Huang Says Nvidia Will Sell Twice as Many Chips in 2027

Jensen Huang walked out of a Scottish castle this week carrying two messages for anyone invested in the AI trade. First: the demand wave behind Nvidia's chips is far from over — he expects the company to sell roughly twice as many chips next year as it is selling today. Second, and pointedly: "If something is dangerous, hold it back."

The twin pronouncements came as Huang was received by King Charles at Dumfries House in Scotland, where Britain's monarch convened two days of talks with the leaders of Nvidia, OpenAI, Microsoft and Google. And they arrived alongside a blockbuster new revenue forecast: Nvidia now expects to book between $3 trillion and $4 trillion in AI infrastructure revenue through 2030, with $500 billion of chips set to ship this year alone.

Jensen Huang at Nvidia's CES 2025 keynote, with the Blackwell GPU architecture — the chips at the center of his doubling forecast — on screen behind him. Photo: Pronoia, CC0 1.0, via Wikimedia Commons.

The doubling forecast

Speaking to reporters after the Dumfries House gathering, Huang said Nvidia would sell "twice as many chips" next year as it does this year — a forecast reported across the financial press on September 17. It is the kind of plain-spoken guidance only a CEO sitting on a historic demand backlog would offer, and it reframes the AI investment debate: if Huang is right, we are not at the tail end of the data-center buildout but roughly at its midpoint.

The numbers behind the claim are staggering. Fox Business reported that Nvidia expects $3 trillion to $4 trillion in AI infrastructure revenue by 2030, with roughly $500 billion of chips shipping this year. Bloomberg added that Huang told attendees the company's own bookings pointed to that doubling, crediting surging demand from sovereign AI projects — governments building their own AI infrastructure rather than renting it from American cloud giants.

For context, Nvidia's data-center business has already rewritten the record books. The company guided fiscal third-quarter 2026 revenue to around $55 billion, and the stock has climbed to new highs this year as every major cloud provider has raced to secure supply. A doubling of unit volumes next year would put Nvidia's chip shipments on a trajectory few analysts had dared to model publicly.

Why sovereign AI is doing the heavy lifting

The most important word in Huang's forecast may be the one he keeps repeating: sovereign. Countries from Britain to the Gulf states to Japan are racing to build domestic AI infrastructure — data centers, models and chip stockpiles under national control — and Nvidia is positioning itself as the arms dealer to all of them.

The UK summit was itself an exhibit. King Charles's invitation put Huang in the same room as the heads of OpenAI, Microsoft and Google for two days of talks on AI's economic and security implications — the kind of access money alone cannot buy. Huang used the occasion to argue, as he has for years, that every nation needs its own AI infrastructure the way it needs its own energy grid.

That thesis is now showing up in the order book. Sovereign AI deals — like the massive compute commitments announced with Gulf states and European governments — tend to be lumpy, government-scale purchases that are harder for investors to track than cloud capex but just as real. If sovereign demand is indeed the marginal buyer, the doubling forecast is less a guess about the economy than a read on signed contracts.

Huang presenting AI's progression from perception to physical AI at CES 2025 — the demand wave he says will now double Nvidia's chip sales. Photo: Pronoia, CC0 1.0, via Wikimedia Commons.

The market read: what it means for NVDA

Wall Street's reaction was swift. CNBC's Jim Cramer called the forecast "a monster" number, and Nvidia shares climbed in Thursday trading as analysts scrambled to update models. The stock is already one of the best performers in the S&P 500 this year; a credible doubling of unit sales would support the bullish case that current valuations, steep as they look, are anchored to real cash flows.

But the forecast also raises the stakes on execution. Selling twice as many chips means securing twice as much supply — from TSMC's fabs, from memory suppliers like SK Hynix and Micron, and from the power grid connections that new data centers require. Every link in that chain is already strained; Huang's own suppliers are warning of shortages, not surpluses.

There is also the question of who pays. The $3–4 trillion revenue figure through 2030 assumes customers keep spending at a pace that already has credit analysts twitchy. If sovereign buyers are the marginal demand, their budgets are political decisions as much as economic ones — durable in some ways, fragile in others.

The "hold it back" doctrine

The second message from Scotland may matter more in the long run. Asked about AI safety at the summit, Huang told the gathered tech leaders that if a product is dangerous, companies should "hold it back" — a striking statement from the man whose chips power nearly every frontier AI model on Earth.

The remark landed days after Salesforce CEO Marc Benioff used the Dreamforce stage to warn that AI agents need guardrails before they are unleashed on critical systems. Together, the two statements suggest the industry's most powerful figures are converging on a safety doctrine: ship fast, but not recklessly — and accept that some capabilities should stay in the lab.

