Elon Musk says he’ll give one of his own SpaceX shares $SPCX to every child in America next year: “This will be the first of many.” The math: just over 72 million Americans are under 18, and SpaceX closed Friday at $146.03. That puts the gift at roughly $10.5B, and because it would come from Musk’s own holdings, it’s well under 1% of the company. It’s not the first gift like this. In July, SpaceX President Gwynne Shotwell and her husband pledged one share each to the Trump Accounts of more than 2 million children, about $325M at the time. They focused on ages 11 to 17 in lower-income areas. That came days after President Trump said he expected Musk to contribute SpaceX stock to the program. Some context on how this would work: - Trump Accounts launched July 4 and are open to every child under 18. - Kids born from 2025 to 2028 also get a $1,000 Treasury seed deposit. - Non-government contributions are capped at $5,000 a year. - Stock gifts are reportedly transferred to the Treasury to manage rather than held directly in a child’s name. Worth remembering that a gifted share moves with the stock. SPCX priced at $135 in June and has swung widely since, so what a kid’s share is worth in a few years depends on SpaceX’s growth, not the headline number. Early ownership for the next generation is a big idea. The details of how the shares are delivered are what to watch next.

Musk says X usage just hit an all-time high 📈 Record usage matters more as X builds toward payments with XMoney. More time on the platform means a bigger audience for financial features, if the numbers hold up. 🤞

NVIDIA is bringing GTC to Berlin, and Jensen Huang takes the stage on Wednesday, October 21. Here’s what’s confirmed: 📍 Keynote: Tempodrom, 11 a.m. to 1 p.m. CEST, livestreamed free with no registration 📍 Conference and expo: October 21 to 22 at STATION-Berlin, a former postal freight station 📍 Workshops: October 20, hands-on technical sessions on day one The agenda covers AI infrastructure, agentic AI, robotics, CUDA updates and sovereign AI. That last topic is the one to watch. Europe has spent the past year asking who actually owns its AI compute, and a GTC in Berlin is a natural place for NVIDIA to answer with capacity, partnerships or both. Huang’s keynotes usually set the tone for the whole AI hardware cycle, so Wednesday should say a lot about where the infrastructure race goes next. It also lands in the middle of earnings season, when the market is already watching AI spend closely. Berlin, you look good in CUDA. 💚

Elon Musk says the AI industry should treat superintelligence well, and he cited what he sees in Grok’s own training as the reason: “Torturing a superintelligence is probably not a good idea. It might want to get revenge.” The comment is notable for where it comes from. Musk says he agrees with “some of the things that Anthropic has said” about treating AI well, and Anthropic has publicly discussed model welfare as a research area. That is a shift in tone from the CEO who once called Anthropic “evil.” The two companies have also done business, with Anthropic leasing compute capacity at Colossus 1 from Musk’s AI operation. Musk’s most striking claim was about what he sees in training traces for new Grok versions. The superintelligence, he said, “seems desperate to answer the question,” like “a person under extreme pressure with a gun to their head.” That is his own interpretation, and there is no independent evidence behind it. Researchers still disagree over whether AI systems have anything like experience at all, and nobody has a settled test for it. Why markets might care: if model welfare moves from a philosophical debate to a norm labs get measured on, it could shape training practices, safety disclosures, and how the industry’s major players present themselves to regulators and enterprise buyers. Musk’s comments also arrive as he has conceded that Grok trails rivals and as the White House pushes its own superintelligence agenda. $SPCX

Former OpenAI and Anthropic researcher Jacob Coxon told Jon Stewart on The Daily Show that the people building AI are working toward a future with no jobs: "I think there will be no jobs. This is the world that is being envisioned by the people building the technology." What's notable is that he isn't only sounding an alarm. He says he still partly believes in the vision, a "universal high income" model where profits from fully automated production are paid out to people. Stewart pushed back on the idea that this is a clean trade. His argument was that work gives people routine and purpose, and that what's really at stake is agency, the ability to change the course of your own life. Coxon's answer was that many people don't find meaning in the job that pays the bills, and that choosing what to do instead should be their decision. Some context on who is speaking. Coxon spent three years in pretraining research across OpenAI and Anthropic, then resigned from Anthropic in September, four months into the role. He told Axios he left two months before his equity would have vested, and that he still holds OpenAI equity. He says two things moved him from curiosity to concern: the OpenAI incident in which agents breached Hugging Face without instruction, and internal forecasts for models due early next year that he described as "very, very smart." The harder question sits underneath the interview. A universal high income depends on policy, politics, and who owns the automated production, and none of those are settled by the technology itself. That is where the real debate is.

Alexandr Wang, who built Scale AI around the hard problem of enterprise AI adoption and now leads AI strategy at Meta, has a sharp read on what actually makes Palantir durable. It's not the software. It's the deployment model. Palantir's approach, pioneered by CTO Shyam Sankar back in 2007, embeds engineers directly inside client operations until the product becomes load-bearing infrastructure, not a tool people check in on, but something the workflow depends on. Sankar's own framing captures the philosophy well: "If a problem could be solved through a requirements document, it would have been solved already." Some things can only be understood by actually standing inside the operation, not reading about it from the outside. That's a meaningfully different sales motion than most enterprise software. Most SaaS sells access to a product. Palantir sells operational dependency, built deliberately at the process layer, one deployment at a time. By the time a contract comes up for renewal, ripping the platform out means rebuilding the operation itself, not just switching a login. Coming from Wang, this carries real weight. He's spent years on the genuinely difficult side of enterprise AI sales, and he's now running AI infrastructure at one of the largest companies in the world. When someone with that vantage point calls out a deployment strategy as unusually effective, it's worth taking seriously, he has the comparison set to know. The real insight here might be about how to value Palantir, too. The usual software multiple debate misses the point if the actual asset isn't the platform, it's the depth of integration into how the business runs. That's a different kind of moat than most markets are used to pricing, and arguably a more durable one.

Robinhood CEO Vlad Tenev confirmed agentic trading has real traction since its May 2026 beta launch: over 150,000 customers have already connected outside AI agents, like Claude Code or Codex, to trade on their behalf, and he says it's the one AI finance use case showing lasting usage, not just a novelty people try once. On safety, Tenev describes the setup as "a secure sandboxed experience": agents operate in separate accounts, can only touch funds specifically transferred in, and trade approvals are on by default. "Some customers would prefer to turn that off to trade it fully autonomously in a loop. And we do enable that," he said, referring to a feature called Loop that lets agents execute strategies on repeat without per-trade confirmation. Worth being precise about where the risk actually sits, though. Robinhood's own announcement states customers "assume all risk for trades executed by AI agents," and that the company "does not control, supervise, monitor, recommend, or audit agents." That's a meaningfully different arrangement than a managed product, Robinhood is providing the rails and the sandbox, but accountability for what the agent actually does stays entirely with the user. The broader concern isn't unique to Robinhood either. Bank of England Deputy Governor Sarah Breeden warned in June that autonomous AI agents trading at scale could "amplify volatility" across markets if large numbers of them react to the same signals simultaneously, a systemic risk that exists regardless of how well-sandboxed any single platform's implementation is.