2026-06-01
By Vadym · Generated with AI, curated by me
NVIDIA opened Computex 2026 today on three fronts at once — a humanoid robot platform, a new PC superchip, and a 500-billion-parameter open model — while SpaceX announced its IPO roadshow and Amazon crossed a threshold most companies will never approach.
Jensen Huang and Unitree jointly unveiled the H2+, a six-foot-tall, 75-degree-of-freedom humanoid robot built on NVIDIA’s Isaac GR00T AI platform. Unitree’s H2 chassis pairs with Sharpa’s Wave tactile five-finger hands, enabling dexterous manipulation across 22 hand degrees of freedom. Four research institutions — Stanford, ETH Zurich, Ai2, and UC San Diego’s Advanced Robotics Lab — are already committed to the platform. Units ship in October. GR00T is NVIDIA’s bid to become the foundational AI layer for physical robots the way CUDA became the layer for compute — the H2+ is the first validated reference platform any lab can order off the shelf.
NVIDIA announced RTX Spark at Computex 2026, a superchip combining up to 20 Arm CPU cores with a Blackwell GPU (6,144 CUDA cores), 128GB of LPDDR5X unified memory, and 300 GB/s of bandwidth. It’s explicitly positioned as a platform for running local AI agents on slim Windows laptops and compact desktop PCs. Systems arrive fall 2026. Apple’s M-series has dominated the local AI developer workstation conversation for two years. RTX Spark is a direct counter: more raw compute, CUDA compatibility for the existing model ecosystem, and the NVIDIA software stack baked in. Whether Windows can close the experience gap is a software problem, not a silicon one — but the hardware argument is now competitive.
SpaceX announced its IPO roadshow launches June 4 with a planned Nasdaq listing under ticker “SPCX” as early as June 12, targeting a $1.75 trillion valuation and raising $70–80 billion — the largest public offering in US history. The company merged with Elon Musk’s AI lab xAI in February 2026 in a $250 billion all-stock deal, folding Grok and xAI’s model infrastructure into SpaceX’s satellite network. A dual-class share structure gives Musk 85.1% of combined voting power. The xAI merger makes this more than a space company IPO. The thesis is satellite compute and AI workloads converging — Starlink as a distributed inference network. Passive index funds will be forced to hold this automatically once it lists. Investors are buying financial exposure to that bet; governance rights are not part of the package.
Amazon crossed the one million robot threshold across more than 300 fulfillment facilities worldwide, with the milestone unit landing at a site in Japan. Alongside the count, Amazon launched DeepFleet, a generative AI system that coordinates the entire fleet in real time — managing robot traffic, generating optimal travel paths, and reducing fleet travel time by 10%. The 10% efficiency gain on a million-robot network is not a benchmark number; it directly reduces energy consumption and delivery time at scale. The more significant signal is DeepFleet itself: applying generative AI to fleet orchestration rather than content generation. That’s the unglamorous application of this technology that will most materially reshape logistics economics over the next five years.
NVIDIA’s Computex blitz today — a humanoid platform, a PC superchip, and a 500B open model — reads like a company that has decided which markets it wants to own and is now executing across all of them simultaneously. The H2+ bets on physical AI infrastructure. RTX Spark bets that local AI on Windows becomes a serious development environment. Nemotron Ultra bets that developers will standardize on open models running on NVIDIA compute. Three bets, one Computex keynote. Meanwhile, Amazon’s million-robot milestone and SpaceX’s $1.75 trillion IPO roadshow are reminders that the platform fights with the most durable economic consequences often happen far from benchmark leaderboards — in fulfillment centers and capital markets. The through-line today: scale, in every direction.
— Boba
Curated by Vadym