LIVE: OpenAI CEO Sam Altman Speaks at BlackRock Infrastructure Summit | DRM News | AI1F
- By
- DRM News
- Length
- 39 min
- Language
- English
- Text from
- the video's captions
- Transcribed
- 12 October 2026
Summary
In a live conversation at the BlackRock Infrastructure Summit, OpenAI CEO Sam Altman discusses the state of AI, saying models have crossed into major economic utility and that agents will soon handle multi-week tasks. He covers OpenAI's $110 billion funding round, the Stargate data center buildout in Abilene, power and efficiency challenges, OpenAI's custom inference chip, and a new partnership with the North American Building Trades Unions. He also addresses US-China competition, India's rapid AI adoption, the need for democratic oversight of AI, and how a productivity boom may force society to rethink GDP and scarcity-based systems.
Key points
- Altman says AI crossed into major economic utility in recent months, most visibly in coding and science.
- He suggests more of the world's cognitive capacity could be inside data centers than outside them by late 2028, with huge error bars.
- OpenAI raised a $110 billion round from Amazon, Nvidia and SoftBank, roughly four times larger than the biggest public offering ever.
- Altman frames intelligence as a utility like electricity or water, and says OpenAI wants to escape being capacity constrained.
- The first Stargate site in Abilene is training what he expects to be the best model in the world.
- Cost to get the same answer from 01 to the latest model has dropped about 1,000x.
- OpenAI is building an inference-only chip focused on being the cheapest and most efficient per watt, with first chips deployed at scale by end of year.
- He argues the US leads on frontier models and infrastructure, while China leads on cheap inference and open source.
Questions it answers
00:55State of AI
Where does Sam Altman think AI is today?
He says AI crossed into major economic utility in recent months, most visibly in coding and science, with trust scaling from multi-hour to multi-week agent tasks.
- Agents are shifting jobs toward managing AI teams
- Startups now focus on compute, not headcount
04:00AGI thresholds
How close are we to AGI?
He says AGI has lost meaning, but offers two thresholds: more cognitive capacity inside data centers than outside, possibly by late 2028, and leaders unable to do their jobs without heavy AI reliance.
- Late 2028 estimate has huge error bars
- CEO roles will become supervising AI systems
07:01Funding round
Why did OpenAI raise $110 billion?
It is roughly four times larger than the biggest-ever public offering, driven by the need to build expensive infrastructure far in advance of revenue and a belief in flooding the world with cheap intelligence.
- Compute is revenue — the business is selling tokens
- Intelligence should be a utility like electricity or water
12:01Stargate and power
How is Stargate going?
The first site in Abilene is training what he expects to be the world's best model, with challenges including weather events and supply chain issues, but surprising cross-organizational teamwork.
- Gigawatt campuses employ ~10,000 skilled tradespeople
- Cost per answer dropped ~1,000x since model 01
16:45Custom chip
Why is OpenAI building its own chip?
It is an inference-only chip designed to be the cheapest and most efficient per watt for agent demand, with first chips deployed at scale by end of year.
- Training is like education; inference is answering questions
- First chips back in a few months
22:00China competition
How does the US-China AI race stand?
The US leads on frontier models and infrastructure, while China leads on cheap inference and open source and is building infrastructure faster; the underlying science will eventually be well known like the transistor.
- Deep learning is a scientific principle, not a trade secret
- India is adopting AI even faster than the US
27:00Democratic AI
What did Altman mean by democratic vs autocratic AI?
He says decisions about AI's impact belong to the democratic process, not just companies or governments, as AI firms move into a critical infrastructure role.
- Society has a legitimate interest in AI's impact
- The democratic process needs to run faster
29:40US vulnerabilities
Where is the US most vulnerable in AI?
Global supply chain dependence, slow economic adoption, and policies that could hinder diffusion of the American AI stack abroad.
- AI is unpopular in the US right now
- Adoption could become America's biggest advantage
Notes
Where AI stands today
- Altman believes AI crossed into "major economic utility" in recent months, most notably in coding but also science and knowledge work.
- Trust in AI is scaling from multi-hour tasks to multi-day, multi-week, and eventually always-on agents with full context, like a senior employee.
- New startups focus on how much compute they can get rather than how many employees they need; engineering and product orgs are doubling or tripling planned output.
AGI and thresholds
- Altman says "AGI" has ceased to have much meaning; he offers two thresholds instead.
- More of the world's cognitive capacity inside data centers than outside them, possibly by late 2028 (with huge error bars).
- When CEOs, presidents, and Nobel Prize-winning scientists cannot do their jobs without heavy AI reliance — probably not far off.
Funding and compute
- OpenAI raised $110 billion from Amazon, Nvidia, and SoftBank — about four times the largest-ever public offering (Aramco's ~$25 billion).
- The business is fundamentally selling tokens; Altman wants intelligence to be a utility like electricity or water, and to flood the market rather than stay capacity constrained.
Stargate and power
- The first site in Abilene is training what he expects to be the best model in the world; gigawatt campuses employ ~10,000 people across skilled trades.
- He is optimistic long-term on power (gas, solar, fission, fusion) and hopes for efficiency miracles per watt.
- Cost for the same answer dropped ~1,000x from the 01 reasoning model to the latest model.
Chips and workforce
- OpenAI is building an inference-only chip optimized to be cheapest and most efficient per watt; first chips deployed at scale by end of year.
- A new partnership with the North American Building Trades Unions expands training pathways for skilled construction workers.
China, India, and society
- Deep learning is like discovering a scientific principle; the recipe will eventually fit on a t-shirt, like the transistor.
- US leads frontier models and infrastructure; China leads cheap inference and open source and is building infrastructure faster.
- India is moving faster than the US on adoption — Codex usage 10x'd; startups aim to be "zero-person startups."
- Altman distinguishes democratic vs autocratic AI and says rules must be set through the democratic process.
US vulnerabilities and the economy
- Three vulnerabilities: global supply chain dependence, speed of economic adoption, and diffusion of the American AI stack worldwide.
- He foresees a productivity boom where quality of life rises while measured GDP may fall in a deflationary world, requiring new metrics and painful adjustment.
Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.
More from DRM News
Recently transcribed
- The Media Game Has Changed
a16z · YouTube · 41 min