted financing needs continue rising without a visible decline in net leverage/interest burden.
• Net debt / annualized revenue fails to trend toward <2.5x as contracted capacity starts.
• Interest remains >20% of revenue after scale-up.
• Mature GPU recontracting economics weaken materially.
• Backlog growth requires progressively worse project financing terms.

Reduce / pass

• Repeated equity/convertible issuance offsets a material share of per-share operating growth.
• Project IRRs under conservative residual-value assumptions fail to clear the blended cost of capital by a meaningful margin.
• +300bp refinancing or 30–40% lower residual values makes base-case common-equity returns unattractive.
• Major customer renegotiations/cancellations materially reduce contracted economics.

Invalidate

> Invalidate if CoreWeave cannot convert exceptional contracted demand into improving fully diluted common-equity economics after financing, depreciation and hardware obsolescence.

Hard invalidators include refinancing failure, major-customer default/loss, or evidence that contracted project returns are structurally below the company's blended cost of capital.

CoreWeave's advantages are specialization, speed and dense GPU-cluster operating expertise. Its disadvantages are financing cost and customer concentration.

• AWS/Azure/GCP/Oracle: lower capital costs, broader ecosystems, deeper balance sheets.
• Nebius/IREN/other neoclouds: compete on speed, power access and customer-specific capacity.
• CoreWeave: strongest current proof of large-scale specialized GPU cloud demand and backlog.

The potential moat is not the GPU itself; it is the combination of power access, deployment execution, orchestration software and customer relationships. The equity moat is weaker because much of the asset base is financeable/replicable if capital is abundant.

1. Build the debt/lease/SPV maturity table using the Q2 10-Q plus September financings; distinguish secured project debt, unsecured/convertible debt, leases and undrawn commitments.
2. On settlement/use-of-proceeds disclosure, quantify how much of the $3.7B Sep convertible actually refinances higher-cost debt versus funding new capex/operations; include the capped-call cost.
3. Track utilization of the 35M-share ATM and fully diluted shares including all convertible series, SBC and equity issuance.
4. Match major customer contracts against GPU generation, facility, capex, delivery date and debt/lease duration; estimate project IRR after interest, depreciation and conservative residual values.
5. Track mature Ampere/Hopper utilization and recontracting price as the empirical test of residual-value assumptions.
6. Model common-equity downside under +300bp funding cost and 30–40% lower residual hardware value.
7. Compare CRWV common-equity economics directly with NBIS/IREN and hyperscaler alternatives before considering capital.

Next full review: after Q3 2026 results / next 10-Q, or earlier on a material financing, customer-contract or refinancing event.