Kerrisdale Capital Management
Sahm Adrangi (founder, CIO)
Activist short seller that publishes detailed teardown reports; in this cycle the marquee call is CoreWeave — a 'debt-fueled GPU rental business with no moat, dressed up as innovation' that it argues is the poster child of the AI-infrastructure bubble.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- No public record — No published return series — performance cannot be verified in either direction.
- Partial 13F — The disclosed book is longs-only and ~45 days lagged — a slice, not the strategy.
Short positions
SHORT — Sept 2025 report; argues ~90% downside (fair value ~$6–13/share) on debt load ($11B), 11–15% asset-backed financing rates, and anchor-customer concentration
Recent moves
Published its CoreWeave activist short in Sept 2025, calling it 'the poster child of the AI infrastructure bubble' with ~90% downside on debt-fueled, moat-less GPU-leasing economics and customer concentration.
Loading disclosed positions…
Our take
Kerrisdale's bear case is the cleanest single-name expression of the AI-infra-debt worry — and it's testable: CoreWeave's $3.6B due by June 30 2026 is a near-term catalyst. The risk to the thesis is the same one that's burned every AI short — anchor-customer mega-deals (e.g. hyperscaler contracts) can refinance the balance sheet faster than the debt math implies.
The most useful public artifact is the 13F, though not for the obvious reason.
Kerrisdale Advisers' Q1 2026 filing (15 May 2026) shows 112 positions worth $262.7M — top names Visa, MercadoLibre, TDS, Amazon. The firm's own site states ~$250M AUM as of Mar 2026, but that page returned HTTP 403 at check time and is unverified; taken at face value the disclosed long book exceeds stated firm AUM, which more likely means the AUM figure is stale or scoped differently than that the longs are levered. Either way the book reads as generic large-cap beta — ~$2.3M average across 112 names, consistent with collateral, borrow and hedging, not conviction alpha. The claimed edge therefore sits entirely in the undisclosed short book: no entry price, no size, no cover date. An activist short monetises the publication-day gap; nothing obliges the manager to hold to target and nothing discloses when they stopped. A reader scoring the thesis by today's price may be scoring a position exited months ago.
- 2026-08Two later published shorts follow CoreWeave — Affirm (Jan 2026) and Everspin (May 2026) — so CoreWeave is no longer the shop's most recent public thesis, and judging the process on that call alone is incomplete.
Thesis
Kerrisdale is a publish-and-move activist short shop: Sahm Adrangi's edge is manufacturing a repricing event on a crowded, narrative-priced name rather than forecasting terminal value. The CoreWeave call (Sept 2025, fair value $6-13, ~90% downside) is the cleanest public expression of the AI-infra-debt worry — and the clearest test of the model's weakness. CRWV traded $78.72 on 21 Jul 2026 against a 52-week range of $63.80-$153.20, never approaching the target.
Concentrated activist short selling with a published research report as the catalyst, alongside a long book. History runs from the 2010-11 China reverse-merger frauds (China-Biotics, Lihua, China Marine Food) through C3.ai (2023), IonQ (Mar 2025), CoreWeave (Sept 2025) and Bitmine Immersion (Oct 2025). Pattern: target a story stock where the equity is a residual claim behind debt or dilution, then publish the arithmetic.
- 2026-07CoreWeave has repriced below the cited levels: it closed $71.77 on 31 Jul 2026 after a new all-time intraday low of $60.55 on 29 Jul, putting the 52-week range at $60.55-$153.20.
Assessment
- The mechanism in the CoreWeave report is specific and checkable — debt service, GPU depreciation, customer concentration — not vague 'bubble' rhetoric.
- Deep pattern-recognition franchise: the 2010-11 China fraud calls established a documented skill at spotting accounting and structure fictions.
- Targets share a genuine structural feature — equity behind heavy debt or dilution — which is a repeatable, not ad hoc, screen.
- Publishes the full argument openly, so the thesis is falsifiable in public rather than asserted from a marketing deck.
- The incentive is publication frequency, not thesis accuracy. A report that produces a one-day drop pays even if the terminal call is wrong — these are not the same skill, and only one is observable.
- No disclosed fund returns at all. The verifiable artifact is the report, not a track record; readers substitute report notoriety for performance and that substitution is unearned.
- The 13F long book (V, MELI, AMZN) is indistinguishable from beta. If it is a meaningful share of firm capital, part of any return is market exposure, not short-side skill.
- Shorting AI infrastructure into a capex supercycle has repeatedly been early-equals-wrong. CRWV's 52-week range of $63.80-$153.20 shows the mark-to-market pain a report-day short faced before any drawdown.
- Single-key-person firm. Adrangi is the research voice, the media presence and the CIO; there is no visible succession or second-signature process.
- 2026-07The drawdown arrived: CoreWeave fell 27.9% in July 2026 ($99.54 to $71.77) against the S&P 500 at -0.1%, alongside Nebius -31.1%, and now sits 53% below the $153.20 high.
Record
Not disclosed — Kerrisdale publishes no fund-level returns, so nothing here is a performance claim. What is checkable: the CoreWeave short was published Sept 2025 with a $6-13 fair value; shares rose on the report's release. CRWV later closed $85.69 on 1 Jul 2026, down ~13.9% that day after a Bloomberg report that Meta plans to launch a commercial cloud service built on its own GPU fleet — precisely the hyperscaler-competition risk the report named. It then traded $73.21 on 17 Jul, $77.93 on 20 Jul and $78.72 on 21 Jul 2026, against a 52-week range of $63.80-$153.20. Direction partially vindicated; magnitude and timing not — the stock never approached $6-13, and a short held from publication would have absorbed a large mark-to-market run-up first. Attribution across the wider book is unknowable from public data. Figures as of 21 Jul 2026.
- 2026-07On timing the record improved: the repricing landed about 10.5 months after the 15 Sep 2025 report and the run-up has fully unwound. On magnitude it has not — $71.77 is still 5.5-12x the $6-13 target.
Risks & fit
- Short squeeze and borrow cost: a crowded short in a retail-heavy AI name can force cover well before the fundamental case resolves.
- Refinancing risk to the thesis, not the company — an anchor hyperscaler contract can term out debt faster than the depreciation math implies.
- Capacity: a sub-$500M asset base limits position size, so returns depend on hit rate rather than scaling a winner.
- Reputational reflexivity — after a high-profile call that does not converge, later reports move stocks less, degrading the core mechanism.
- Legal and regulatory exposure is a standing feature of activist short publishing in every jurisdiction it operates in.
The thesis class — neoclouds as financially engineered residual equity behind depreciating hardware — is falsified if CoreWeave services debt out of contracted revenue rather than new capital raises, holds gross margin as first-generation GPUs roll off, and diversifies beyond its top customers, while trading durably above the Sept 2025 report price. That would mean the mechanism is wrong, not merely early. It is confirmed if a maturity is met only by refinancing on materially worse terms — no specific 2026 maturity event is verified here. Watch contracted-revenue-to-debt-service coverage, not the share price.
Readers analysing the AI-infrastructure credit cycle who want the bear mechanism laid out with specific arithmetic, and anyone studying how activist short research actually monetises — the gap between a report's market impact and its terminal accuracy. It is not a window into a verifiable return stream: the fund discloses none, the 13F shows only a longs-only, ~45-day-lagged US equity slice, and the short book that defines the firm is invisible in filings.
Not disclosed. Kerrisdale is a private fund and publishes no fee schedule; activist short funds typically carry a management fee plus performance fee, but no figure for this firm is public and none should be assumed.