Jensen just dropped the biggest consumer hardware announcement in years. The RTX Spark chip -- built on N1X CPU (ARM, co-developed with MediaTek) + Blackwell GPU -- puts 128GB unified memory and RTX 5070-class graphics into laptops. Microsoft unveiled Surface Laptop Ultra as the launch vehicle. NVIDIA framed this as "the first completely re-engineered PC in 40 years." AMD fell 4.9%, Intel down similarly, Qualcomm dropped 8.5%. Dell and HP both up 7%+.
Also announced: Vera Rubin in full production (early customers: Anthropic, OpenAI, SpaceX/xAI, Dell, Oracle, CoreWeave) and NVIDIA Vera CPU designed specifically for AI agents -- Huang called it "our new major growth driver" targeting a $200B market.
NVIDIA launched Cosmos 3 at GTC Taipei -- the first fully open omnimodel that combines vision reasoning, world generation, and action prediction. Mixture-of-transformers architecture. Natively generates text, images, video, ambient sound, and robot actions with physics accuracy. Reduces physical AI training cycles from months to days. Also launched the Cosmos Coalition with Agile Robots, Black Forest Labs, Runway, and Skild AI.
NVIDIA unveiled a standardized humanoid robot reference platform: Unitree H2 chassis (6 ft, 150 lbs, 31 DOF) + Sharpa Wave tactile hands (22 DOF) + Jetson AGX Thor T5000 compute (Blackwell GPU, 2,070 FP4 TFLOPS). Stanford and UCSD among first academic users. Reuters reports NVIDIA plans similar partnerships with US, European, and Korean robot makers. Geopolitical angle: US lawmakers pushing to ban Unitree robots from government-funded research due to Chinese military ties.
Cognition closed Series D (May 27): $1B+ raised, $26B post-money valuation (up from $10.2B in September). ARR hit $492M. Enterprise customers include Goldman Sachs, Citi, Mercedes-Benz, Dell, US Army, US Navy. CEO Scott Wu says Devin writes 89% of Cognition's own code. Mercedes-Benz reduced an 8-month legacy migration to 8 days. Cognition acquired Windsurf earlier in 2026.
Anthropic shipped Opus 4.8 (May 28) with Claude Code v2.1.154. Key feature: Dynamic Workflows -- Claude writes a JS orchestration script, spins up hundreds of subagents in parallel, consolidates results. Research preview for Enterprise/Team/Max plans. Opus 4.8 claims 4x fewer self-generated code defects vs 4.7. Bridgewater Associates highlighted its ability to proactively flag input/output issues. Early practitioners are documenting the sharp learning curve.
Jensen posted a recap of the keynote: "AI learned to generate. Then it learned to reason. Now it can do useful work." Lists Vera Rubin, Vera CPU, RTX Spark, Agent Toolkit, Cosmos 3. The comment section is flooded with generic congratulations and spam.
Multiple high-engagement Threads posts today about Jensen's keynote -- @aiposthub (6,579 likes, 54 comments), @knews_taiwan (4,526 likes), @qndzy (551 likes, 60 comments). Most are surface-level recaps or hype. Threads audience skews less technical than X.
Medium post "What Opus Got Wrong Building a Claude Code Dynamic Workflow" is getting traction. The author documented exactly where Opus 4.8 guesses wrong about the new feature API. Practitioners are sharing real experiences vs the marketing pitch.
The Cognition raise is generating debate across X and LinkedIn about whether $26B for a coding agent company is justified at $492M ARR (~53x revenue multiple). The "Devin writes 89% of its own code" stat is getting the most attention and the most skepticism.
The RTX Spark announcement is a supply chain event, not just a product launch. Frame it through the lens of who wins and who loses in the silicon food chain:
This is your unique lane. Most takes will be product reviews. Your take should be the semiconductor and supply chain analysis.
Three announcements today that form one coherent strategy most people are covering separately:
NVIDIA is doing for robotics what they did for AI training: owning every layer of the stack. The Cosmos Coalition (Agile Robots, Skild AI, Runway, BFL) is the robotics equivalent of the CUDA ecosystem moat. The geopolitical wrinkle with Unitree adds a spicy dimension -- NVIDIA solving the "Chinese robot hardware with US compute" problem through cybersecurity gatekeeping on the Jetson chip.
Broadcom reports Wednesday after close -- the last major semi to report this cycle. Expected: $22B revenue (+47% YoY), $2.40 EPS (+52%). AVGO is at an all-time high ($460). HSBC just raised PT to $600. Options market pricing a 9% swing. Then Friday's jobs report feeds directly into the June 16-17 Fed meeting. Two data points that together tell us whether AI infrastructure spend is still accelerating or whether the semi rally is outrunning fundamentals.
Pairs naturally with the semiconductor thesis. Time it for Wednesday morning pre-earnings for maximum relevance.
Threads is blowing up with NVIDIA/Computex content today but the conversation is almost entirely from media accounts and tech news aggregators. @aiposthub has 6,579 likes on a recap post. @knews_taiwan has 4,526 likes. The signal-to-noise ratio is low -- lots of "wow" reactions, very little analysis. There's a clear opening for a builder's perspective that explains what the RTX Spark and Cosmos 3 actually mean for developers and investors.
Between Cognition at $26B, Claude Code Dynamic Workflows, and NVIDIA's Agent Toolkit (NemoClaw, Nemotron 3 Ultra, OpenShell, CUDA-X Skills), "AI agent infrastructure" is crystalizing as a category this week. Nobody on Threads is connecting these dots yet. Most Threads content treats each announcement as isolated news.