Nvidia China Probe and Cheap Agent Models Test Compute and Software

The setup — A board investigation into Nvidia ($NVDA) CEO Jensen Huang's relationship with an alleged Chinese intelligence operative complicated the chipmaker's push to loosen US export restrictions during Monday's session, straining the case for our AI compute build-out theme. Meanwhile, Meta ($META) launched an entry-level model priced at 10 cents per million input tokens, strengthening our AI software and agents theme by driving down the cost of running automated software. The evidence splits between hardware governance risks and falling software input costs.

What changed for the themes

Our AI software and agents theme gained support on Monday from fresh evidence that falling inference prices are commoditising model access. Meta launched Muse Ember, an entry-level model built to execute agent tasks at 10 cents per million input tokens, which undercuts rival lightweight models such as Haiku and Luna. Meanwhile, Anthropic released its frontier Claude Fable 5.2 and Mythos 5.2 models during Monday's session (per The Verge), and after the close set an IPO price range of $70 to $78 per share (per CNBC) that could raise up to $63 billion at a valuation above $1 trillion (per MarketWatch). Cheaper token prices and faster frontier models lower the cost of building software agents for companies that sit where work is done, including Salesforce ($CRM), ServiceNow ($NOW) and Adobe ($ADBE), all three of which our readings flag as cheap against their own history.

Our AI compute build-out theme faced a new governance and export-policy complication even as semiconductor production expands. Before Monday's open, Nvidia's board opened an investigation into CEO Jensen Huang's affair with Alice Huang, an executive at Singapore-based chip buyer Megaspeed whom US intelligence reports identified as a suspected Chinese spy (per CNBC and The Verge). Megaspeed has previously been probed for allegedly supplying Nvidia chips to China (per CNBC), and the scandal threatens to complicate Nvidia's lobbying against US export controls (per TechCrunch), bearing directly on whether export rules close markets. After Monday's close, President Trump said the FBI is looking into Huang (per CNBC), while Taiwan Semiconductor Manufacturing ($TSM) reported a jump in October sales early Tuesday (per CNBC).

Our robotics and physical AI theme saw a regulatory hiccup on Monday alongside ongoing production expansion. Federal safety regulators reported on November 9 that Waymo robotaxis continued to illegally pass stopped school buses despite a software update rolled out nearly a year earlier to fix the issue. Our dashboard still reads the evidence on whether safety concerns are holding back deployments as too thin to call, leaving the broader theme supported by factory and vehicle production.

Nothing material changed in Monday's session for our AI platforms and cloud theme or our AI power and critical minerals theme.

Risks to watch

Export rules that close foreign markets and leadership turmoil now test the compute build-out directly. With the FBI looking into Jensen Huang, a class-action lawsuit emerging (per CNBC), and Nvidia lacking a clear successor (per CNBC), any tightening of US export rules or restrictions on buyers like Megaspeed would confirm the threat to chip sales.

Credit tightening for debt-funded cloud builders also strains both compute and cloud platforms, following last week's neocloud delays and voter pushback. By our dashboard, hyperscaler capital spending is running 107.6% above last year — 2.22 standard deviations above its 10-year average, a level we label saturation — while high-yield credit spreads have widened 27 basis points over the past month to 342 basis points. That credit strain hits builders differently depending on how they pay for capacity. Nvidia holds more cash than debt and turns 40% of revenue into free cash flow after stock pay, whereas CoreWeave ($CRWV) spent 298% of its operating cash flow on capital expenditures over the past year while its long-term debt rose 185%, and Oracle ($ORCL) is burning cash with rising debt. Even though Goldman Sachs CEO David Solomon said on Monday that investor appetite for AI infrastructure financing remains strong (per CNBC), CoreWeave's bond yields above 16% show the strain is already real.

Finally, our readings flag company-specific warnings: Super Micro Computer ($SMCI) carries warning language in its filings, EQT ($EQT) faces declining revenue, and USA Rare Earth ($USAR) shows heavy share dilution.

The companies

Nvidia ($NVDA), which rides our AI compute build-out theme, fell 4.3% to $178.67 on Monday. That drop was about 3.9 times its typical daily move after allowing for the broader rally across our tracked companies, following the pre-market news that its board opened an investigation into CEO Jensen Huang's relationship with an alleged Chinese spy (per CNBC). During the session, Huang also pulled out of the ElevenLabs Summit after earlier withdrawing from an Anthropic event (per TechCrunch). Monday's selloff contrasts with our valuation reading: Nvidia sits at the 10th percentile of its own valuation history, priced for about 15% annual cash-flow growth after delivering 94% a year over the past three years, though its price trend is still falling and a few customers buy much of what it sells.

CoreWeave ($CRWV) rose 5.6% to $48.98 and Nebius ($NBIS) gained 9.2% to $141.45. Before Monday's open, coverage highlighted that both neoclouds had lost about half their value over the past month amid credit risk concerns ahead of earnings (per CNBC), and nothing in the day's filings explains their Monday rebound. Our readings on both remain adverse: CoreWeave carries warning language in its latest filing, burns cash equal to 188% of revenue after stock pay, and saw operating profit fall short of interest expense that reached 21% of revenue, even as contracted backlog grew 245%.

Several other closely watched companies rose without a specific catalyst in the day's filings or coverage. In robotics, Aeva Technologies ($AEVA) climbed 11.8% to $12.30; it filed a 10-Q after Friday's close, though we do not have its text, and our readings flag cash burn of 601% of revenue after stock pay. In our AI power and critical minerals theme, Energy Fuels ($UUUU) rose 9.8% to $7.31 — filing an 8-K with undisclosed contents after Monday's close — while GE Vernova ($GEV) gained 7.1% to $829.34, Denison Mines ($DNN) added 6.9% to $2.34, and USA Rare Earth ($USAR) rose 6.5% to $8.76.

The backdrop

Monday's session was a broad rally everywhere outside of Nvidia. Across the 67 companies we track, 63 rose and only 4 fell, with the median stock up 5.6% led by Coherent ($COHR) up 16.8%, Astera Labs ($ALAB) up 13.7% and Advanced Micro Devices ($AMD) up 13.3%. The S&P 500 ETF ($SPY) gained 2.4%, the Nasdaq 100 ETF ($QQQ) rose 3.5% and the Nasdaq Composite added 2.7%. The Semiconductor ETF ($SMH) climbed 5.1% despite Nvidia's drop, while the Software ETF ($IGV) rose 2.6% and the Technology Sector ETF ($XLK) gained 3.7%. Volatility retreated as the $VIX fell 22.3% to 19.92, the 10-year Treasury yield held flat at 4.86%, the US dollar index slipped 0.6% to 101.65, and WTI crude fell 4.4% to $86.85.

What to watch

• CoreWeave ($CRWV) earnings on Tuesday, November 10, after the close: Results and financing commentary will test whether tightening credit for debt-funded cloud builders is slowing capacity expansion or forcing higher borrowing costs.
• Applied Materials ($AMAT) earnings on Thursday, November 12, after the close: Wafer-fab equipment orders and China revenue disclosures will test our AI compute build-out theme and show how export rules are affecting toolmakers.
• Nvidia ($NVDA) board investigation and November 17 earnings: Any board findings, FBI actions, or changes to US export policy following the Megaspeed and Jensen Huang probe ahead of next Tuesday's results.
• Anthropic's IPO pricing and reception: Expected next week at $70 to $78 a share (per CNBC and MarketWatch), which will test public-market appetite for frontier model builders carrying $895.4 billion in purchase commitments.