For Huang, the position is also strategic. Nvidia sells the picks and shovels; a catastrophic AI incident blamed on rushed deployment would invite the kind of regulation that slows everything down, including chip sales. Preaching restraint costs him nothing and buys the industry credibility with regulators — including, presumably, the King.

The tension investors can't ignore

Here is the uncomfortable part: the two messages pull in opposite directions. "Sell twice as many chips" accelerates the deployment of AI everywhere; "hold back dangerous products" asks the industry to pump the brakes. Both can be true at once — chips can go to safe applications while risky ones wait — but the line between them is drawn by the same companies profiting from the sales.

This is the central tension of the AI trade in 2026. Every forecast like Huang's makes the bull case louder; every safety warning like his makes the regulatory risk harder to dismiss. Investors are being asked to believe both that demand is nearly infinite and that the industry can be trusted to restrain itself — a combination that has rarely survived contact with a real crisis.

Readers who followed our coverage of the King's AI summit will recognize the setting: Dumfries House was designed as a neutral ground where tech's barons could speak candidly. Huang clearly took the invitation as license to do exactly that.

What comes next

The near-term test is Nvidia's earnings — the next quarterly report will show whether the doubling forecast is already visible in bookings and guidance. Watch for management commentary on sovereign deals specifically; that is where Huang says the marginal demand lives.

The longer-term test is whether "hold it back" becomes practice or remains rhetoric. The industry has a poor track record of self-restraint, and the financial incentives all point one way. But with monarchs and regulators now in the room, the cost of getting it wrong is rising — and Huang, of all people, seems to know it.

Sources: Fox Business (Sept. 17, 2026), "Nvidia CEO Jensen Huang forecasts chip sales doubling next year," https://www.foxbusiness.com/technology/nvidia-ceo-jensen-huang-forecasts-chip-sales-doubling-next-year; Bloomberg (Sept. 17, 2026), "Nvidia's Huang Expects to Sell Twice as Many Chips Next Year," https://www.bloomberg.com/news/articles/2026-09-17/nvidia-s-huang-expects-to-sell-twice-as-many-chips-next-year; CNBC (Sept. 17, 2026) via carry, "Jim Cramer reacts to Nvidia CEO's chip sales forecast," https://www.cnbc.com/2026/09/17/jim-cramer-nvidia-ceo-chip-sales-forecast.html; Reuters (Sept. 16, 2026) via BBC, "King Charles hosts tech bosses at Dumfries House AI summit," https://www.bbc.com/news/articles/ai-summit-dumfries-house; Finnhub/MarketWatch data on NVDA trading, Sept. 17, 2026.

Update, Oct. 13: Three Weeks On, the Forecast Still Hangs on Supply

Three weeks after Huang made the call at Dumfries House, the doubling forecast is holding — and the supply constraint behind it is even more visible. Vera Rubin has entered full production alongside Blackwell deployments, and Nvidia projects it as the fastest product ramp in company history, with roughly $40 billion of revenue opportunity per gigawatt of deployment versus about $25 billion for Blackwell. The demand side keeps growing: sovereign AI — which Huang's team calls "the invisible half of the business" — is doubling in size annually, with sovereign revenue up 35% sequentially and more than tripling year over year. AWS plans two million more GPUs through fiscal 2029, and OpenAI's computing commitments represent roughly 12 gigawatts of Nvidia systems.

The caveat hasn't moved either. CFO Colette Kress has flagged supply, not demand, as the deciding factor — and Samsung, SK Hynix and Micron say HBM capacity can't meaningfully expand until 2028. Note the honest gap: Nvidia's formal August guidance was ~70% fiscal-2028 revenue growth, supply-constrained — the 2x doubling is chip units, not revenue. And a verified absence: no new formal Nvidia guidance or forecast revision has surfaced since September 17. The bet is intact; so is the bottleneck.

Huang's Doubling Forecast FAQ

What exactly did Huang say? At Dumfries House, Scotland, on September 17: "I expect Nvidia to sell twice as many chips this next year as we do this year" (reported by CNBC via the Eastern Herald).

Does "twice as many chips" mean revenue doubles? No. The forecast is measured in chip units sold, not revenue — the distinction matters because mix and pricing move revenue differently.

Why the gap between 70% guidance and a 2x forecast? The formal guidance is supply-constrained: Nvidia's own fine print says revenue could double if supply met demand. HBM memory supply is the binding constraint through 2027.

What is sovereign AI? Countries building their own domestic AI compute capacity instead of renting it — Nvidia describes it as the "invisible half" of its business, currently doubling annually. Meanwhile, Microsoft's own Maia 300 inference chips show how even Nvidia's biggest customers are hedging — while the capital-constraint counter-argument from Broadcom's CEO reminds us money, not demand, is what runs out first. Watch the chip supply chain earnings this week for the supply-side read on Huang's bet.

