Nvidia reported $96.2 billion in Q2 fiscal 2027 revenue on August 26. Up 106% year-over-year. Data center: $89 billion. EPS $2.22. Gross margin 75%. Jensen Huang on the call: "Compute is revenue." He then guided fiscal 2028 at approximately 70% revenue growth — supply-constrained, demand "much higher" than what the company can ship.

Supply-constrained. At $96 billion a quarter. Worth sitting with for a second.

Thirty-six hours later, the Wall Street Journal reported Nvidia had quietly paused its $500 billion cloud-financing program. Not ended — paused. The program launched in July 2026. Nvidia arranged credit backing from major U.S. financial institutions for AI cloud providers in exchange for a cut of their revenue, plus a $105 billion guarantee to help OpenAI lease computing infrastructure. It ran for fewer than two months before someone pressed stop.

And on August 27, the same day as the pause report: Nvidia agreed to buy Hugging Face for $12.9 billion.

Three things in 72 hours. They're not separate stories.

Source spread

• NVIDIA Newsroom — Q2 FY2027 earnings [hype] — Official press release. $96.2B revenue, $89B data center, 75% gross margin, EPS $2.22.
• WSJ via WKZO — Financing pause [skeptic] — First to report the cloud-financing pause. Attributes the stop to employee antitrust concerns and "how much control the company could exert over customers."
• Yahoo Finance — Financing details [skeptic] — Detailed breakdown of the circular-deal structure and the $105B OpenAI compute guarantee.
• CNBC — Hugging Face acquisition [builder] — Reports Nvidia agreed to buy at $12.9B. Context on why the company turned down a $500M Nvidia offer at $7B in 2025.

Pros & cons

What's real:

• The $96.2 billion quarter is audited. The 106% YoY growth is real, and at this scale, additional percentage points represent billions in new revenue. The 70% fiscal 2028 guidance is Huang's, not a Wall Street estimate — he's rarely wrong about his own supply picture.
• Pausing before regulators act is the correct move. The circular-deal logic is clear: Nvidia finances cloud companies → those companies use the money to buy Nvidia chips → Nvidia profits → finances more companies. That's exactly the structure the DOJ and FTC have learned to notice. Nvidia's own employees flagged it. Getting out ahead is competent legal risk management, not a strategic retreat.
• Hugging Face is the cleaner version of the same ambition. Buying the open-model distribution layer doesn't carry circular-revenue-sharing logic. The concern shifts from antitrust to neutrality — and neutrality is a slower, messier problem.

What deserves scrutiny:

• "Supply-constrained with demand much higher" is Jensen Huang framing his own market. It positions the limit as a production ceiling rather than a demand ceiling. Convenient and hard to verify from outside.
• The HF acquisition closes the circle differently: Nvidia sells chips, arranges compute financing (remaining deals), owns the model-hosting layer, takes inference revenue. That's vertical integration. It won't trigger the circular-deal concern, but it will attract regulatory attention of a different flavor — eventually.
• Hugging Face's value to builders has been neutrality. Google's models, Meta's models, Anthropic's weights — all live there on equal terms. That's the product. Nvidia buying it changes the incentive structure on day one, even if the policy doesn't change.

What builders need to know

• The Hugging Face acquisition is not closed. Models, APIs, and hub behavior are unchanged today. Don't emergency-migrate anything.
• If you depend on Hugging Face for model hosting, the Transformers library, or inference — watch pricing and default options over the 12 months after the deal closes.
• The $500B financing pause doesn't affect API pricing or builder access to Nvidia hardware directly. It was a revenue-sharing arrangement between Nvidia and cloud providers.
• Huang's 70% fiscal 2028 growth guidance implies continued GPU scarcity. If you're planning compute capacity for 2027, model in the assumption that prices don't drop significantly.
• Benchmark the "supply-constrained" framing against your actual experience getting GPU allocation. If you're not supply-constrained, the constraint isn't at your layer.

Further reading

• NVIDIA Q2 FY2027 press release — the audited numbers
• CNBC — Nvidia doubles Q2 revenue — earnings call context and Jensen Huang's guidance statements
• CNBC — Nvidia agrees to buy Hugging Face for $12.9B — acquisition report
• Yahoo Finance — Nvidia pauses AI cloud financing — financing program details and circular-deal concerns
• Our earlier piece on the Hugging Face sale talks — background on what HF is and why the neutrality question is the whole thing