Huang says Nvidia can grow about 70% next year — and cybersecurity is AI's next big market

Start with the weight of that 70% figure. Huang is talking about roughly 70% revenue growth in the next fiscal year, and it comes on top of an already enormous base. The justification is a single sentence: demand for AI compute continues to outpace available capacity. That sentence has been repeated for two years, but the context is different this time — it lands just as the market has begun doubting the returns on AI capital spending. The data-center debate has been running for months: tech companies are pouring vast sums into AI infrastructure while actual profits lag the investment, and the AI bubble narrative has taken hold. Huang's response is not to argue about demand but to move the conversation from chip prices to utilization: Nvidia executives argue their systems have long-lasting value because of high utilization and favourable economics, with financing partnerships involving financial firms supporting adoption.

The second judgement worth remembering is the market he names as next: cybersecurity. This matters more than it sounds, because it pushes AI commercialization past productivity tools like code and copy into offence and defence — a domain with extremely strong, recurring willingness to pay. The logic is not hard to follow: if frontier models are already capable of finding and exploiting real software vulnerabilities, the same capabilities will necessarily be used to defend. Third-party evaluators have been observing the same trend; assessing frontier models' actual ability to discover and exploit vulnerabilities is one of the focus areas for organizations like METR. For enterprise and government buyers this creates a new purchasing rationale — not what AI lets you do, but what it stops others from doing to you. Huang also listed robotics and autonomous driving as incremental directions for the next several years, and that physical-AI thread connects neatly with Hyundai Motor Group's autonomous driving data flywheel strategy announced the same day.

The third piece is pricing — and it is the detail that best reveals Nvidia's strategy shift. Vera Rubin at roughly $40,000, Blackwell at about $25,000, Hopper at about $18,000: those three tiers trace a clear upward curve. What they track is not a linear gain in chip performance but a change in the unit of sale — from selling accelerator cards to selling a complete AI factory system. For Nvidia that has two consequences. Order sizes and software lock-in both rise; and the defensive surface narrows — once you are not one link in a supply chain but the general contractor for a customer's AI infrastructure, the customer's switching cost becomes your moat. Huang also singled out the Anthropic relationship: Nvidia invested in the company in November 2025 and is now gaining business from it. That combination of investment plus supply has become a standard play in today's AI chip competition.

Finally, there is his answer on circular financing, the line that travelled furthest: invest $1 and get $100 back — if that is circular financing, we will do more of it. The backstory is that as Nvidia increasingly invests in and buys from cloud providers and model companies, observers question whether this mutual funding loop amplifies risk and eventually manufactures phantom demand. Huang did not dodge it at the Goldman conference; he redefined it as a normal form of capital leverage. The market reaction shows up in analyst moves: on September 13 Piper Sandler initiated coverage of five AI chip stocks — Nvidia, Broadcom, AMD, Arm and Marvell — all at overweight, on the argument that both Nvidia and Broadcom are poised for strong revenue growth driven by insatiable demand for AI compute. The same batch of research also carried the opposite view: Nvidia's share price has lagged the triple-digit gains clocked by AMD and Intel this year, which turns valuation into a catch-up argument. Bulls and bears are ultimately arguing about one variable — whether this capex cycle converts into durable revenue.