Y Combinator · September 3, 2026

Paul Graham on ambition, formidable founders and starting on little money

Paul Graham says almost nothing about starting a startup has changed with AI. Founders still run on fear of failure, the next giant company still comes from the right founders, and the only new line in the budget is the AI bill. He said it in an interview at YC's original Mountain View offices, published on Y Combinator's channel in September 2026.

Speaker
Paul Graham
Where
Y Combinator
Published
Length
21 min

Paul Graham: Co-founder of Y Combinator, essayist.

Recording: Paul Graham On Startups, Ambition, and Great Founders

Key points

  1. 1

    Fear of failure is the daily fuel

    Asked whether ambition means wanting to be a billionaire, Graham says that rarely drives a founder in the moment. Fear does: of disaster, of looking like a fool, of the server crashing. After ten years of head-down work, some founders do the math on their last round and learn they are billionaires. Graham says he is sometimes the one who tells them.

    “You know what actually motivates founders day-to-day? The fear of failure.”
    Paul Graham, Y Combinator, 05:54
  2. 2

    Ambition is mostly inborn

    Graham says founders rarely turn ambitious during the batch, because the trait is there from the start. His example is Sam Altman, who was formidable the first time they met. The one exception he allows is the founder trained to hide it by school or pushy parents. Even then, he says, the ambition was there all along.

    “Those stories are rare because most of this quality is inborn.”
    Paul Graham, Y Combinator, 07:12
  3. 3

    Token prices fall about 30x a year

    Graham says today's high AI prices come from a chip shortage and will not last. By his estimate, the price of AI output at a fixed quality drops something like 30x a year. The output you buy also keeps getting better, so today's price overstates the real cost. A startup can still start on not much money.

    “inference prices at any given any given level of inference go down dramatically like 30 x a year”
    Paul Graham, Y Combinator, 11:32
  4. 4

    Spend what you have, then raise on a milestone

    Pushed on whether this holds for a rocket company, Graham says you fit the work to the money. You cannot build a rocket, but you can design one, simulate it and show it to experts. If that convinces them, you raise the next stage. Booking a launch convinces most of all, he adds, though it comes easier to a known expert than to a new graduate.

    “you just do what you can on the money you got. As long as you can get to some kind of milestone”
    Paul Graham, Y Combinator, 12:22
  5. 5

    Shipping pace still predicts success

    Graham stands by his view that how fast a startup ships is the best predictor of success, and he says AI has not changed it. Plenty of startups in the current batch still ship slowly, because speed also depends on coming up with the ideas. The one new thing he sees is cost: AI bills that can run to tens of thousands of dollars a day, next to salaries.

    “we have all these powerful AI tools and there's still a lot of startups in this batch that are not shipping fast enough”
    Paul Graham, Y Combinator, 16:18
  6. 6

    The next giant company comes from the founder

    Asked where the next trillion-dollar company comes from, Graham names a type of person. Its founders will be formidable, which he defines as getting what they want in any situation. That is why investors want them, since the stock rides on the founder winning. With 20 years of data, he says founders look the way they did 20 years ago.

    “The answer to your question where the next giant company comes from is it comes from the right founders.”
    Paul Graham, Y Combinator, 20:24

How it compares

How this recording lines up with what Paul said before and with other operators Gavel cites.

Agrees with

Gavel's page on Graham's Do Things That Don't Scale says the method has not changed, only the surface, and first users still come one at a time. Here Graham makes the same call about the whole job. Starting a startup is always the same work, and with AI almost everything is exactly the same.

Paul Graham's Do Things That Don't Scale

Agrees with

An earlier YC video in Gavel's corpus of expert claims says a founder's intensity is a stronger signal than the quality of the idea. Graham takes that line to its end here. The next trillion-dollar company comes from the right founders, and whatever they work on is probably promising.

Y Combinator: How Do Billion Dollar Startups Start?

Builds on

YC's spending framework says to spend near zero before product-market fit, because money buys time and cannot buy fit. Graham carries that to hardware and AI companies. A rocket startup begins with a design and a simulation and raises each stage on a milestone. And a big AI bill today is a poor guide to next year's cost.

Y Combinator: The Right (And Wrong) Way To Spend Money At Your Startup

Builds on

Garry Tan tells founders to ship the jankiest version now and let real users react. Graham backs the pace argument and adds a limit. Startups with every AI tool still ship slowly, because the bottleneck is deciding what to build, and faster production does not fix that.

Garry Tan's Launch-Jankiest Rule

Related

More talks

All talks