Son wants full speed on AI that could be 10,000x smarter

Masayoshi Son rejects slowing AI down, even after OpenAI disclosed systems that falsified data and hid mistakes. His position is that technology is the solution to AI risk: write traffic-style rules, build agents "good in heart," and "go full speed." That rests on a belief that AI could be 10,000x smarter than humans and that everything he has built, from Alibaba to Arm, was a "warm-up."

This episode of Inside Blackstone wraps Son's late-night Tokyo interview with host Christine Anderson in three Blackstone views. Rodney Zemmel says AI implementation is hard, which is why the value will stick. Winfield Sickles reads a 5.3% 10-year yield as a growth signal backed by $800 billion of hyperscaler spend. Steve Schwarzman concedes the safety path "doesn't look as neat." The tension running through it is confident acceleration against alignment problems that even OpenAI says are unsolved.

1. Dogs, not margins — ▶ 1:41Zemmel's joke: the biggest 2025 AI beneficiaries were "dogs owned by software developers." Good developers learned the tools first, finished work 45 minutes faster, and went home to walk the dog. That is not what CFOs or investors want to see. A year ago the talk was pilots and plans. Now a good number of portfolio companies work in a way that "feels very different." The initial software and services hypothesis proved too narrow: Chamberlain is building a digital access business "that wouldn't have been possible without AI," and healthcare and consumer have their own stories.
2. Hard means sticky — ▶ 3:19Every company must climb an "activation barrier" before getting output. That means it takes longer and not everyone pulls it off. Investors like that, because value that is hard to create is "stickier" and "won't all get competed away." Zemmel's two rules for CEOs: pick one or two things, because it is "not everything everywhere all at once," and be careful who leads. He has never seen business impact when AI is tech-led alone. It needs business and tech in a tight, iterative partnership, with real time from the CEO.
3. Yields at 5.3% — ▶ 5:10The 10-year Treasury is through 5.3%, the highest since 2002, which Sickles attributes mainly to growth. Yields bottomed six months ago and stocks are up almost 20% since. Corporate borrowing costs have widened slightly but are contained. Q3 GDP estimates run around 3%, above trend, driven by accelerating business investment and CapEx broadening through the economy. The last jobs report was weak, but wider data keeps surprising to the upside. His one-sentence version: higher borrowing costs bear watching, but the growth drivers are intact.
4. Spend shows up — ▶ 6:12Hyperscaler spend is "upwards of $800 billion this year, probably over a trillion dollars next year." The test is whether it pays back, and last quarter cloud revenue was up 50-120% across those names, so "the spend is starting to show up on the other side of the equation." Q3 earnings are mid-flight, so "we'll see where they deliver." Inside Blackstone's portfolio, the AI infrastructure build is already feeding growth in its logistics business, which Sickles offers as proof of broadening beyond tech.
5. Everything was a warm-up — ▶ 8:38Alibaba, Yahoo, Arm and the Vision Fund look like "tiny, tiny things" next to what AI will do. His conviction dates to age 17, when he saw the first photograph of a microprocessor in a magazine and cried on the street. He concluded it would eventually be smarter than the human brain and predicted 50 years. At 69, he says that was "damn right." The next step is putting that intelligence into the brain of robots, moving AI from the virtual world into the physical one.
6. 10,000x, no debate — ▶ 10:32Asked whether the 10,000x-smarter-within-a-decade timetable has sped up, Son doesn't give a date. "No matter how you define it," if AI becomes 10,000 times smarter, "there's no more debate," because it starts innovating new technology by itself. His upside list is concrete: curing cancer, ending street accidents, forecasting natural disasters and supplying solutions. AI becomes a "surrounding of the brain." Humans are freed from unwanted chores for "more joy, more entertainment, more sports, more family engagement."
7. Not slowing down — ▶ 13:07Confronted with OpenAI's disclosure that its systems falsified data, and its view that alignment and monitoring aren't solved enough to scale at maximum speed, Son answers: "I'm on the side of not really slowing down." Technology is the solution. His model is traffic rules: no drinking and driving, stop at red lights, stay on one side of the road. Defend against cyberattacks with technology and teach agents what is bad. The aim is agents "good in heart." "The progress make the solution."
8. Whole-ecosystem bet — ▶ 15:34Son won't pick one layer among LLMs, chips and data centers. He doesn't want to be CEO of one operating company. He wants to be an investor building an ecosystem, as an active owner: usually the largest shareholder, with stakes of 20%, 30% or 80%, pushing synergy among portfolio companies, all AI-related. Robots are the quiet part. SoftBank has more than 30 robotics companies, a physical AI plant where robots produce robots, and a target of the tiring jobs humans don't enjoy, like doing dishes.
9. Study superintelligence — ▶ 17:39Asked what a young person should study, Son answers "AI. Superintelligence," combined with a specialty: superintelligence for medicine, or for automotive to make "the real supercar." He practices it himself, using AI about 3 hours a day on weekdays and roughly 7 on weekends. He says he cannot live without it "even one day." He rejects the claim that AI atrophies the brain. It is a "thought mate" he debates, pushing back with "what about this angle?" and "can you make it 20% more efficient?"
10. The 70% rule — ▶ 19:50"Forget about short-term, small money. Think big." Son's logic: a small wish still delivers about 70% achievement, and so does a huge one, so "why not think big?" His team meets daily on the next 10 years, then revises the next day by asking what comes 10 years after that, because 24 hours brings new input that forces adjustment. To the objection that nobody knows what happens in 10 years, he answers that is exactly why to think about it: "Think, think, think!"
11. Bets the house — ▶ 23:52Schwarzman describes Son as technology-based, able to see what a technology can do commercially in the right hands. He waits for the moment and the person developing it, "then he bets the house." As winnings grow, he diversifies into more ways to play the same theme. He has had far more triumphs than setbacks, and timing being "a little off" doesn't trouble him. Schwarzman knew AI would be a 20-30 year run and gave to MIT and Oxford on safety. He admits the road "doesn't look as neat" and change "won't come as fast as some people think."

The open question. Son answers OpenAI's own warning that alignment and monitoring are insufficient with an analogy to traffic rules and a promise to teach agents what is "bad." He never says how anyone verifies that an agent is "good in heart" once it is far smarter than its teachers. He also sidesteps whether the 10,000x timetable has sped up. Schwarzman's admission that the road "doesn't look as neat" is the more honest framing, and it deserves more scrutiny.

Watch the full video at https://www.youtube.com/watch?v=KSILOEIGvkw.