On September 10, Nvidia CEO Jensen Huang took the stage at the Goldman Sachs Communacopia+ Technology Conference in San Francisco and named cybersecurity as AI's next major application scenario. His logic was direct: AI is automating computer programming, which means code is being written faster, exploited faster, and patched faster. The entire supply and demand structure of cybersecurity is shifting beneath the industry's feet.
“What better way to create demand than by manufacturing a problem?” Huang told the audience.
He said it with a smile. The market heard it as a thesis.
The HBM That Isn't There: A Formula Built on Speed
The gap Huang describes is measurable, not hypothetical. According to CrowdStrike, AI-enabled attacks rose 89% in the past year, and the fastest eCrime breakout time has dropped to 27 seconds. At human speed, that's not defense — that's forensics.
The numbers explain why Nvidia moved before Huang spoke. On September 1, CrowdStrike unveiled SafeMind, an agentic cybersecurity system built on Nvidia's Nemotron open models. The system runs Red Tempest (offensive simulation) and Blue Solano (defensive response) in a continuous coevolution loop — two AI models training against each other until attacks fail.
The price-performance claim is the part worth noting. CrowdStrike's internal evaluation showed Blue Solano, built on Nemotron 3 Super, delivered higher accuracy than leading frontier models at 99% lower cost. It can be retrained on a single GPU in hours.
That's not a product announcement. That's an argument about where AI value accrues: not in the biggest model, but in the model trained on the right data. CrowdStrike brought 15 years of incident response telemetry. Nvidia brought the architecture to use it.
The Hugging Face Loop: $13 Billion and a Denial
The cybersecurity pivot is one half of Huang's message. The other half was a defense — one he didn't need to give but chose to.
On September 3, Nvidia announced it would acquire Hugging Face for approximately $13 billion, including a $1 billion retention program for employees joining Nvidia. The platform hosts 18 million developers, 3 million models, and 200,000 enterprise customers.
Investors immediately asked the obvious question: is this circular financing? Nvidia invests in AI companies, those companies buy Nvidia chips, Nvidia books revenue, Nvidia's stock rises, Nvidia invests again.
Huang's response: “This isn't circular, because we put in a little money and a lot more comes back”.
He added that he “doesn't take any risk,” because he only invests when customers are already queuing for the product.
The logic is seductive. It also collapses under light pressure. If Nvidia's investment decisions help create the demand that justifies the investment, the queue isn't evidence of independent demand — it's evidence of Nvidia's own gravity. The company is not just a chip supplier. It is, increasingly, its customers' banker.
Consider Nscale, a two-year-old British AI cloud firm. It is negotiating a $3.5 billion raise ahead of a U.S. IPO. Nvidia is expected to supply roughly $2 billion of that financing, even as Nscale has ordered 194,000 of Nvidia's upcoming Vera Rubin GPUs. The chip supplier becomes a major shareholder in the company placing the order, weeks before public investors are asked to back a valuation roughly double its last private round.
The structure isn't new — Nvidia used it with CoreWeave before its 2025 IPO. It has since applied the same approach to other infrastructure buyers including Nebius and Nokia.
Huang says he's not taking risk. But when your equity stake and your customer's purchase order are the same number, risk isn't the issue. Clarity is.
A Supply Chain Bet, Not a Product Launch
The cybersecurity thesis and the circular financing denial aren't separate stories. They are the same strategy viewed from different angles.
Nvidia's cuOpt decision-optimization library, Palantir's Ontology, and Cisco's AI Factory now form a three-way sovereign AI deployment, announced September 9. Nvidia is using Palantir's Ontology to manage component allocation across a global manufacturing network with millions of parts. The company fine-tuned its Nemotron 3.5 Lightning model for supply-chain workflows and found it outperformed its larger Nemotron 3 Ultra model for the specialized task.
The pattern: smaller, domain-trained models beating larger general models, running on infrastructure the customer controls.
This matters for the cybersecurity market because it suggests the winner won't be the company with the biggest model. It will be the company with the best data and the tightest feedback loop. CrowdStrike has the data. Nvidia has the architecture. Cisco has the enterprise deployment channel.
For AI infrastructure investors, the question Huang didn't answer is whether this strategy validates Nvidia's long-term demand — or masks it. If cybersecurity becomes a real market, AI compute gets a new revenue floor. If it's mostly a narrative, the circular financing critique gets louder, and vendor-financed IPOs start to look like a way to distribute risk rather than build it.
P.S. Nvidia reports quarterly earnings in November. Watch whether management breaks out cybersecurity-related revenue — or keeps it inside the data center segment.
Frequently Asked Questions
Q: What did Jensen Huang say about cybersecurity?
A: At the Goldman Sachs Communacopia+ Technology Conference on September 10, Huang said cybersecurity “very likely” will become AI's next major application scenario. He argued that AI is accelerating code generation, which means code is exploited and patched faster, fundamentally shifting the industry's supply and demand structure.
Q: What did Huang mean by “manufacturing a problem to create demand”?
A: Huang acknowledged that AI accelerates the creation of security vulnerabilities. He said there are responsible ways and less admirable ways to create demand, but didn't specify which approach Nvidia takes.
Q: Why is Nvidia acquiring Hugging Face?
A: Nvidia announced the $13 billion acquisition on September 3. Hugging Face hosts 18 million developers and 3 million models. The deal extends Nvidia's reach into open-source AI and includes a $1 billion employee retention program.
Q: What is circular financing, and is Nvidia doing it?
A: Circular financing refers to a supplier investing in customers who then buy the supplier's products, inflating apparent demand. Huang denied the practice, saying Nvidia only invests when customers are already queuing. However, Nvidia's expected $2 billion investment in Nscale — which has ordered 194,000 Nvidia GPUs — follows a similar structure to its CoreWeave deal.
Q: What is CrowdStrike SafeMind?
A: SafeMind is an agentic cybersecurity system built on Nvidia's Nemotron open models. It includes Red Tempest (offensive simulation) and Blue Solano (defensive response) in a coevolution loop. CrowdStrike claims 99% lower cost than frontier models with higher accuracy.
Q: What does the cybersecurity pivot mean for Nvidia investors?
A: If cybersecurity becomes a major AI market, it provides a new revenue floor for AI compute demand. If it remains mostly narrative, the circular financing critique gains weight as vendor-financed IPOs distribute risk to public investors.
Q: What is Nscale, and why does its IPO matter?
A: Nscale is a British AI cloud firm seeking $3.5 billion ahead of a U.S. IPO. Nvidia is expected to supply $2 billion. The company has contracted for 194,000 Nvidia GPUs and claims $103 billion in total contracts. Its largest deal — $45 billion with Anthropic — followed rejections from Microsoft and Google.
Q: When does Nvidia report earnings?
A: Nvidia reports quarterly earnings in November 2026. Investors should watch whether the company breaks out cybersecurity-related revenue or keeps it inside the data center segment.
