Jensen Huang says Nvidia will sell twice as many chips next year, and ties safety back to the product line
On September 17, 2026, Nvidia chief executive Jensen Huang said one sentence in Scotland: I expect Nvidia to sell twice as many chips as this next year as we do this year. There is nothing technically complicated about the sentence; what gives it weight is who said it and where. Nvidia never discloses total chip volumes, so a company publicly putting a doubling forecast on the record effectively lays out where the industry's capital spending is heading next year. His reasoning was equally plain: AI contributes so much to the benefits of different industries and different economies, and in almost every country Nvidia operates in, people want to invest in AI.
Start with the order of magnitude. Nvidia does not disclose total chip volumes, and the closest public reference is what Huang said last autumn: the company shipped 6 million Blackwell GPUs in four quarters. A little earlier, the company said it expected about 70% growth in the fiscal year ending January 2028, roughly 673 billion dollars, a figure that pushed analyst expectations up across the board. Put the two together and the doubling line is not a standalone slogan but a spoken version of a longer growth forecast. Note also that Nvidia's products have long since stopped being just data centre GPUs: it also makes central processors, switch chips, chips for optical networking, chips for laptops, Jetson chips for robots and cars, and the chip inside Nintendo's Switch 2. Doubling refers to volume, not revenue, and not to any single product line.
The second thing worth noticing is the room. According to CNBC, Huang was joined by representatives from Google DeepMind, OpenAI and Anthropic, with AI safety among the topics, and part of the purpose of his trip to the United Kingdom was to discuss it. That explains why a remark about sales volume ends up landing on safety. Huang's phrasing is that when a product is not safe, we should hold it back and keep engineering it. Semantically that is a fairly conservative position: it treats safety as a question of product maturity rather than of research tempo. An unsafe thing, on this reading, is a product held back until it is finished, not an entire frontier research line stopped.
To gauge the weight of the safety remark, look at what surrounds it. In recent weeks a group of AI researchers warned that humans are no longer able to control the most advanced artificial intelligence, and the debate over risks to humanity reached a fever pitch; Anthropic chief executive Dario Amodei publicly urged a pause on frontier model development, and AI chip stocks fell as a result. At the same time, the direction from Washington is to maintain the lead over China, with President Donald Trump pushing back on calls to slow down. Huang sits precisely where the two lines cross: he is both the person saying demand will double and the person saying unsafe products should be held back. The two positions are not contradictory, but they serve different judgements - the first assumes growth continues, the second assumes growth should be orderly.
One more detail deserves its own paragraph, because it bears on whether the forecast can be checked. Huang said the number of chips will double, and Nvidia never publishes a total chip count, offering shipment figures for a specific part only occasionally as a reference point. That makes the forecast literally falsifiable but hard to falsify in practice: what outsiders can observe is revenue, shipment claims for particular product lines, and customers' own capital spending guidance. For people making technical decisions, the more practical use is to treat it as a directional signal. If the supply side itself expects volumes to double, then compute pricing, delivery lead times and power provisioning are unlikely to ease next year, and procurement and capacity planning should proceed on that assumption.
🤔 Frequently Asked Questions
What exactly did Huang say?
He said: I expect Nvidia to sell twice as many chips as this next year as we do this year. His reason was that AI contributes so much to the benefits of different industries and economies that countries want to invest in AI. He spoke at a summit in Scotland convened by King Charles III, and according to CNBC one purpose of the trip was to discuss AI safety.
Doubling of what exactly?
Of the number of chips, not revenue and not any single product line. Nvidia does not disclose total chip volumes; the closest public reference is what Huang said last autumn, that the company shipped 6 million Blackwell GPUs in four quarters.
What growth had Nvidia already guided to?
About 70% growth in the fiscal year ending January 2028, roughly 673 billion dollars. The guidance pushed analyst expectations up across the board, and Huang's spoken forecast can be read as the same trajectory expressed in a volume sense.
What did he say about safety?
When a product is not safe, we should hold it back and keep engineering it. That is a product-maturity framing, arriving alongside calls from Anthropic's chief executive and others to slow frontier development and a US administration position of maintaining the lead over China.
🛠️ Recommended Tools
- Unit ConverterUnit conversions for compute and bandwidth are the daily reality of this beat: GPU shipments, data centre power and network bandwidth all appear together, and manual conversion is where orders of magnitude go wrong.
- AI Token CounterThe direct consequence of doubling volume lands on inference cost. Counting actual tokens per call first, then comparing per-million-token prices across parts, is how you project the monthly bill after scaling.
- Timestamp ConverterThis story spans fiscal calendars: the fiscal year ending January 2028, last autumn, September 17. Aligning dates across time zones and fiscal years is a common source of error, and one conversion saves a round of checking.
Summary
On September 17, 2026, Nvidia chief executive Jensen Huang said at a summit in Scotland convened by King Charles III that the company will sell twice as many chips next year as this year, citing AI's contribution to the benefits of different industries and economies and the resulting appetite to invest. Nvidia does not disclose total chip volumes; the closest public reference is what Huang said last autumn, that it shipped 6 million Blackwell GPUs in four quarters, and the company had already guided to about 70% growth in the fiscal year ending January 2028, roughly 673 billion dollars. According to CNBC, representatives from Google DeepMind, OpenAI and Anthropic were also at the summit, with AI safety among the topics. On safety, Huang said that when a product is not safe, we should hold it back and keep engineering it. The context is a recent warning from a group of AI researchers that humans can no longer control the most advanced systems, calls from Anthropic's chief executive to slow frontier development, and a US administration position of maintaining the lead over China. Note that the doubling refers to volume rather than revenue, that Nvidia publishes no total, and that the forecast is therefore hard to falsify in practice. All figures and quotes come from CNBC reporting, with nothing inferred.
Sources: CNBC: Jensen Huang says Nvidia will sell twice as many chips next year