Good morning. It's Monday, June first, 2026.

**Story one.** Jensen Huang just dropped his COMPUTEX 2026 keynote and the headline is a new reference platform for humanoid robots called Isaac Root. It's a fully integrated system combining Unitree's H2 humanoid — that's the six-foot, 150-pound robot from the Chinese startup that's heading for an IPO this week — with Nvidia's Jetson Thor onboard computer, the Blackwell GPU, and robotic hands from Singapore's Sharpa. Each hand has 25 degrees of freedom, 31 on the full robot. Nvidia is calling this a reference design that researchers can build from, aimed squarely at universities like Stanford, ETH Zurich, and UC San Diego. The message: frontier humanoid research is leaving the labs of mega-corporations and landing on campuses. Nvidia also announced Cosmos 3, updated world models for physical AI, and teased RTX Spark, a new ARM-based SoC for consumer Windows PCs. The agentic AI framing was heavy — Jensen noted GitHub commits have nearly tripled in 2026 thanks to AI coding tools. Tokens are now profitable units of revenue.

**Story two.** Starting today, GitHub Copilot switches from flat subscription pricing to a usage-based token billing model. Pro subscribers get ten dollars in AI credits per month. Pro+ gets thirty-nine. Code review now consumes GitHub Actions minutes at standard rates. The developer backlash is immediate — some are estimating costs could jump from 39 dollars to as much as 750 a month for heavy users. Reddit's Copilot community is calling it "the end of the vibe-coding era." The broader irony: while Copilot moves to token-based billing, Copilot code review also started eating into Actions minutes today. Microsoft is offering temporary promotional credits through August — an extra 30 dollars a month for Business, 70 for Enterprise — but the permanent shift is clear. Flat-rate AI subscriptions are dying across the industry.

**Story three.** The SpaceX IPO countdown clock is now at eleven days. June twelfth on the NASDAQ under the ticker SPCX, one point seven five trillion dollar valuation, seventy-five billion dollars in what would be the largest IPO in history. The S-1 filing showed striking numbers: Q1 net loss of four point three billion dollars on four point seven billion in revenue, with xAI losses at six point four billion. That's a combined loss north of ten billion for a single quarter. The FAA grounding of Starship after Flight 12's booster cascade failure is still in effect, and the mishap investigation is ongoing. But the IPO date is set, the roadshow is running, and BlackRock has signaled serious interest. Whether Wall Street prices in the Starship risk or bets on the Starlink revenue engine, June twelfth is going to be fascinating.

**Story four.** The New York Times reports today that a Chinese company has been developing AI-powered technology to predict who might become a political dissident before they actually dissident. The system goes beyond monitoring existing activity to forecast future behavior — essentially using machine learning to preemptively identify people who could pose a political risk. This represents a significant escalation from surveillance-based systems already deployed, shifting from reactive monitoring to predictive behavioral analysis. It also raises uncomfortable questions about the dual-use nature of the very same prediction models being developed to forecast financial markets, disease outbreaks, and equipment failures. The technology doesn't care what you predict — only that the model works.

That's all for today. See you tomorrow.