At G20 event, Sam Altman says AI use is 'nonnegotiable' and argues for more data centers
- Length
- 32 min
- Language
- English
- Text from
- the video's captions
- Transcribed
- 13 October 2026
Summary
At a G20 event, Sam Altman traces his path from AI enthusiast to OpenAI's co-founder, describing the scaling-law insights that led from GPT-1 to GPT-4 and soon a new model. He argues that AI adoption is non-negotiable for countries, comparing it to electricity, and predicts an unprecedented boom in entrepreneurship. He calls for abundant, inexpensive AI backed by massive data center buildout, while flagging cybersecurity, biosecurity, and concentration of power as risks that could set the technology back.
Key points
- Sam Altman describes being an AI and sci-fi nerd who gave up on AI in school, then refocused after deep learning began working around 2012 and co-founded OpenAI at the end of 2015.
- He says the moment it became clear AI was bottling knowledge came around GPT-4 in March 2023, which OpenAI had internally for about eight months before launch.
- Altman predicts AI will drive the greatest boom in entrepreneurship and small business creation the world has ever seen, citing a laundromat owner who ran his business with ChatGPT.
- He argues that countries adopting AI is non-negotiable, comparing rejecting AI to rejecting electricity, while allowing that countries will differ on regulation and on building versus renting data centers.
- He projects token usage growing from a global average of 100,000 tokens per month to roughly 100 billion tokens per month in another six and a half years, requiring vastly more infrastructure to keep AI abundant and inexpensive.
- He identifies cybersecurity and biosecurity as urgent challenges, and warns that failing to make AI abundant could concentrate power and worsen inequality.
- Altman positions OpenAI as pragmatic centrists who want to be reliable partners, avoid catastrophic risks, respect sovereignty, and stay on team humanity.
- He frames the history of technology as one continuous exponential built by successive generations, not isolated revolutions, and says missing any part of it leaves a country unready for what comes next.
Questions it answers
00:16Altman's origin story
How did Sam Altman get into AI and start OpenAI?
He was an AI and sci-fi nerd who gave up on the field in school, returned after the 2012 deep learning breakthrough, and co-founded OpenAI at the end of 2015.
- OpenAI's first day of work was January 2016
- Deep learning getting better with more compute drew him back to AI
04:12From GPT-1 to GPT-4
When did it become clear AI was capturing knowledge?
Around GPT-4 in March 2023, after OpenAI had the model internally for about eight months, even internal skeptics concluded the approach would go very far.
- GPT-3 in 2020 was the first model over a usability threshold
- Scaling laws suggested a trillion dollars of compute would yield something amazing
06:06Entrepreneurship boom and non-negotiable adoption
What does Altman say AI means for entrepreneurship and for countries?
He predicts the greatest boom in entrepreneurship and small business creation ever, and says countries' use of AI is non-negotiable, like electricity, though they may build or rent data centers as they choose.
- A laundromat owner ran his business with ChatGPT
- Coding agents accelerated things in 2025; persistent agents are next
10:35One continuous revolution
How should governments think about AI compared to past technologies?
He views history as one continuous exponential rather than isolated revolutions, and says countries that fail to enable their people with AI won't invent the next level of scaffolding.
- Each generation builds on the scaffolding of the last
- There will be bigger revolutions after AI
14:22OpenAI's positioning
What does Altman say differentiates OpenAI?
He describes OpenAI as pragmatic centrists, reliable partners that respect sovereignty, avoid catastrophic risks, don't want to be the only company, and are firmly on team humanity.
- He compares disliking some AI uses to disliking some uses of electricity
- OpenAI wants the whole global economy to succeed
19:08Electricity analogy and risks
Why does Altman call AI adoption non-negotiable, and what could go wrong?
He compares rejecting AI to rejecting electricity, predicts people will stop noticing AI in a decade, and stresses that cybersecurity and biosecurity must be navigated urgently or adoption will be set back.
- A kid growing up today will never be smarter than AI
- Too much concentration of power is another major risk
24:29Token growth and infrastructure
How much compute growth does Altman project?
He projects per capita token use rising from 100,000 to about 100 billion tokens per month in another six and a half years, meaning far more data centers are needed to keep AI abundant and cheap.
- Top user went from 100,000 to hundreds of billions of tokens per month
- He wants AI as cheap as an hour of light at night
28:59Optimism and worry
Is Altman optimistic about AI's future?
He says worry is how society anticipates problems, as with fire safety, and remains optimistic that human ingenuity can mitigate AI's downsides given the huge economic and personal value at stake.
- He met someone who says they wouldn't be alive without ChatGPT
- Doomers five years ago wouldn't have expected things to go this well
Notes
Sam Altman's path to OpenAI
- Altman was a self-described AI and sci-fi nerd who studied AI in school, gave up when nothing worked, and pursued entrepreneurship instead.
- The 2012 breakthrough in deep learning, which improved with more compute, drew him back; OpenAI was announced at the very end of 2015 with work starting January 2016.
- Early work included training a robot hand to solve a Rubik's cube and playing video games; unsupervised learning progress came around 2018 with GPT-1, and GPT-3 in 2020 was the first model over a threshold of usability.
The GPT-4 turning point
- Altman dates the consensus moment that AI was bottling knowledge to GPT-4, around March 2023, which OpenAI had internally for about eight months before launch.
- He says the new model launching soon is another step forward, capable of discovering new knowledge, doing science, and creating complex software.
AI and the economy
- He predicts the greatest boom in entrepreneurship and small business creation ever, illustrated by a laundromat owner who used ChatGPT for marketing, legal contracts, and vendor relations.
- 2025 brought acceleration with coding agents and white-collar work agents; he expects long-term persistent agents that act as virtual collaborators.
- He compares rejecting AI to rejecting electricity: adoption is non-negotiable, though countries may choose different regulations and whether to build or rent data centers.
OpenAI's positioning
- Altman calls OpenAI the pragmatic centrists, avoiding blind optimism and doomism, wanting to be a reliable partner that respects sovereignty while constraining catastrophic risks.
- He says OpenAI does not want to be the only company or suck up all the value, and is firmly on team humanity.
Infrastructure and abundance
- Token usage grew from a top user at roughly 100,000 tokens per month in early 2020 to hundreds of billions per month by mid-2026; the global average went from zero to 100,000 tokens per month.
- Naively projecting forward, per capita use could reach about 100 billion tokens per month, requiring much more infrastructure to keep AI abundant and inexpensive.
- He draws on the falling cost of an hour of light at night, from 5 hours of labor to near-zero, as the model for making AI abundant.
Risks and optimism
- Cybersecurity is a top challenge requiring urgent action; biosecurity challenges are bigger ones yet to come.
- Too much concentration of power is another risk; failing to make AI abundant would make it a hotly competed, expensive commodity.
- He argues worry is warranted, citing fire safety as an example of humanity learning from accidents, and remains optimistic that downsides can be mitigated.
Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.