Nvidia's $96 billion quarter is real but a physicist warns some demand may be propped up by financial engineering

Nvidia just posted one of the most extraordinary quarterly earnings reports in corporate history. Revenue of $96.2 billion, up 106% year on year. Net income of $59.7 billion. Forward guidance of $108 billion for the next quarter. And one of the sharpest AI researchers in the room is not impressed. He is worried.

The circular financing warning

Alexander Wissner-Gross, a physicist and AI researcher, raised a question on the Moonshots podcast that cuts beneath the headline numbers. How much of Nvidia's explosive demand is genuine and how much is circular, kept alive by Nvidia itself deploying billions of dollars to ensure its own customers can afford to buy its products?

Nvidia has been deploying capital in checks ranging from $5 billion to $20 billion to ensure its customers can keep buying chips. The loop Wissner-Gross described is straightforward: Nvidia posts record profit, deploys billions in customer financing, customers buy Nvidia GPUs with that capital, Nvidia books the revenue as organic growth and Wall Street prices in continued expansion.

"I would not like to discover two to three years from now that some or a large portion of all of this demand at the infrastructure layer was being artificially propped up through financial engineering," he said.

Is circularity a problem or a feature?

Not everyone agrees. AI investor Dave Blundin pushed back, arguing that every economy is circular by nature and that the AI economy is becoming self-sustaining in a way that makes comparisons to traditional market dynamics beside the point.

Even Wissner-Gross stopped short of predicting a crash. "If there is a correction, it will be a mini winter, not a full winter," he said, because compute is too fundamental to the global economy to collapse entirely. Nvidia's own guidance supports the optimists, with the company projecting roughly 70% sales growth in fiscal 2028 against Wall Street's 44% consensus.

Nvidia's real vulnerability: TSMC

The podcast identified Nvidia's most significant structural weakness as its near-total dependence on Taiwan Semiconductor Manufacturing Company. Nvidia represents roughly one-third of TSMC's output, an arrangement that constrains its ability to diversify manufacturing without risking its most important supplier relationship.

Salim Ismail, a former Singularity University executive, argued Nvidia's strategic response is a pivot from chipmaker to operating system for artificial intelligence, potentially including an acquisition of Hugging Face, the open-source AI model platform, in a move he compared to Microsoft's purchase of GitHub.

Blundin added that the real new moat is not software but interconnect. Nvidia's acquisition of Mellanox gave it control over the high-speed networking needed to bind together GPU clusters of 100,000 to several million units, a hardware advantage that AMD, Google and Amazon cannot easily replicate.

OpenAI's Jalapeno chip changes the race

Competitive pressure is building at the inference layer. OpenAI's new custom chip, Jalapeno, developed with Broadcom, reportedly delivers between 1.5 and 1.9 times more AI work per watt and up to 3.6 times lower latency than Nvidia's current architecture, at half the power consumption. A claimed 54-times throughput increase was cited on the podcast, though the figure has not been independently verified.

If Jalapeno's performance claims hold, OpenAI would have strong economic incentive to rent out cheap inference capacity to competitors, potentially becoming a hyperscaler that serves its rivals.

America vs China: two roads to AGI

Wissner-Gross offered a sharp framing of the US-China AI divergence. American labs are revenue-maximizing while Chinese labs are not. OpenAI pivoted away from video generation because code commands higher revenue per token. Chinese labs, facing enterprise trust barriers, pivoted to video, a global product with no security concerns. Video generation now represents approximately 70% of all AI token consumption in China, growing faster than code generation grew in the United States.

The strategic question is which path reaches artificial general intelligence first. America optimizes for better code, which could accelerate recursive self-improvement. China optimizes for world modeling, which could accelerate embodied AI and robotics.

Wissner-Gross's financial engineering warning sits uneasily against the collective optimism of a group that holds significant positions in AI infrastructure. But even he concedes a correction would be brief. The investment question is not whether to hold AI infrastructure but which companies have genuinely diversified across the stack and which are, as Blundin said of Apple, coasting on past glory.