KEY POINTS
- Nvidia CEO Jensen Huang confirmed Vera Rubin — a seven-chip, six-trillion-transistor AI platform built on TSMC 3nm — has entered full mass production at COMPUTEX 2026.
- The announcement extends Nvidia's dominance in data center AI, where it holds 81% market share and reported $75.2 billion in quarterly data center revenue, up 92% year-over-year.
- Traders should watch Broadcom's earnings on June 3 and Nvidia's next quarterly guidance for signals on whether the AI infrastructure spending cycle has peaked or still has room to run.
Jensen Huang took the stage in Taipei on Sunday and gave the semiconductor trade its next catalyst: Vera Rubin is in full production.
The platform — seven custom chips, six trillion transistors, more than 18,000 components on a single board — represents Nvidia's most ambitious engineering project to date, with 40,000 engineers across the company contributing to its development. Built entirely on TSMC's 3nm process, Vera Rubin succeeds the Blackwell architecture that powered Nvidia's record fiscal 2026, a year in which revenue surged 65% to $215.9 billion and data center revenue alone hit $75.2 billion in a single quarter.
The timing of the announcement — during COMPUTEX's opening keynote, with the PHLX semiconductor index already up nearly 80% year-to-date — tells you everything about where the AI hardware arms race stands heading into the second half of 2026.
The Full Stack Play
Vera Rubin is not a single chip but a complete system architecture. The flagship configuration, Vera Rubin NVL72, connects 36 Vera CPUs and 72 Rubin GPUs through sixth-generation NVLink Switch for scale-up, paired with ConnectX-9 SuperNICs and Spectrum-X Ethernet Photonics for scale-out. Huang described it as an end-to-end designed platform, meaning Nvidia controls silicon, interconnect, networking, and software from top to bottom.
That vertical integration matters for Nvidia's margin story. When a hyperscaler buys a Vera Rubin rack, they are buying the entire stack from one vendor — a dynamic that has kept Nvidia's data center gross margins well above 70% even as it ships at record volume.
Huang also unveiled RTX Spark, a consumer-focused "Superchip" designed to bring AI agents, content creation, and gaming to portable devices. It is Nvidia's most direct shot yet at Apple's M-series silicon and Qualcomm's Snapdragon platform in the personal computing market. Further out, Huang mapped the roadmap: Rosa Feynman Spark processors arrive in 2028, with the next consumer generation targeted for 2030.
Agentic AI Takes Center Stage
Beyond hardware, the keynote signaled a strategic pivot toward agentic AI — autonomous software agents that can reason, plan, and use external tools without human intervention. Huang declared that AI is moving from its generative phase into an agentic era, announcing a two-layer system with advanced reasoning and tool orchestration capabilities.
Nvidia released an updated Agent Toolkit centered on NemoClaw, alongside Nemotron 3 Ultra, a 550-billion-parameter mixture-of-experts model purpose-built for long-running autonomous agents. The DGX Station for Windows, powered by the GB300 Grace Blackwell Ultra Desktop Superchip, will deliver 20 petaflops of FP4 performance with up to 748 gigabytes of memory. Systems from Dell, HP, ASUS, MSI, and Super Micro ship before year-end.
The Valuation Question
The semiconductor index's 80% rally in 2026 has been almost entirely a story about AI infrastructure spending. Deloitte estimates the AI chip market will reach $500 billion this year, with up to half of the $1.32 trillion global semiconductor industry's revenue coming from AI data center demand. Nvidia alone has secured a $1 trillion order pipeline.
But concentration risk is rising. Nvidia's 81% share of the AI accelerator market makes every earnings cycle a sector-wide event. And the company's own investment behavior signals it sees adjacencies as critical: Nvidia has deployed more than $40 billion in equity investments in 2026, including $2.1 billion in data center operator IREN and $3.2 billion in Corning.
Broadcom reports fiscal Q2 earnings on June 3, and Wall Street expects AI semiconductor revenue of $10.7 billion, up 140% year-over-year. That report will be the first major test of whether the custom ASIC market — Broadcom's domain — is growing fast enough to diversify the AI trade beyond Nvidia. For now, Huang's keynote gave bulls exactly what they wanted: production confirmation, a consumer roadmap, and a software ecosystem play that locks in the next generation of AI spending.
