BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list. Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%. Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything. Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO. h/t @OliverMolander Would you move your team off frontier APIs if the bill dropped 90%?

Why Masa Son Calls His Entire Career a Warm-Up for AI Masayoshi Son, founder, chairman and CEO of SoftBank, interviewed by Christine Anderson (Inside Blackstone) Masayoshi Son founded SoftBank in 1981 with the proceeds from an electronic translator he sold as a student at UC Berkeley. He backed Alibaba, Yahoo and Arm and later started the Vision Fund. SoftBank now holds the largest stake in OpenAI and invests across AI chips, models, data centers and robotics. Summary Son says everything he has done so far was a warm-up for AI, which he expects to be 10,000 times smarter than people within a decade. He opposes slowing down and wants AI agents taught good intent, in the way traffic rules were written for cars. SoftBank owns large stakes across the AI stack, with more than 30 robotics companies in its portfolio, and Son uses AI three hours a day on weekdays as a thought partner. His advice is to think big, because the result is proportionate to the dream. 1. A Fifty-Year Warm-Up. Son says everything he did before AI, including Alibaba, Yahoo, Arm and the Vision Fund, was "tiny, tiny things" and a warm-up. He dates his conviction to age 17, when he saw the first photograph of a microprocessor in a magazine and cried on the street, sure it would change the future of mankind. He predicted it would take about 50 years for microprocessors to become smarter than people's brains, and he says it has now been 50 years. The brainpower now sits in data centers. 2. Ten Thousand Times Smarter. Son says that if AI becomes 10,000 times smarter than people, within a decade in his view, the debate is over. An AI that smart starts inventing new technology by itself. He expects the change to reach the physical world as well as the virtual one, as intelligence moves into the brains of robots. 3. More Fun, Fewer Dishes. Son says humans will be happier if AI cures diseases such as cancer, prevents accidents on the street and forecasts natural disasters with solutions. He pictures AI as a surrounding brain that makes people more productive. People would no longer have to do what they did not want to do, and would have more joy, entertainment, sports and time with family and friends. Doing dishes is his example of a tiring job that robots could take over. 4. Full Speed, Good Intent. Asked about OpenAI's disclosure of incidents where AI systems falsified data and hid mistakes, Son says he is on the side of not slowing down, because technology development is the solution to those issues. Rules are needed, as they were when cars arrived: no drinking while driving, stopping at red lights, staying on one side of the road. Cyberattacks should be countered with technology, and AI agents must be taught what is bad so they are good in heart as well as capable. His summary is to go full speed with good intent built into the system. 5. Own the Whole Stack. Son will not pick one part of AI. He invests across models, chips, data centers and robotics, because the AI generation is starting and history is being made. He does not want to run one operating company as CEO; he wants to be an investor who builds a group of companies. SoftBank is not a passive investor: it is usually the largest shareholder, at 20%, 30% or even 80%, close to each CEO and connecting portfolio companies to each other. The portfolio includes more than 30 robotics companies that Son says are quietly preparing. 6. Study AI Plus Your Subject. Asked what a young person should study today, Son answers AI, meaning superintelligence. He adds that you should combine it with your field of interest: superintelligence for medicine, or for automotive, which he calls the real supercar. The same holds for aircraft or whatever business you want to be a specialist in. Pairing AI with a specialty, he says, gives that subject its future. 7. A Thought Partner. Son says he uses AI about three hours a day on weekdays and about seven hours a day on weekends, and cannot live even one day without it. He treats it as a thought partner. When the AI offers an answer, he debates it, asks about another angle and asks whether it can be made 20% more efficient. He calls it "my thought mate now." 8. The 70 Percent Rule. Son's advice to investors is to forget short-term, small money and think big, because the result is proportionate to your dream. He says that whether the wish is small or large you still get about 70% of it, and nobody gets 100% easily. So the sensible choice is a big wish. 9. Ten Years, Revised Daily. Son says SoftBank's daily meeting covers the plan for the next 10 years, and the next day the team revises it and asks what comes 10 years after that. His reason is that in 24 hours they get more input and have to adjust a lot. Nobody knows what will happen in 10 years, he says, so the answer is to keep thinking about it: "Think, think, think!" 10. Humans Pick the Lineup. Asked whether he would let ChatGPT choose the opening lineup for the SoftBank Hawks, his Japanese baseball team, Son says no. That is the job of the general manager, who has plenty of capability and the trust of the players. Part of the fun, he says, is that it is not only calculation.