Burry Dumps Anthropic Stake, Doubles Down on Nasdaq Shorts While Adding JD, Tencent, and BYD
nashnova research
Hedge fund manager Michael Burry publicly passed on Anthropic's IPO, doubled his Nasdaq and semiconductor short positions out to 2027, yet bought JD, Tencent, and BYD — betting that on-device small models will siphon cloud compute demand and compress AI valuations sooner than the market expects.
What happened on Anthropic's first day of trading?
Anthropic priced at $83, opened at $111.50, and closed at $97.17 — up 17% from the IPO price but down nearly 13% from the open.
The broader market sold off hard: the Dow fell 1,101 points, the S&P 500 posted its sixth straight losing session, and the Philadelphia Semiconductor Index sat 35.63% below its all-time closing high — deep in a technical bear market.
This means → The largest AI IPO in history failed to lift sentiment in the compute sector; it landed at the market's most fragile moment.
Why did Burry skip Anthropic and add to his shorts?
Burry publicly passed on the deal during the roadshow, writing on Substack that the offering was "visibly pumping index futures." He would neither subscribe nor buy after listing.
On November 12, he doubled his Nasdaq 100 index put position — strikes roughly 20% out of the money, expiring in early 2027 — after the index bounced near 27,000. He compared the October 6 Nasdaq 100 peak to the March 2000 dot-com top.
He also holds Nvidia puts at a $140 strike expiring February 2027 and repurchased VanEck Semiconductor ETF puts he had previously trimmed.
In plain terms = He is not shorting one company. He is betting the entire AI compute chain — chips to index — weakens over the next eighteen months.
Nvidia's earnings looked strong — what caught Burry's eye?
Nvidia's quarterly revenue rose 103% year-over-year to $115.8 billion, but Burry focused on how fast the cash comes back: accounts receivable hit $81 billion, and days-sales-outstanding climbed from 60 to 64 days.
Nvidia attributed this to Rubin chip shipment timing and longer payment terms for certain investment-grade customers.
This means → Revenue is growing, but collections are slowing — Burry reads lengthening payment terms as a possible precursor to order deferrals.
What does "on-device small-model diversion" actually mean?
Burry's core thesis: even as enterprises expand AI adoption, locally deployed small models will siphon inference tasks — the actual work of running a model to produce answers — that would otherwise require paid calls to cloud-hosted frontier models.
Example: a law firm retrieves clauses from internal archives with a local model, then sends only the most complex cases to a cloud frontier model. Nvidia has already used roughly 2-bit quantization to run a 74-billion-parameter mixture-of-experts model on a device with just 24 GB of memory.
In plain terms = More and more routine tasks no longer need expensive cloud compute — and that directly undercuts the revenue assumptions of compute suppliers.
What risk does Anthropic itself face?
Anthropic's prospectus discloses procurement and lease commitments of roughly $900.9 billion, almost entirely for compute and infrastructure. Of that, $70.7 billion comes due within twelve months after September 2026.
Revenue is generated mainly on a pay-per-use basis. If high-frequency simple tasks are diverted to local models, the remaining call volume and unit price may not cover these fixed obligations.
This means → If utilization rates and lease prices fall in tandem, the first to feel cash-flow pressure will not be chip makers but data-center operators servicing GPU-backed debt — and their private-credit lenders. Burry noted that leverage in the AI financing stack reaches as high as 76×.
If he is shorting AI, why is he buying Chinese stocks?
On the same day, Burry disclosed seven buy trades, including additions to JD.com, Tencent, and BYD — all Hong Kong-listed.
JD.com: He estimates each ADR corresponds to nearly $20 in cash and marketable securities, with earnings per share on track for $3. The buy case rests on the cash cushion plus earnings power.
Tencent: He added during the post-earnings pullback, citing its market position and hard-to-replicate business assets. Excluding new AI products, operating profit rose 8% year-over-year to ¥83.3 billion, but advance payments for compute procurement pushed free cash flow to negative ¥69.1 billion. Burry had previously paused buying on the negative FCF; the price drop changed his calculus.
BYD: He added at a low in the stock, citing its long-term competitive edge over rival automakers.
This reflects a clear framework: Burry is not bearish on everything — he is cautious on the AI capex cycle but willing to own businesses that can be valued on existing operations once the price falls far enough. The key verification point is how much paid inference actually stays in the cloud. If on-device models divert traffic faster than expected, compute-stock valuation compression will arrive ahead of most investors' timelines.
市场有风险,内容仅供研究参考,不构成投资建议。
