Nvidia expects to sell twice as many chips next year as it ships this year, chief executive Jensen Huang told reporters at an AI summit hosted by King Charles in Scotland this month, according to CNBC. The pledge backs the 70 percent fiscal 2028 revenue growth forecast laid out by chief financial officer Colette Kress in August and puts the company’s supply chain on the hook for a doubling with no slack built in.
The stock sits roughly 6 percent below its May all-time high of $235.74, and the gap between Huang’s volume promise and Wall Street’s consensus says a lot about how the market is pricing AI infrastructure. Based on the consensus projection for fiscal 2027 revenue of about $396 billion, 70 percent growth would push annual sales to roughly $673 billion, a number that would make Nvidia by far the largest semiconductor company in history by revenue.
Demand beyond the hyperscalers
Huang’s confidence rests on a demand pattern Kress described on the August 26 earnings call as stretching well past the big cloud companies to what she called a massive market of sovereign governments, neocloud providers and enterprises. Non-hyperscaler customers already account for roughly half of Nvidia’s data center business and are growing at about 100 percent annually, Huang said on the same call.
He also said AI had reached its inflection point, with the number of companies needing large GPU clusters expanding dramatically. Analysts who listened to the call flagged the geographic spread of AI infrastructure spending, from Asia and Europe to the Middle East, as one of its clearest takeaways. Sovereign AI projects in particular have moved from talking points to signed procurement deals over the past year, with national computing programs announced in France, Japan, Saudi Arabia and the United Arab Emirates.
Memory is the bottleneck
The constraint is physical. Kress acknowledged on the August call that the fiscal 2028 outlook reflects available supply, and Huang told analysts that growth would be a lot higher without production limits. Memory scarcity is the biggest bottleneck, and she noted the AI buildout itself is driving much of that shortage across the semiconductor supply chain.
The shortage is visible in consumer prices. DRAM costs have climbed sharply this year as data center buyers absorb supply, pushing up prices for laptops, phones and game consoles. Gross margins are expected to bottom in the fourth quarter of fiscal 2027 at 71 to 72 percent, then settle at 72 to 73 percent in fiscal 2028, Kress said, as rising memory costs keep pressure on profitability.
Memory suppliers have responded with capacity expansions, but new fab capacity takes years to qualify, and the same fabs feed both AI accelerators and consumer devices. That tension has made memory pricing one of the most watched indicators in the hardware supply chain this year, and it explains why Kress flagged the shortage unprompted on the earnings call rather than burying it in the outlook.
What doubling means in practice
Nvidia does not disclose total chip shipment figures, so the doubling claim cannot be checked directly against public numbers. What can be checked is the revenue math, and it implies the average selling price of what Nvidia ships stays roughly flat even as volume doubles. That would mean the mix of products, from flagship rack-scale systems down to mid-range accelerators, holds steady rather than shifting toward cheaper parts.
The company’s fiscal 2028 guidance also assumes its newest architecture ramps on schedule. Production transitions have historically been where Nvidia’s supply forecasts slip, and customers ordering system-level deployments book capacity months in advance. Any delay in the next generation would push revenue between quarters without changing the annual total, but it would test the credibility of the doubling pledge.
The skeptics
Not everyone accepts the forecast. Analysts tracking AI capital spending have questioned whether data center builders can absorb that much hardware, noting that power availability, not chip supply, has become the binding constraint at many planned sites. Grid connections for new data centers are queued years out in several markets, and some large projects have already pushed back completion dates.
There is also the question of who ultimately pays. Most of the demand Huang describes is financed by debt or by circular arrangements in which chip suppliers invest in their own customers. If AI revenue at the application layer does not catch up to infrastructure spending, the doubling could meet a financing wall rather than a demand ceiling.
Competition adds a further variable. Custom accelerator programs at Google, Amazon and several startup chipmakers are maturing, and hyperscalers have been explicit about shifting a portion of inference workloads onto in-house silicon. Nvidia’s response has been to bundle more of the stack, from networking to software, so that the customer decision stays about the platform rather than the chip alone.
For now the market has mostly chosen to believe the company. Huang has a track record of underpromising on supply and overdelivering on demand, and the August guidance already exceeded what analysts had modeled. The stock’s partial recovery from its May peak suggests investors have accepted the growth story while withholding judgment on the memory squeeze. The next test comes with quarterly results, where shipment volumes and memory costs will show whether the doubling is on track, and with the October Fed decision, which will set the macro backdrop the whole buildout is financed against.