Public transcript

We Interviewed Sam Altman at TreeHacks 2026

By: TreeHacks
Length: 44 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Sam Altman returns to Stanford for a TreeHacks interview, reminiscing about dorm life, steam tunneling, and his first startup Looped, which he funded through the first Y Combinator batch. He traces the origins of OpenAI around the idea that scaling deep learning mattered, recalls the GPT moments that convinced him something big was happening, and predicts current sophomores will graduate into a world with AGI. The session closes with audience questions on competing with OpenAI, its cash burn, his happiness, and OpenAI's custom silicon plans.

Key points

Sam Altman lived in Donner at Stanford and recalls steam tunneling as a freshman, including an incident where a professor's car fell into an open manhole.
His first startup, Looped, began as a late-night dorm project and was funded after applying to the first Y Combinator batch; he says business plan competitions were a waste of time.
He argues Stanford's special access to VC funding is mostly over, and dropping out felt low-risk because Stanford let students stop out for eight years.
OpenAI started in 2014-2015 around the single idea that scaling deep learning mattered, at a time when AGI was seen as 100 years away.
He says the GPT series was the moment he knew something important was about to happen, and that current sophomores will graduate into a world with AGI in it.
He believes the next ChatGPT moment after coding agents will be AI getting things done across all knowledge work, and that one-person startups will reach the revenue of medium-sized companies.
On OpenAI's cash burn, he says it is not serious because model growth correlates with future revenue growth and there is plenty of capital willing to fund it.
He reveals OpenAI has a big custom silicon program and an inference chip it is excited about, trains on Nvidia, and would prefer not to build data centers itself.

Questions it answers

00:00Stanford memories

What does Sam Altman remember about his time at Stanford?

He was a Donner resident with fond memories of move-in day, steam tunneling (where a professor's car fell into an open manhole), and all-nighters in the basement of Gates.

Steam tunneling led to a professor's car falling into an open manhole.
His best Stanford experience was building projects with friends rather than clubs.
All-nighters in Gates became happy memories in retrospect.

02:21Looped and Y Combinator

How did his first startup get started and funded?

Looped began as a late-night dorm project, won a business symposium, and he applied to YC's first batch through PG's form. He raised over $30 million and says dropping out felt low-risk because Stanford allowed stopping out for eight years.

Business plan competitions were a waste of time; YC changed what it took to get funded.
Stanford's special VC access is mostly over now.
He advises pursuing whatever offers the highest rate of learning and impact.

10:27OpenAI origins

How did OpenAI get started?

After AlexNet in 2012 and watching deep learning scale, he and co-founders including Greg Brockman and Ilia pushed the unpopular idea that scaling deep learning mattered, starting as a pure research lab in Greg's apartment.

The pitch was that others were undervaluing deep learning and scale.
The core ideas of OpenAI's first four years were written on a whiteboard in week one.
OpenAI was never meant to be a product company initially.

18:00GPT moments and caution

When did OpenAI know it had discovered something big?

The GPT series was the turning point; by GPT-2 he knew something important was coming. GPT-2 was initially withheld out of precaution, though he says that looks silly in retrospect.

Dota, the robot hand, and language models felt like miracle after miracle.
He believes in slight caution but warns against an AI capability overhang.
Coding agents were the next ChatGPT moment; all knowledge work is next.

22:03AGI predictions

What does Altman predict about AGI and society?

He predicts current sophomores will graduate into a world with AGI, with science automated and startups and jobs transformed, though human adaptability and desires will persist.

One-person startups may reach medium-sized company revenue.
He is not a jobs doomer; humans will still find fulfilling work.
An AI CEO of OpenAI will likely outperform him at some point.

34:24Q&A: competition, money, happiness, hardware

What did Altman say in the audience Q&A?

He says Google's slow reaction let OpenAI exist, that the cash burn is not serious given revenue growth, that he is genuinely happy thanks partly to meditation, and that OpenAI has a big custom silicon program with an exciting inference chip and a deep Nvidia partnership.

Startups should build things only possible with AI, not copies of ChatGPT.
He wakes early, works, plays with his kid, then faces chaotic days.
He would prefer not to build data centers himself.

Notes

Stanford memories

Sam Altman was a Donner resident and has fond memories of his first day on campus.
Early in freshman year he went steam tunneling; a friend left a manhole cover off and a professor's car fell in.
His best experience was hacking on projects with friends; in hindsight he would have done fewer clubs and more building.

Looped and Y Combinator

Looped started as a late-night dorm project, not intended as a company, and won a business symposium competition in spring 2005.
He applied to YC's first batch via a form from PG that had to be pasted into an email unchanged.
Before YC, founders were told they needed MBAs, business plans and VC connections; he calls business plans and business plan competitions a waste of time.
He raised over $30 million, which he jokes is about 17 GPUs today.

Dropping out

Stanford let students stop out for eight years, which made dropping out feel low-risk and helped get his parents on board.
He advises working on whatever offers the highest rate of learning and impact, and says the default path has gotten relatively worse while the option space has expanded.

OpenAI origins

He worked in the AI lab between freshman and sophomore year and found AI "was not it" at the time.
AlexNet in 2012 and subsequent scaling results convinced him deep learning was an attack vector worth pushing.
The pitch was simply: something important is happening with deep learning and scale, and others are undervaluing it.
Co-founders: Greg Brockman (known via Stripe), Ilia (stalked at a conference), Wojciech, and connections to John and Elon.
OpenAI was not meant to be a company; the core ideas of the first four years were written on a whiteboard in Greg's apartment in the first week.

GPT and caution

The GPT series was the moment he knew something important was coming; GPT-2 was initially withheld.
He favors slight caution but warns of an AI capability overhang: companies that don't adopt AI fast enough will lose to fully AI-run companies.

AGI and society

Current sophomores will graduate into a world with AGI in it; startups and jobs will be totally different.
He is not a jobs doomer: human desires to be useful, compete, create status and express creativity will persist.
He expects an AI CEO of OpenAI to outperform him at some point soon.

Q&A highlights

Competing with OpenAI: Google's slow reaction let OpenAI exist; startups should do something totally new, not copy ChatGPT.
Cash burn: not serious; spend a billion this year to make $3 billion next year and capital is available.
Happiness: he is very happy now, credits meditation, and reframes the opposite of a bad experience as no experience.
Undervalued areas: a new architecture as big as transformers over LSTMs, and products only possible with AI at the core.
Hardware: big custom silicon program, an exciting inference chip, deep Nvidia partnership for training, and reluctance to build data centers itself.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.