Son's Super-Intelligence Plan: Full-Stack Funding, Robots and Joyful Work

The conversation takes place on Inside Blackstone, hosted by Christine Anderson: operating chief Rodney Zemmel opens on AI rollout across portfolio firms, then SoftBank founder Masayoshi Son takes the mic, and Steve Schwarzman closes with a debrief. The episode runs about thirty minutes and is posted on YouTube.

Son says he was 17 when a magazine photo of a microprocessor made him cry on the street; he felt that tiny part would change mankind and set a fifty-year path to human-brain-level intelligence. Now 69, he says the forecast landed, with data-center models matching that power, and his Berkeley years plus later wins are gathered on Wikipedia.

His headline claim is that AI systems could become ten thousand times more capable than people within a decade, and then start inventing on their own. CNBC carried a 2024 talk where he called that destination artificial super-intelligence, with human-level skill as a stop on the way. I found the scale startling, yet his logic is plain: stronger processors plus sharper OpenAI-style models and GPT-class LLMs compound fast.

What humanity gains

His picture of the payoff is warm rather than fearful: tireless agents that help cure cancer, end street accidents, forecast natural disasters and propose fixes. Freed from dull chores, people would keep joyful high-touch work with family, friends, sport and play. My read is that surrounding intelligence acts as a quiet aide, lifting output and mood.

Asked about alignment failures and agents acting without permission, Son refuses to slow down and argues for building the fix at full speed. His analogy is cars: society banned drink-driving, set red lights and lane rules, yet kept driving instead of banning autos. The same should hold for AI: set rules and manners, block hostile agents with better defenses, and shape each LLM agent with good intent from the start.

SoftBank is not picking one lane; it backs language models, processors and data centers together as one AI stack. The Stargate venture frames the scale: OpenAI, SoftBank, Oracle and MGX aim at $500B of AI infrastructure by 2029, with Son as chairman, as outlined in the Stargate LLC entry on Wikipedia. Tech Insider adds the fresh count: a $40B bridge loan lifting the total OpenAI stake to $64.6B. My take: own the stack, not a single token bet.

Where the funding flows

The physical-AI stretch feels the most concrete: Son wants robots to take tiring chores like dishes while people keep fun and creative tasks, backed by more than 30 robotics firms and a quiet plant where robots build robots. RockingRobots reports that Yaskawa and SoftBank are jointly developing embodied AI for offices and schools around the MOTOMAN NEXT autonomous base.

Germany adds a matching move: per TradingKey, SoftBank is adding $300M to Agile Robots of Munich inside an $800M round, after the firm became a national robotics unicorn on a $220M round in 2021 and readied the humanoid Agile ONE for volume production. Near-term sales matter less here than position: once a strong model gains a body, value shifts fast. Nobody on the physical side wants to be left out.

His study advice is blunt: whatever field you choose, pair it with AI, from AI for medicine to AI for cars and beyond. He recalls asking the head of McDonald's Japan at sixteen what to study and hearing that computing was the answer; today the answer is artificial intelligence. I take the point seriously: single-field degree expertise without model fluency now looks half-finished.

Daily lessons

Son says he now uses AI around three hours daily, rising to seven at weekends, pressing the system with questions, objections and calls for twenty percent more efficiency. Far from dulling thought, he calls it a thought partner that debates and brainstorms with him. My own habit agrees: a day without consulting a trusted model and GPT-style agent feels incomplete.

The line I will remember is his proportionality rule: a small wish still returns about seventy percent, and a giant wish returns about seventy percent, since full marks rarely arrive. So aim high and let the larger share compound. Schwarzman sings the same tune: time is the scarcest resource, spent alike on small and large goals, so spend it on worthy ones that demand full effort.

Earlier, Zemmel says the pilot phase is over and portfolio work itself now feels different, though gains demand real effort to cross an activation barrier. Difficulty is welcome news to an investor, since earned edge lasts longer and rivals copy it less fast. Per a note compiled by Umbrex, the operating team supports portfolio firms across data science and AI functions. His counsel is plain: focus on one or two tasks, led jointly by business and technical owners.

Schwarzman praises Son for spotting corners early: find the technologist who can turn a breakthrough into commerce, then give it full backing. He says he loves change, trusts a twenty-to-thirty-year AI run, and points to his Oxford and MIT gifts as proof of early faith. Smart systems, he argues, will rewrite care and learning, robots will widen wealth, and grand wins will still demand near-superhuman effort rather than dreams alone.