Product2 distinct publishers3 min readPublished
Nvidia's CEO claimed the milestone and dismissed it in the same answer, without ever defining it. For anyone who has to write acceptance criteria next week, that is reason enough to strike the word from the template.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The box on the vendor questionnaire reads "describe your path to AGI," and the person scoring the answers has nothing to score with. Three paragraphs come back, and they are neither right nor wrong. Everything else on the form has a unit attached, so the AGI line gets a checkmark and travels into the contract as decoration.
The zero percent answer from the podcast is the useful part of the record [6]. The claim held up as long as it stayed general, but it fell apart once someone attached an actual threshold to it [5], which is exactly what happens to the word in a procurement review.
The one version of AGI that ever had to bind two parties is written in money. Set the reported $100 billion profit threshold [8] against Nvidia's own quarter: $96.2 billion in fiscal Q2 revenue, up 106% year over year [9], works out to 96.2% of that figure [1]. The units do not match: revenue against profit. That mismatch is the finding. The only definition precise enough to litigate is one a chip vendor's income statement is already brushing against.
Underneath it, vocabulary is doing the work definitions cannot. OpenAI's chief research officer, Mark Chen, told Time the company is "80% of the way" to AGI [10], a percentage of a target Sam Altman has himself called "not a super useful term" [11]. Anthropic's Dario Amodei calls it imprecise, a marketing term, and says "powerful AI" instead [12]. Add the house coinages the Verge lists and there are six labels in circulation besides AGI, roughly one per lab [2].
Teams tell themselves that watching the AGI timeline tells them when to redo the plan. What actually moves a plan is the thing Huang put in AGI's place, and read closely it is a supply claim rather than a capability one. On the same call, CFO Colette Kress said memory pricing has exceeded the company's prior expectations and is headed higher next year [14], with the shortage expected to bottleneck Nvidia at least through the early part of 2028 [15]. PCMag also notes that the two leading model companies have yet to prove they are profitable [17]. Tokens per dollar, and who can get parts, is the portion of this you can put in a capacity plan.
The forcing function takes an afternoon. Search your roadmap and your RFP template for AGI and its house variants, delete the word, and hand each sentence back to whoever wrote it with one requirement: that it name a task and the measurement that settles whether the task got done. A sentence that survives that test was doing work; one that collapses tells you which slide existed just to make the quarter feel bigger.
Ranked by verification strength, evidence, and original report placement.
On Nvidia's earnings call on Wednesday, CEO Jensen Huang said "for many tasks, we could say that we've already achieved AGI."
Huang called such milestones "kind of senseless at this point" and did not provide a precise definition or benchmark for AGI on the call.
Huang said what matters for the industry is that AI is doing productive and useful work, that AI is generating profitable tokens, and that more compute would generate more profitable tokens and more profit.
In March, on Lex Fridman's podcast, Huang said "I think we've achieved AGI"; Fridman proposed a definition of an AI system able to start, grow and run a successful tech company worth more than $1 billion.
Responding to that definition, Huang said "the odds of 100,000 of those agents building Nvidia is zero percent."
OpenAI's charter defines AGI as "highly autonomous systems that outperform humans at most economically valuable work."
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 26, 2026
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Washington's secret AI test is coming for open weights, and release dates go with it2 distinct publishers
invest
The 81% Problem: AI's Star CEOs Are Polling Badly With The People They Need To Hire1 distinct publisher
product
Nine AI leaders, nine majorities of distrust: the floor onboarding copy cannot lift1 distinct publisher
leadership
Data center opposition is now a siting cost, and the industry is pricing it as a PR line1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One transcript, quoted twice
The core of this story is unusually clean: PCMag and The Verge transcribe the same earnings call and agree word for word on the two lines that matter, including the dismissal. Everything at a remove gets thinner fast. The $100 billion profit trigger arrives with a "reportedly" and only in The Verge's telling; the Chen and Altman figures are relayed from a Time interview neither outlet conducted; and PCMag's profitability point about OpenAI and Anthropic is asserted without figures.
Compute is adopted; the milestone is not
Nothing in this reporting shows a system tested against any definition of AGI — Huang's supporting detail is that agents run autonomously and improve "recursively," which is an assertion, not a measurement. What is genuinely measurable sits one layer down in the same call: a $96.2 billion quarter and a memory market so tight that Nvidia guides scarcity into 2028. Demand for the hardware is documented; arrival at the milestone is not.
Declared, then made uncheckable
A milestone claimed and retired in the same answer cannot be falsified, which is the whole problem. Huang's own March remark — the odds of 100,000 such agents building Nvidia are "zero percent" — undercuts the announcement more effectively than any critic, and the executives who trade on the term concede it is imprecise while continuing to forecast against it. Note the gap sits in the claims, not the coverage: The Verge's framing is openly skeptical, and PCMag pairs the boast with a bubble caveat.
The shovel seller sets the scorecard
Huang's replacement metric — useful work, profitable tokens, more compute means more profit — is also Nvidia's order book, and he offered it on a call reporting 106% growth. On the other side, per The Verge, the only AGI definition with a number attached is a reported $100 billion profit trigger governing OpenAI's relationship with Microsoft, which makes the word a contractual lever rather than a scientific one. PCMag adds the incentive to keep talking about productivity at all: the two leading labs have yet to show a profit.
Sure what was said, not what it means
Two independent transcriptions of a public earnings call put the quotes beyond reasonable doubt, and Nvidia's own numbers and guidance are on the record. The uncertainty is downstream: the profit threshold is second-hand, the Time-sourced timelines are someone else's interview, and the central term has no definition either outlet can pin down — so any assessment of whether the claim is true has nothing to grip.