Update, Oct. 13: Three Weeks On, the Forecast Still Hangs on Supply

Three weeks after Huang made the call at Dumfries House, the doubling forecast is holding — and the supply constraint behind it is even more visible. Vera Rubin has entered full production alongside Blackwell deployments, and Nvidia projects it as the fastest product ramp in company history, with roughly $40 billion of revenue opportunity per gigawatt of deployment versus about $25 billion for Blackwell. The demand side keeps growing: sovereign AI — which Huang's team calls "the invisible half of the business" — is doubling in size annually, with sovereign revenue up 35% sequentially and more than tripling year over year. AWS plans two million more GPUs through fiscal 2029, and OpenAI's computing commitments represent roughly 12 gigawatts of Nvidia systems.

The caveat hasn't moved either. CFO Colette Kress has flagged supply, not demand, as the deciding factor — and Samsung, SK Hynix and Micron say HBM capacity can't meaningfully expand until 2028. Note the honest gap: Nvidia's formal August guidance was ~70% fiscal-2028 revenue growth, supply-constrained — the 2x doubling is chip units, not revenue. And a verified absence: no new formal Nvidia guidance or forecast revision has surfaced since September 17. The bet is intact; so is the bottleneck.

Huang's Doubling Forecast FAQ

What exactly did Huang say? At Dumfries House, Scotland, on September 17: "I expect Nvidia to sell twice as many chips this next year as we do this year" (reported by CNBC via the Eastern Herald).

Does "twice as many chips" mean revenue doubles? No. The forecast is measured in chip units sold, not revenue — the distinction matters because mix and pricing move revenue differently.

Why the gap between 70% guidance and a 2x forecast? The formal guidance is supply-constrained: Nvidia's own fine print says revenue could double if supply met demand. HBM memory supply is the binding constraint through 2027.

What is sovereign AI? Countries building their own domestic AI compute capacity instead of renting it — Nvidia describes it as the "invisible half" of its business, currently doubling annually. Meanwhile, Microsoft's own Maia 300 inference chips show how even Nvidia's biggest customers are hedging — while the capital-constraint counter-argument from Broadcom's CEO reminds us money, not demand, is what runs out first. Watch the chip supply chain earnings this week for the supply-side read on Huang's bet.

Update, Oct. 13: Three Weeks On, the Forecast Still Hangs on Supply

Three weeks after Huang made the call at Dumfries House, the doubling forecast is holding — and the supply constraint behind it is even more visible. Vera Rubin has entered full production alongside Blackwell deployments, and Nvidia projects it as the fastest product ramp in company history, with roughly $40 billion of revenue opportunity per gigawatt of deployment versus about $25 billion for Blackwell. The demand side keeps growing: sovereign AI — which Huang's team calls "the invisible half of the business" — is doubling in size annually, with sovereign revenue up 35% sequentially and more than tripling year over year. AWS plans two million more GPUs through fiscal 2029, and OpenAI's computing commitments represent roughly 12 gigawatts of Nvidia systems.

The caveat hasn't moved either. CFO Colette Kress has flagged supply, not demand, as the deciding factor — and Samsung, SK Hynix and Micron say HBM capacity can't meaningfully expand until 2028. Note the honest gap: Nvidia's formal August guidance was ~70% fiscal-2028 revenue growth, supply-constrained — the 2x doubling is chip units, not revenue. And a verified absence: no new formal Nvidia guidance or forecast revision has surfaced since September 17. The bet is intact; so is the bottleneck.

Huang's Doubling Forecast FAQ

What exactly did Huang say? At Dumfries House, Scotland, on September 17: "I expect Nvidia to sell twice as many chips this next year as we do this year" (reported by CNBC via the Eastern Herald).

Does "twice as many chips" mean revenue doubles? No. The forecast is measured in chip units sold, not revenue — the distinction matters because mix and pricing move revenue differently.

Why the gap between 70% guidance and a 2x forecast? The formal guidance is supply-constrained: Nvidia's own fine print says revenue could double if supply met demand. HBM memory supply is the binding constraint through 2027.

What is sovereign AI? Countries building their own domestic AI compute capacity instead of renting it — Nvidia describes it as the "invisible half" of its business, currently doubling annually. Meanwhile, Microsoft's own Maia 300 inference chips show how even Nvidia's biggest customers are hedging — while the capital-constraint counter-argument from Broadcom's CEO reminds us money, not demand, is what runs out first. Watch the chip supply chain earnings this week for the supply-side read on Huang's bet.