How much does it cost to train a frontier AI model?
Guess before you scroll — the answer is below
The answer
$100,000,000
Roughly $100M.
Training a frontier AI model costs on the order of $100 million in compute alone, up from a few million a few years ago. The cost is doubling roughly every year, which is why the number of organisations able to do it keeps shrinking.
How to get close to this number
Multiply hardware by time. Training a frontier model occupies tens of thousands of specialised processors for weeks to months, and renting that capacity costs a few dollars per chip-hour. Chips times hours times rate gives a figure in the tens to hundreds of millions.
The growth rate in the explanation is what matters more than the level: roughly doubling every year. A cost that doubles annually rises by a factor of a thousand in a decade, which means a capability available to a few dozen organisations today was available to thousands a few years ago and will be available to a handful soon.
That is the structural consequence worth naming: the number of organisations able to do this keeps shrinking. When the cost of participating in a field grows exponentially, the field consolidates regardless of anyone's intent — and the compute figure is the mechanism behind every argument about who controls this technology.
Where this number comes from
| Source | Stanford AI Index Report |
|---|---|
| Figure refers to | 2024 |
| Last checked | |
| Unit | dollars |
Numbers move. If this one has drifted from its source, let us know.
Think you can do better?
Five fresh questions every day. No account needed.
Related questions
- What is the median U.S. fixed broadband download speed, in Mbps?
- How many petaflops does the fastest supercomputer on the June 2026 TOP500 list run at?
- How many charge cycles does a smartphone battery last?
- How many minutes does a fast charger take to charge an EV to 80%?
- How many active Android devices are there worldwide?
- How many active satellites were orbiting Earth in mid-2026?