An AI startup CEO says his company spends $10 million to $50 million a year in compute for each researcher. And now some tech candidates are reportedly asking how much compute comes with the job before they accept an offer.
Salary. Bonus. Equity. Benefits. And…GPUs?
Welcome to recruiting in the AI era.
$10M to $50M in Compute Per Researcher?
The number that got my attention came from Pim de Witte, CEO of AI startup General Intuition.
In an interview about building General Intuition, de Witte said the company spends roughly $10 million to $50 million per year in GPU resources per researcher.
That number sounds bananas.
But when you look at what’s happening across frontier AI companies, it isn’t as far-fetched as you might think.
And I think there’s a bigger recruiting story hiding inside that giant number.
Apple: A Small Team With a Huge Compute Budget
Apple gives us one of the more interesting comparisons.
Reporting on Apple’s foundational-model work described a small team of roughly 16 people with training costs reaching millions of dollars per day.
Now, I wouldn’t take a daily training budget, multiply it by 365, divide it by 16, and call the answer “compute compensation per researcher.” Apple didn’t disclose that.
But the comparison still matters.
It shows how a surprisingly small group of people can have an enormous amount of computing power behind them.
Thinking Machines Is Going Even Bigger
Then there’s Mira Murati’s Thinking Machines Lab.
In 2026, Thinking Machines and NVIDIA announced a multi-year partnership to deploy at least one gigawatt of next-generation NVIDIA systems.
That’s an enormous amount of infrastructure for frontier AI work.
Again, we can’t divide the value of that infrastructure by the number of researchers and call it a per-person benefit. That’s not how the companies reported it.
But it shows where the competition is heading: elite people paired with enormous computing resources.
OpenAI and Anthropic Are Spending Billions on Compute
The same pattern shows up at the biggest AI labs.
The Stanford 2026 AI Index, using Epoch AI data, estimates OpenAI’s annual compute spending at roughly $16.3 billion in 2025 and Anthropic’s at roughly $6.8 billion.
Those aren’t per-researcher numbers. They’re company-level estimates that include different kinds of compute spending.
But they give us a sense of scale.
At frontier AI companies, compute isn’t just another IT expense. It’s one of the key resources employees need to do the work.
Meta Is Thinking About “Compute Per Researcher”
This connects to something I wrote about recently.
In 7 Lessons on Recruiting AI Talent (From Meta’s Head of AI), I looked at the idea of “compute per researcher.”
Meta CEO Mark Zuckerberg has talked about compute per researcher as a competitive advantage in the race for AI talent.
But I think the idea is bigger than GPUs.
Compute is really a proxy for resources per employee.
Top candidates don’t only ask, “What will you pay me?”
They also want to know:
- What tools will I have?
- How much budget will I control?
- How much support will I get?
- Can I move quickly?
- Will this company help me do my best work?
AI researchers just happen to put a giant dollar sign on those questions.
NVIDIA Makes the Recruiting Connection Obvious
NVIDIA CEO Jensen Huang took this idea one step further.
As Business Insider reported, Huang said he’d be “deeply alarmed” if an engineer earning $500,000 wasn’t consuming at least $250,000 worth of AI tokens per year.
Think about that.
The AI resources available to an employee could be worth half their base salary.
Business Insider also reported that NVIDIA was aiming to spend as much as $2 billion on tokens for its engineering workforce.
Huang even described a new recruiting question:
“How many tokens come with my job?”
That’s when this stops being an AI infrastructure story and starts becoming a talent acquisition story.
Candidates Are Starting to Ask About Compute
And some candidates apparently aren’t waiting for employers to bring it up.
Business Insider reported that some tech candidates are asking during interviews how much AI compute they’ll have access to if they take the job.
Compute could become another thing candidates consider alongside the familiar pieces of an offer:
Salary + Bonus + Equity + Benefits + Compute
I don’t think that means employees will suddenly choose $20,000 worth of GPUs instead of $20,000 in salary.
The evidence is stronger for something simpler: access to AI compute is becoming a work resource and recruiting perk.
What If You’re NOT Recruiting Frontier AI Researchers?
This is where I think the trend gets useful for most talent acquisition teams.
Most employers aren’t hiring AI researchers who need $50 million in GPUs. They don’t need to.
Frontier AI companies are showing us an extreme version of a much more ordinary question:
What does your company give employees to help them do great work?
For an AI researcher, the answer might be millions of dollars in compute.
For a recruiter, it might be AI sourcing tools, automation, better job description software, and fewer hours of manual work.
For a designer, it could be premium creative AI tools.
For a salesperson, it could be AI prospecting tools and better customer data.
Those resources have value to employees, even if they never appear on a paycheck.
Should AI Tools Show Up in Your Job Descriptions?
I think the answer is increasingly yes.
If your company gives employees meaningful access to AI tools, tell candidates.
Don’t write:
“You’ll work in an innovative, technology-driven environment.”
Please. We’ve all read that sentence 900 times.
Tell candidates what “innovative” actually means.
Which AI tools can they use? Does the company pay for premium versions? What technology will they have from day one? Are there AI or token budgets? What annoying tasks can they automate?
Those details tell candidates far more about what working for you will actually feel like.
And as I wrote in our guide to recruiting AI talent, the best candidates want horsepower. They want to know whether they’ll have the tools, support, budget, and freedom to do their best work.
Maybe the lesson from $50 million-per-researcher compute budgets isn’t that every employer needs more GPUs.
Maybe it’s this:
The resources attached to a job are becoming part of the job offer.
Why I Wrote This
At Ongig, we spend a lot of time looking at what candidates actually need to know before they apply for a job.
Salary matters. Benefits matter. But candidates also want to understand what they’ll have available to succeed once they get there.
AI compute is an extreme example, but the recruiting lesson applies to almost every employer. If your company gives people great technology, AI tools, support, or other resources that make their jobs better, don’t hide those details behind vague language like “innovative workplace.”
Ongig helps employers turn those specifics into clearer, more useful job descriptions and job postings at scale.
Request a demo to see Ongig in action.
FAQs
What does compute per researcher mean?
Compute per researcher describes how much computing power is available to an AI researcher. That can include GPUs, AI tokens, cloud infrastructure, and other resources used to train and run AI models.
Do AI companies really spend $10 million per researcher?
General Intuition CEO Pim de Witte has said his startup spends roughly $10 million to $50 million per researcher per year in GPU resources. That’s an extreme frontier-AI example, not an industry-wide average.
Is AI compute becoming an employee perk?
There’s growing evidence that access to AI compute is becoming a valuable work resource and recruiting perk. Some tech candidates are reportedly asking about compute access when evaluating jobs.
Why should talent acquisition teams care about AI compute?
Because compute points to a much larger recruiting trend. Candidates want to know what resources they’ll have to succeed. For one employee that’s GPUs. For another, it could be premium AI software, automation, better data, or other tools.
Should employers mention AI tools in job descriptions?
Yes, if those tools meaningfully affect the employee’s work. Specific information about technology, AI tools, budgets, and resources can give candidates a clearer picture of what they’ll actually have available on the job.
