Yesterday at the UN General Assembly, President Trump announced that the U.S. will refer to artificial intelligence as "super intelligence" in official documents. A new name reopens an old question: what can today's AI actually do, and what should we call it? Speakers debated that question on the RAISE stage in Paris this July. @ylecun argued that current systems excel at language and code but still lack a cat's understanding of the physical world. He questioned whether scaling today's models leads to general intelligence at all. @istoica05 noted there is no clear threshold for what makes a model "too powerful," and argued that many concerns about AI risk are overblown. President @EmmanuelMacron and @arthurmensch made the case for European independence and open, shared development. U.S. Under Secretary of State @jacobhelberg set out the administration's view: let builders move as boldly and as fast as possible. The people shaping this technology see it from different vantage points and reach different conclusions. Those conversations belong in the open, in front of everyone they affect. All full sessions are available to watch free and ungated on the RAISE YouTube channel. What would you call it?

Two words are quietly separating the companies winning with AI from the ones just spending on it: value maxing. @julien_lepine, Director, Technology at @AWSFrance, made that distinction on our stage at RAISE Summit. AWS powers the cloud infrastructure behind a huge share of the AI being built and deployed today, which puts Lépine in the room for how enterprises are actually using it, not just experimenting with it. He noted that the real question companies need to answer isn’t which model to use, it’s how to fundamentally change the way they work. bit.ly/3V4lfWL and Generali in Italy are already restructuring operations around AI and seeing revenue returns in the hundreds of millions. The companies getting those results aren’t chasing “token maxing,” burning through usage for its own sake. They’re focused on value maxing: making sure every bit of AI spend converts into real business outcomes. Is your organization still measuring AI adoption by usage, or by actual value returned? To hear the full conversation, watch it on our YouTube channel. Link in comments.

Every time you send a prompt to a closed AI model, you’re not just getting an answer back. You’re handing something over. @vipulved, Co-Founder & CEO of @togethercompute, made that case on our stage at RAISE Summit. Together AI runs the AI acceleration cloud behind some of the fastest-growing generative AI workloads in the industry. Going from $0 to $100M in ARR faster than almost any company in the space, built specifically to give businesses more control over how and where their models run. Vipul’s argument is that proprietary data is one of a company’s most valuable strategic assets, and every time it’s sent through a closed model API, that business takes on real risk. As these models get smarter, that risk compounds, where the data flowing through the API isn’t just being processed, it’s effectively training someone else’s intelligence. Feed a closed model your data long enough, and you’re not just paying for a service. You’re handing over your IP. Do you think enterprises are underestimating this risk, or is it a fair trade-off for the convenience of closed models? Full conversation on our YouTube channel. Link in comments.

Starting a company used to require a team, a budget, and technical cofounders. On stage at RAISE Summit 2026, @mcuban and @antonosika described the moment those requirements stopped applying. "A million new projects? Every single week?" Cuban asked. "Yeah. Every single week," Osika answered, along with more than 700 million monthly visits to the applications people had built. Cuban, an investor in @lovable, wasn't reacting to the scale for its own sake. He was naming what it replaces: "Having started a bunch of companies, one of the brilliant things about Lovable is you can be an entrepreneur, have had that idea... but taking that next step to get started" no longer needs the infrastructure it used to. The number wasn't the headline. The shift in who gets to have that idea and act on it was.

"There's not going to be any one company, Google included, that can solve the problems up and down the stack." Amin Vahdat, Chief Technologist/SVP, AI and Infrastructure at @Google, revealed offstage the quiet truth behind every conversation we hosted at RAISE Summit 2026. He'd just spent twenty-five minutes on stage walking through what it takes to build intelligence at scale, energy, silicon, infrastructure, models, applications. From the podium it sounded like an engineering problem. Offstage, it became an admission: if the stack runs that deep, no single keynote can cover it either. AI's biggest breakthroughs aren't happening in silos. They're happening at the intersection: hyperscalers and startups, researchers and builders, infrastructure and application. No single lab, however powerful, owns the whole stack. The ecosystem moves forward together, or it doesn't move forward at all. That's the room we set out to build at RAISE, where the people shaping each layer can come together. Watch Amin's full keynote on our YouTube channel. Link in comments.

Congratulations to RAISE 2026 speakers @annadgoldie and @Azaliamirh, who were named to @TIME 100 AI 2026 last week. The co-founders of @RicursiveAI built AlphaChip at Google, a deep reinforcement learning agent that generates superhuman chip layouts in hours rather than weeks. It's already deployed across five generations of Google's TPU, Waymo chips, Pixel phones, and Axion CPUs. And the recursive part is already running: the next generation of AlphaChip was trained on TPUs whose layouts were designed by the previous version. At Ricursive, they're going further. Their tools run up to 1,000x faster than leading commercial tools. The vision is what they call the "designless era", where any company, from frontier AI labs to robotics startups, gets custom chips built for their specific workload without needing hundreds of chip design experts in-house. As Anna put it at RAISE Summit: "It shouldn't take a year or more to design a chip. And it shouldn't involve hundreds or thousands of human experts." Watch the full session: bit.ly/4xB1LaD