Nvidia's $5 Trillion Milestone: Orders, Demand & Bubble Debate

Five trillion on the tape

Nvidia closed October 29, 2025 more than 3 percent higher, taking its market capitalization to roughly $5.03 trillion, the first close any company had recorded above five trillion dollars. The move came the same day President Trump signaled, ahead of a planned meeting with CEO Jensen Huang, that Nvidia's Blackwell chips might get an export path to China that had not previously existed, adding a geopolitical tailwind on top of an AI rally already in progress. [1]

Asked directly about bubble comparisons, Huang said the companies buying Nvidia's chips "are generating real revenues" and that the products underneath the spending "are profitable," a defense he repeated through the fall as the round-number milestones kept arriving faster than the skepticism could catch up. [1][3]

Atlas interpretation: Nvidia had already crossed four trillion dollars in market capitalization on July 9, an intraday peak that closed just under the mark and looked, at the time, like a plausible pause. The climb from four trillion to five took about sixteen weeks, the same stretch in which Nvidia's chip export dispute with China resurfaced, a new round of multi-billion-dollar compute deals landed, and a public argument about whether any of the spending was sustainable started in earnest. None of it slowed the stock down; the case against it just got louder alongside the case for it. [1][9]

What filled the four months

Huang used Nvidia's GTC conference in Washington, D.C., the day before the close, to put a number on the order book. "This is how much business is on the books. Half a trillion dollars worth so far," he said, describing combined 2025 and 2026 orders for Blackwell GPUs, the coming Rubin generation, and networking gear. The keynote also detailed strategic investments Nvidia had committed to OpenAI, Intel and Nokia, tying chip sales to equity stakes in the companies buying them. [3]

The order book sat on top of results that were already record-setting. For the quarter ended July 27, 2025, reported August 27, Nvidia's data center revenue reached $41.1 billion, up 56 percent from a year earlier, with zero H20 chip sales into China during the quarter. The core business was accelerating before any of the October announcements landed. [4]

The quarter Nvidia was actually trading through when it closed above five trillion dollars ended October 26 and was reported November 19: data center revenue of $51.2 billion, up 66 percent year over year, with Huang telling investors "Blackwell sales are off the charts, and cloud GPUs are sold out." The five-trillion close had priced in a quarter the company had not yet reported, and the quarter came in ahead of it. [5]

Atlas interpretation: Some of the four months' commitments were Nvidia's own; some belonged to a competitor bidding for the same customers. AMD's own six-gigawatt supply agreement with OpenAI, signed October 6, carried a warrant for up to 160 million AMD shares tied to deployment milestones. Nvidia's investors read a rival needing to hand over equity, not just win a purchase order, as confirmation of scarcity rather than competition: whichever company's logo ended up on the chip, the buyer had to pay in more than cash to secure supply. [1][3]

The paradox the market picked

Nvidia lost roughly $600 billion in market value in a single session in January, on the theory that DeepSeek had shown frontier models could be trained and run for a fraction of the presumed cost, and that cheaper models would need fewer chips to serve. Nvidia's own statement that same day called DeepSeek's R1 "an excellent AI advancement" rather than a threat. Microsoft CEO Satya Nadella went further on social media the next day, writing "Jevons paradox strikes again" and arguing that as AI got cheaper to run, people would use dramatically more of it, not less. [8][7]

Atlas interpretation: Jevons paradox, named for the 19th-century economist who noticed that more efficient steam engines increased coal consumption rather than cutting it, was Nadella's bet that efficiency and total demand would move together, not in opposite directions. Through the rest of 2025, Nvidia's own results were the running scoreboard for that bet: data center revenue kept setting records even as the industry talked openly about cheaper inference, distillation, and smaller models doing more of the work. Cheaper tokens did not translate into fewer tokens sold; they showed up as more applications built on the same underlying compute. [7][4]

Atlas interpretation: By October 29 the market had settled on the opposite reading from January's: efficient inference meant more inference happening, and more inference meant more chips to run it on. Nobody disputed that DeepSeek's efficiency claims were real; the five-trillion-dollar close was a rejection of the inference that efficiency would shrink the chip market, not of the underlying result. Nine months separated two very different verdicts on the same fact. [1][5]

The bubble argument, still live

The bubble argument did not go away after the milestone; if anything it sharpened. Michael Burry's Scion Asset Management disclosed, in a filing signed November 3, roughly $1.1 billion in put options against Nvidia and Palantir, a bearish bet made before Burry deregistered the fund as an investment adviser a week later. On social media he argued that Nvidia's biggest customers, big cloud providers including Meta and Oracle, were depreciating GPUs over five to six years on their books against a real useful life he put closer to two or three years, a gap he estimated understated their combined depreciation, and overstated their profits, by roughly $176 billion between 2026 and 2028. [6]

Nvidia and its bulls answered in kind. Huang's response to bubble questions at GTC, that customers "are generating real revenues" and that the products underneath the spending "are profitable," was one version of the counterargument: demand for compute was showing up in reported results, not only in orders on a slide. Wolfe Research analyst Chris Caso made a related case from the order book itself, arguing the $500 billion figure implied upside to consensus 2026 revenue estimates rather than a company inflating its own backlog. [6][3]

Atlas interpretation: Both sides were arguing from facts that were not in dispute: Nvidia's revenue and order book were real and growing, and big cloud providers were using a depreciation schedule Burry considered too generous. What separated the two readings was not the data but its meaning, whether reported profits reflected the true economics of the buildout or were flattered by an accounting choice that would eventually have to reverse. That argument remained open as of this writing, and a five-trillion-dollar close settled none of it. [6]