Contrarian Corpus
short seller research note initial thesis
2025-09-01 · 30 pages

CoreWeave, Inc. CRWV

CoreWeave is a debt-fueled, undifferentiated GPU rental stopgap with 71% Microsoft concentration and sub-WACC returns; fair value is $6–13, or 88–95% downside.

Thesis

Kerrisdale is short CoreWeave (CRWV), arguing the 'AI hyperscaler' narrative masks an undifferentiated, heavily levered GPU rental business propped up by tactical hyperscaler contracts. Microsoft (71% of 2Q25 revenue) already declined a $12B expansion option and signed its next deal with neocloud rival Nebius, confirming Satya Nadella's 'one-time thing' framing. CoreWeave carries ~$20B net debt, is set to burn $19B of cash in 2025 and over $40B through 2028, and earns just 2% ROIC — below its cost of capital. Its sell-side 'illustrative' contract assumes utopian 80% EBITDA margins and 75% residual value from five-year-old GPUs, assumptions a former engineer dismissed as 'totally bullshit.' On 8x 2028E EBIT or 0.9x book value, fair value is $6–13 per share, implying 88–95% downside.

SCQA

Situation

CoreWeave is a publicly traded 'neocloud' that rents Nvidia GPUs to hyperscalers under multi-year take-or-pay contracts, trading near $112 at a $75B enterprise value after a 30% one-week rally.

Complication

Microsoft (71% of revenue) declined a $12B expansion and signed rival Nebius, exposing CoreWeave as an interchangeable stopgap; $40B of debt and aggressive depreciation flatter sub-WACC returns on rapidly obsolete GPUs.

Resolution

Short CRWV. Ignore headline growth and value the company on commodity rental economics: 0.9x book value or ~8x 2028E EBIT, reflecting thin margins, weak moat, and commodity competition from DGX Lepton and AMD.

Reward

Fair value of $6–13 per share versus ~$112 current, implying 88–95% downside as hyperscalers internalize compute, custom silicon proliferates, and the debt-fueled growth narrative collapses.

The three reasons

  1. 1

    Microsoft (71% of revenue) declined expansion and signed rival Nebius — a 'one-time thing'

  2. 2

    CoreWeave will burn $19B cash in 2025 and carry >$40B debt by 2028 on 2% ROIC

  3. 3

    'Illustrative' contract assumes utopian 80% margins and 75% residual value from five-year GPUs

Primary demands

  • Short CRWV; reprice to commodity GPU-rental economics reflecting sub-WACC returns

KPIs cited

Customer concentration
Microsoft = 71% of 2Q25 revenue; 77% of YoY revenue growth; 70% of all revenue growth since launch
ROIC
2% in 2024A; projected 1–5% through 2028E — below cost of capital
Net debt
~$20B in 2025; projected >$40B total debt by 2028
2025 cash burn
$19B expected free cash burn in 2025; $40B cumulative through 2028
Net debt / EBITDA
Peaks ~6.0x in 2025
Unit economics EBIT margin (GB200 NVL 72)
20.5% at 6yr useful life → 12.4% at 5yr → 0.1% at 4yr
Debt service coverage ratio
Only 1.21x in Year 1 of illustrative contract — thin even take-or-pay
Valuation
0.9x 2025E book → $6/share; 8x 2028E EBIT → $13/share

Pattern membership

Where this document fits across the library's 12 rhetorical / structural patterns. Orange cells are present in this deck; neutral cells are not.

Composition what's on the 30 slides

Visual + textual elements counted across every slide in this deck. Hover a box for what that element is; click to see every slide in the corpus that uses it.

Slide gallery ·

All 30
No slide inventory yet

Pass-2 extraction may still be in progress for this deck.

Notes

Classic Kerrisdale short-report format: text-heavy research note with embedded tables/charts, not a slide deck. Title 'Artificial Returns' is the rhetorical hook. Key rhetorical moves: (1) Satya Nadella 'one-time thing' quote used as central CEO-quote contradiction; (2) before/after cash flow waterfall (p.21 CoreWeave's 'illustrative' contract vs p.23 Kerrisdale's redrawn version); (3) margin-cliff table on p.12 showing EBIT collapsing from 20.5% → 0.1% as GPU useful life assumption shortens. No named human author on cover — only 'Kerrisdale Capital, September 2025'. presentation_date set to 2025-09-01 as month-only placeholder since exact day not disclosed. Not a fraud report; thesis is structural overvaluation + levered commodity economics, mapped to multiple_rerating + other.