Kimi K3 (2.8T MoE) + NVIDIA LeRobot GR00T 1.7. Massive open models for long-context reasoning meet physical AI platforms for humanoid robots. More frontier inference + real-world deployment = exploding GPU/edge compute demand. The race isn't just training anymore. It's agents that think, see, and act. #ComputeRace# #AIAgents#

AMD just crossed $1 trillion. The AI compute race keeps minting giants. According to Reuters, @AMD reached the milestone for the first time on Monday, becoming the fourth US chipmaker to do so, alongside Nvidia, Broadcom and Micron. Investors are betting on AMD’s expanding role in AI computing. A trillion-dollar vote of confidence in the silicon powering the buildout.

AI companies that let their ego get in the way and do not buy compute from SpaceX will fall behind in the AI race. The main reason is bc AI is becoming a compute race and Earth has limits like power, cooling, land, permits, water. Every major AI company is fighting for the same finite resources, and building new data centers takes years. Space does NOT have those limits. And SpaceX is already proving there’s massive demand for its AI infrastructure. Companies are already paying $ billions for access to its compute bc the fact is there simply isn’t enough capacity on Earth. Now think about what happens when SpaceX expands beyond ground-based clusters. The long-term vision is orbital AI compute powered by continuous solar energy in space, with no water needed for cooling, no crowded power grid, and room to scale far beyond what Earth can support. If that vision works, the cost and capacity advantages will become enormous. If you’ve been following the AI race, you know this is the game of iteration. The company with more compute can train larger models, run more experiments, improve faster, and serve customers at lower cost. History shows that every major computing shift creates winners and losers. Companies that ignored the cloud spent years trying to catch up. And I believe this shift will be multiple times bigger. The AI companies that secure SpaceX compute early will gain a compounding advantage. The ones that do not will eventually find themselves competing with one hand tied behind their back. The future leader of AI will be about who has access to the most compute. And my bet is SpaceX is the answer.

This week was about agentic AI. xAI launched Grok 4.5. Meta released Muse Spark 1.1. OpenAI introduced ChatGPT Work. AI is moving from answering questions to completing workflows. More agents mean more inference. More inference means more GPU demand. The AI race is becoming a compute race.

⚡️SPACE DATA CENTERS ARE NO LONGER SCI-FI Polymarket now price a 36% chance that a space-based data center launches by the end of next year. The AI compute race is getting so extreme, even orbit is becoming part of the infrastructure story.

TL;DR: A new five-platform environment lets you train and evaluate agents that combine GUI operation with coding, and it reveals that even top models look far more capable than they actually are under strict behavioral testing. Title: RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents URL: Key points 🖥️ Covers Ubuntu, macOS, Windows, Android, and Web 🔍 "Recreation" tasks: rebuild a running reference app from scratch, no source code 🧪 RecreationBench: 250 tasks scored via both programmatic and visual assertions 📈 Fine-tuning on 35k recreation trajectories lifts OOD benchmarks by up to 17.9 points 🏆 Top model GPT-6 Astra scores 58.1% overall ⚠️ Yet it passes every programmatic test on only 2.8% of tasks 🔧 Agents flexibly swap Electron apps for native GTK/AppKit without losing fidelity It's a sharp reminder that surface-level recreation and true behavioral fidelity are still worlds apart. #AIAgents# #ComputerUse#

Huge T3 Code release just dropped. Over 380 changes. Quick highlight reel of my favorites: - in-app PR reviews - `npx t3 triage` to debug and fix issues with T3 Code - `npx t3 connect` support on Mac (run 1 command to make a Mac into a remote dev machine) - 93.8%+ performance improvement for rendering long threads - 80%+ reduction to data transferred and stored for threads - In-app terminal performance overhaul - Remote server upgrades are 5x faster and 100x more reliably - Remote projects can now be opened in VS Code via ssh with 1 click - OpenVSX support for custom themes! Use your favorite VS Code themes directly - Codex skill discovery is way more reliable - OpenCode skills actually appear in app now - Claude can discover project-level skills now - Fixed really annoying auto scroll behaviors and edge cases - Claude Code compaction recommendations now surfaced in app - Codex permission requests for MCP/ComputerUse stuff now surfaced in app - Tons of improvements to Windows app - More polish than I can fit into one tweet Sorry this release took so long. I hope it was worth it. Check out the nightly builds if you want to get these features when they're (mostly) done instead of waiting for us to cut a stable release 🫡

🚨🇨🇳 PENTAGON’S WORST NIGHTMARE: CHINA’S J-36 BECOMES FIRST SIXTH-GENERATION FIGHTER PROTOTYPE SEEN MAKING SHARP MANEUVERS China has publicly demonstrated something no other sixth-generation fighter program has yet revealed: a giant tailless prototype performing a sharp turn-climb. New footage of the aircraft unofficially known as the J-36 challenges the assumption that a stealth aircraft of this size must handle like a slow flying-wing bomber. 🔸 The J-36 is significantly larger than China’s J-20 and uses an unprecedented three-engine configuration. Its size points toward a long-range air-dominance platform carrying large fuel reserves, powerful sensors and heavy internal weapons loads. 🔸 Unlike conventional fighters, the J-36 has no vertical stabilizers or horizontal tailplanes. Removing them reduces radar-reflecting surfaces but also eliminates much of the aerodynamic leverage normally used to control pitch and yaw. 🔸 Chengdu compensates with computerized flight controls and numerous movable surfaces along the wing’s trailing edge. Recent prototypes have also displayed exhaust configurations widely interpreted as two-dimensional thrust-vectoring nozzles, potentially allowing the engines themselves to help control the aircraft. 🔸 The latest footage shows the huge aircraft entering a sharp turn while climbing and remaining stable throughout the maneuver. The exact speed, angle of attack and G-load remain unknown, but its ability to change direction rapidly without losing control is clearly visible. 🔸 It demonstrates something more important for J-36's intended role: China appears to be combining the range, payload and stealth of a much larger aircraft with far greater maneuverability than previously expected from a tailless design. 🔸 The footage may show as many as the fifth distinct flying J-36 airframe. Visible changes to the nose, intakes, landing gear and exhaust system suggest Chengdu is testing multiple configurations rather than relying on a single technology demonstrator. China’s sixth-generation prototypes have been publicly flying since December 2024. The U.S. Air Force says secret experimental aircraft have already completed hundreds of flight hours, but the final F-47 remains in engineering development and has yet to appear publicly. The real warning for the Pentagon is therefore bigger than one impressive maneuver. China is rapidly building, flying and refining multiple versions of a sixth-generation combat aircraft while the American public has seen only illustrations of its future rival. What should worry the Pentagon more: the J-36’s agility or the speed at which China is testing new prototypes?