Huang Says Nvidia Now Sells Over $40B of Gear per Gigawatt

Nvidia chief executive Jensen Huang said the Nvidia equipment that goes into one gigawatt of AI data-centre capacity has risen from about $18 billion to more than $40 billion, in an episode of the Bg2 Pod published on 15 October 2026. He put the rise down to Nvidia selling far more than GPUs into each site, and said: “my sense is that that’s going to keep on going up.”

Huang also estimated Nvidia’s share of the capex of the large cloud providers at probably about 40%, and its share outside them, in on-premises and neocloud data centres, at probably about 80%. Six years ago, he said, “Our percentage of the world’s capex was in single digits.”

Business Pill · CAPEX AND DEPRECIATION

A one-minute explainer of capex and depreciation: why the lifetime of the equipment a company buys up front drives its profits. It teaches the general idea only and says nothing about any company in this story.

What Huang Said About the Gigawatt

Huang spoke about 53 minutes into the episode, in a chapter titled “50 GW of Chips, 35 GW of Power,” with hosts Brad Gerstner and Clark Tang. He was answering a question about how much compute fits into the land, power and shell that might come online, after Tang cited about 50 GW of announced chips and 35 to 40 GW of land, power and shell.

He described Nvidia’s offer widening in steps. “We used to be a GPU company, then we became the largest and fastest growing networking company,” Huang said. “We’re now going to be one of the largest CPU companies, surely the fastest growing one.” He added computing for storage and for knowledge infrastructure, and said “now we’re in security”.

The amount Nvidia offers into one gigawatt went from about “18 billion or so per gigawatt” to “20 somewhat billion per gigawatt” to “now 40 plus per gigawatt,” he said. By our arithmetic, more than $40 billion is more than twice the $18 billion level. In the passage we read he gave no date for the $18 billion level.

What Was Posted on X

Rohan Paul’s 15 October 2026 post summarising Huang’s Bg2 Pod remarks on Nvidia’s dollar content per gigawatt, the relay that first surfaced the figures; the quotes in this piece come from the episode itself.

A clip posted by @firesidealpha on 15 October 2026 in which Huang says Elon Musk is ‘launching 100 launches a year right now’ and ‘you could put Vera Rubin’s in there and it’ll justify it’, calling Nvidia compute ‘fungible’ as a satellite payload. The post does not name the venue of the clip; the remark is Huang’s own view of SpaceX’s demand, not a disclosed order.

Nvidia’s Share of Each Capex Dollar

Huang tied the rise to two sources of share gains: customers shifting towards Nvidia’s computing platform, and Nvidia offering more into each site. He said “our market share is actually increasing” because of both.

“Right now with the world’s CSPs we’re probably running about 40%,” he said, referring to the large cloud service providers. Outside them, in on-premises and neocloud deployments that he described as half of Nvidia’s business, he put Nvidia’s percentage of capex at probably about 80%.

He said both halves of the business are growing, doubling every year, and added: “our percentage of the world’s capex is in fact even growing.”

Watch the passage · Bg2 Pod, 53:14

Why He Says Buyers Are Shifting

Huang linked the shift to power. When the world’s available power is limited, he argued, buyers move to the most productive computing, and he said Nvidia’s architecture “generates the most tokens per watt”. “We’re the most rentable,” he said.

He said his fellow chief executives “all have high expectations for revenues next year” and need to deliver them, which in his account pushes them towards the most productive compute for the power they have.

The comments add to Nvidia’s own disclosures. Nvidia’s latest 10-Q, for the quarter ended 26 July 2026, lists $366 billion of commitments, $56 billion of additional commitments and $108.5 billion of maximum exposure under guarantees, which by our arithmetic add up to $530.5 billion, as we reported on 11 October. On 14 October, Bloomberg reported that Huang would halt chip supply to a customer developing unsafe AI.

The Structural Read

As we read it, the per-gigawatt figure and the share figures are one claim seen twice: if Nvidia sells more items into each site, its slice of the buyer’s capex rises, which is the link Huang draws between offering more and growing share.

As we read it, his two share estimates describe two kinds of buyer: probably about 40% with the large cloud providers and probably about 80% outside them, in on-premises and neocloud deployments, by his estimates.

As we read it, his argument makes power the limit buyers work within: when power is short, he said, buyers move to the most productive compute per watt, and he argued that is Nvidia’s.

Three Implications

CLOUD PROVIDERS Huang put Nvidia at probably about 40% of the large cloud providers’ capex, his estimate on the episode.

NEOCLOUDS AND ON-PREMISES BUYERS He put Nvidia’s share outside the large cloud providers at probably about 80%, in what he described as half of Nvidia’s business.

INVESTORS He said the Nvidia offering into one gigawatt went from about $18 billion to more than $40 billion, and that his sense is that it will keep going up.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Nine Layers of AI.

The framework argues that “the AI economy is not a list of companies racing each other. It is a layered industrial stack”.

As we read it, Huang’s list, from GPUs to networking, CPUs, storage and security, describes one supplier selling into more layers of that stack inside each site; the framework does not discuss these figures.

What Is Not Established

The figures are Huang’s spoken estimates, given with hedges such as “about” and “probably”; we have not matched them to an Nvidia filing. He did not say which products or costs are counted in the per-gigawatt figure beyond the categories he listed.

We read the passage through YouTube’s auto-generated captions, which garble some words, so the quotations follow the captions as published. The share figures are percentages of the buyers’ capex as Huang described them, not Nvidia revenue figures.

The Bottom Line

Jensen Huang said on the Bg2 Pod, published on 15 October 2026, that the Nvidia equipment in one gigawatt of AI capacity has gone from about $18 billion to more than $40 billion, and that he expects it to keep rising.

He estimated Nvidia’s share of capex at probably about 40% with the large cloud providers and probably about 80% outside them, up from single digits six years ago.