Nvidia beats in Q3, raises Q4 outlook - stock dips as AI demand stays "sold out"
Nvidia slipped 3.15% on Thursday even after a clean beat and stronger guide. It wasn't alone - the rest of the "Magnificent Seven" faded with the broader market. Call it a risk-off day, not a fundamentals problem.
Q3 snapshot
- Revenue: $57.01B vs. $55.2B consensus
- EPS: $1.30 vs. $1.26 consensus
- Year-ago: $35.1B revenue and $0.81 EPS
- Data center: $51.2B vs. $49.3B consensus
- Gaming: $4.3B vs. $4.4B consensus
Q4 outlook
Nvidia guided $65B ±2% for Q4 revenue versus ~$62B expected. Management's tone was confident: "Blackwell sales are off the charts, and cloud GPUs are sold out," CEO Jensen Huang said. CFO Colette Kress added that Blackwell Ultra is now the leading architecture across customer categories, while demand for the prior Blackwell generation remains strong.
One notable line item: revenue from the China-specific H20 chip was "insignificant." That helps frame the demand picture as primarily ex-China for now.
Why the stock fell anyway
- Macro tape: Broad tech weakness pressured the group; early gains faded.
- Positioning: After a run where market cap briefly topped $5T, even strong prints can trigger profit-taking.
- Supply constraints: With cloud GPUs "sold out," upside is capped by availability, not demand. Good problem, but still a cap near term.
Accounting debate: depreciation
Investor Michael Burry criticized large AI spenders for allegedly understating data center depreciation, which could inflate earnings. Kress pushed back on the call, pointing to CUDA extending the useful life of accelerators - even 6-year-old A100s remain productive thanks to software updates.
For finance teams, the takeaway is simple: expect continuing scrutiny on useful lives, residual values, and refresh cycles across the AI stack. Model sensitivity around service lives can materially swing EBIT and FCF.
Capital flows and competitive context
- Shareholder moves: Peter Thiel's hedge fund reportedly exited a ~$100M NVDA stake pre-earnings; SoftBank sold ~$5.8B to fund its own AI bets.
- Rival narrative: AMD CEO Lisa Su pegs the data center market at up to $1T by 2030. AMD shares are up 82% YTD and 58% over 12 months; Nvidia is up 37% YTD and 25% over 12 months.
What to update in your model
- Topline: Move Q4 to the midpoint of $65B ±2%; maintain a supply-constrained posture for cloud GPUs.
- Mix: Keep data center as the primary driver; gaming slightly below consensus is not thesis-changing.
- Capex pass-through: Track hyperscaler capex guides - they set the ceiling on accelerator demand timing.
- Depreciation: Revisit useful life assumptions for GPUs and systems; software-enabled longevity can reduce depreciation drag versus prior cycles.
- China: Treat H20 as minimal near-term; focus on US/EU/ROW deployments.
- Positioning risk: After the $5T milestone, sentiment and flows matter; expect higher volatility around prints and macro data.
Risks to watch
- Supply bottlenecks limiting shipments despite demand.
- Customer concentration among top cloud buyers.
- Export restrictions and regional product variants diluting mix.
- Accounting scrutiny on depreciation and useful lives across the ecosystem.
- Competitive cadence from AMD and custom silicon.
Bottom line
Demand isn't the issue - supply is. Nvidia beat, raised, and still sold off on a market-wide downdraft and elevated expectations. For now, the AI flywheel keeps spinning: "AI is going everywhere, doing everything, all at once," as Huang put it. Model shipments to availability, not appetite.
Your membership also unlocks:
