Nvidia CEO Jensen Huang has a message for elite software engineers or AI researchers whose annual AI token bill comes in at $5,000: He will, in his words, “go ape something else.”
Speaking on the All-In Podcast on the last day of GTC 2026, Huang laid out a thought experiment. “Let’s say you have a software engineer or AI researcher, and you pay them $500,000 a year,” he said. “We do that all the time.” His mandate: that $500,000 engineer should be consuming at least $250,000 worth of tokens annually. When asked if Nvidia is spending around $2 billion a year on tokens for its engineering team, Huang answered: “We’re trying to.”
And then he extends it further with the CAD tools analogy: “this is no different than one of our chip designers who says guess what I’m just going to use paper and pencil.”
The quotes have been everywhere. But the coverage has mostly stopped at the headline. We did the math on what Jensen’s dream actually could mean for the companies selling those tokens: specifically, whether the demand he’s describing would make AI’s two biggest money-losers profitable overnight.
| Scenario | Engineers | New Token Revenue | OpenAI P&L Impact | Anthropic P&L Impact |
| Scenario 1: Jensen’s Literal Dream Only $500K+ engineers150K × $250K | 150,000 | $37.5B | –$7B (covers half of the losses) | –$0.5B (nearly breakeven) |
| Scenario 2: The Trickle-Down Proportional budgets150K×$250K + 550K×$75K+ 1.2M×$17.5K | 1.9M | $100B | +$4B | +$7B |
| Scenario 3: Full Jensen Every U.S. SWE at 50% of total comp | 4.4M | $358B | +$50B | +$38B |
Who actually earns $500,000?
Jensen’s thought experiment starts with a $500,000 engineer or AI researcher. That number is real… at well-paying employers like Nvidia as well as at frontier AI labs themselves. But that’s far outside of common compensation for the broader software engineering workforce. BLS tracks data scientists separately, with about 246,000 employed at a median salary of $112,590, and computer and information research scientists, the closest proxy for frontier AI researchers, at 40,300 jobs and a median of $140,910. Neither group is pulling $500K in total comp outside of top-tier tech firms.”
Still, the $500,000 likely figure refers to total compensation: base salary plus stock grants, bonuses and benefits. The median U.S. software engineer earns nothing close to that. Bureau of Labor Statistics data from May 2024 pegs the median salary for software developers at $130,160, with 1.79 million professionals employed nationwide. Glassdoor puts the average at roughly $149,000. Levels.fyi, which tracks total compensation including equity, shows a median of $190,100.
The $500,000 total comp threshold is essentially Staff-level or above at a FAANG company or late-stage unicorn. For reference: a Staff Software Engineer at LinkedIn carries a median total comp of roughly $475,000. At Google, L6 engineers (Staff) median around $490,000–$530,000.
Things that that wow, this is too hard. That thought is gone. This is going to take a long time. That thought is gone. We’re going to need a lot of people. That thought is gone. —Huang
The United States has approximately 4.4 million people working in software engineering, according to multiple industry analyses. Of those, the number earning $500,000 or more in total compensation is roughly 100,000 to 200,000. That is somewhere between 2% and 5% of the total.
Jensen even signaled this when he said “we do that all the time.” Nvidia does. Most companies don’t. A mid-level engineer at a healthcare IT firm in Ohio is not pulling down $500K in total comp. And the math gets absurd fast: if you apply Jensen’s 50% token ratio to the median total comp of $190,000, you get a $95,000 annual token budget. For an engineer whose base salary is $130,000, the company would be spending more on AI tokens than on the person.
What would Jensen’s dream do to AI economics?
While investors continue to worry about an AI bubble, the four largest tech companies will spend an estimated $320 billion on AI infrastructure in 2026. R&D funding in the field has eclipsed what the U.S. spend on the Apollo program. Nvidia just posted record quarterly revenue. OpenAI is on pace for $25 billion in annualized revenue. Anthropic hit $19 billion. The money is flowing.
But where is the return on all of that spending? Goldman Sachs says AI’s economic impact so far is basically zero in the U.S. Chief Economist Jan Hatzius said “we don’t actually view AI investment as strongly growth positive,” he said.
Nobody says that building looks really heavy. Nobody. Wow, that mountain looks too big. Nobody says that. Everything that’s too big, too heavy, takes too long, those thoughts, those ideas are all gone. —Huang
OpenAI and Anthropic, the two largest pure-play AI API providers, are both losing enormous sums of money. If Jensen’s vision of massive token consumption materialized, what would it do to their balance sheets?
First, the current state of play. Anthropic surpassed $19 billion in annualized revenue in March 2026, up from $9 billion at the end of 2025. OpenAI hit an estimated $25 billion annualized run rate in February 2026. Combined, that’s roughly $44 billion in run rate.
But neither company is profitable. OpenAI’s own internal projections show $14 billion in losses for 2026, with cumulative losses of $44 billion expected through the end of 2028 before turning a profit of $14 billion in 2029. Anthropic projects reaching positive cash flow by 2028, with estimated 2026 losses in the range of $4–$6 billion.
First, the compute curve. The shift from generative AI to reasoning models required roughly 100x more computation. The shift from reasoning to agentic systems required another 100x. “In just two years, computation went up by a factor of 10,000x,” Huang said. And in his view, scaling has barely started.
Second, the revenue model changes. “People pay for information, but people mostly pay for work,” he said. “Agentic systems get work done.” Chatbot queries generate token revenue. Agents that replace labor generate enterprise revenue, and enterprises pay accordingly.
Third, the distribution layer. Huang predicted that “every single enterprise software company will also be a value-added reseller” of tokens from Anthropic, OpenAI and others. This is what Huang calls a “logarithmic expansion” of their go-to-market. In this framing, token demand doesn’t scale linearly with engineer headcount. It scales with every SaaS product that embeds AI into its workflows.
To model the impact of Jensen’s vision, I built three scenarios ranging from “what he literally said” to “what he’s implicitly projecting.” The market share assumptions use a 45/30/15/10 split across OpenAI, Anthropic, Google and others, and a 40% gross margin reflective of current AI inference economics.
Assumptions: Market share split of 45% OpenAI / 30% Anthropic / 15% Google / 10% others. Gross margin of 40%, consistent with current AI inference economics. OpenAI 2026 operating loss of $14B (per The Information). Anthropic 2026 operating loss estimated at $5B. Revenue figures represent new token revenue from engineer consumption, additive to current run rates.
The breakeven number
At current gross margins of roughly 40%, it takes just 12% of “Full Jensen” to make Anthropic profitable, and about 22% to break OpenAI even.
In dollar terms: OpenAI needs approximately $78 billion in new token revenue to erase its $14 billion annual loss at 40% gross margin—assuming its 45% market share. Anthropic needs roughly $42 billion, but given its smaller loss footprint, it reaches breakeven at a much lower threshold of total market expansion.
Even Scenario 1, where only the roughly 150,000 U.S. engineers who actually earn $500K+ adopt $250K token budgets, generates $37.5 billion in new annual revenue. That alone nearly closes Anthropic’s gap and covers roughly half of OpenAI’s annual burn. The “literal” version of what Jensen described, applied only to the people he was literally describing, is already transformative.
Methodology and sources
Engineer population: 4.4 million U.S. software engineers (multiple industry sources including Lemon.io, DQYDJ, GoRemotely); 1.79 million BLS-counted software developers (BLS OEWS May 2024). Compensation data: Median salary $130,160 (BLS); median total comp $190,100 (Levels.fyi); FAANG Staff+ total comp $500K–$700K (Hakia/Levels.fyi analysis). Revenue data: Anthropic $19B annualized run rate (Bloomberg, March 2026); OpenAI $25B annualized run rate (Sacra/The Information, February 2026). Loss estimates: OpenAI $14B projected 2026 loss (The Information); Anthropic $4–$6B estimated 2026 loss (author estimate based on reported burn rate and revenue trajectory). Gross margin of 40% based on reported AI inference economics. Market share split is illustrative. Jensen Huang quotes from All-In Podcast, final day of GTC 2026.
