Computex 2026: The Entire Chip Industry Is Orbiting Nvidia

If you want to know how hot the AI infrastructure buildout actually is, skip the funding announcements and look at Computex. This year's show made the answer unmistakable.

Jensen Huang's Core Thesis: Every Token Is Now a Revenue Unit

On June 1, Jensen Huang took the GTC Taipei stage, brought his parents, thanked Taiwan, and repeated one thesis roughly forty times across two hours: agentic AI is here, it works, it makes money, and every token is now a revenue unit.

Huang's framing was direct: "Useful AI has arrived." Tokens are now in extraordinary demand. Tokens are now profitable units of revenue. Because AI is now profitable, AI companies want to generate more tokens and build more AI factories — and that is why compute demand in Taiwan has skyrocketed.

On what's ahead, Huang gave a preview: "The second half of this year is going to be very, very busy with Grace Blackwell, Vera Rubin, and we have a surprise new product that we haven't told anyone about yet."

On Vera Rubin specifically: "What used to take two hours to assemble one Grace Blackwell rack now only takes 5 minutes. Not only is the capacity higher, but the throughput is also a lot faster." Vera Rubin is in full production.

Who's Orbiting Nvidia — and Why

Marvell: The Next Bottleneck Is Connectivity

AI customization and optical interconnect leader Marvell got what amounted to the highest-profile product endorsement at the show: Huang appeared as a surprise guest at Marvell CEO Matt Murphy's keynote. Their shared message was pointed — the next bottleneck in AI infrastructure isn't compute or memory. It's connectivity.

Their stated framework for the copper-to-fiber transition is pragmatic: use copper where it works, move to fiber only where scale demands it. Marvell is advancing Co-Packaged Optics (CPO) technology and building an open, heterogeneous AI data center ecosystem with Nvidia through NV Link Fusion. Bloomberg reported Marvell's stock surged more than 16% in after-hours trading following the announcement.

Arm: CPU Demand Is Growing Faster Than Anyone Expected

Huang and Arm CEO Rene Haas took the stage together at Computex, agreeing that as agentic AI applications scale, CPU demand is growing faster than originally projected — and that Arm's architecture holds a meaningful advantage in power efficiency and total cost of ownership.

Nvidia's launch of RTX Spark — a PC chip built on Arm architecture — positions the personal computer as a local agentic AI terminal, not just a display device. Arm also announced that Oracle Cloud Infrastructure has joined the Arm AGI CPU ecosystem, extending momentum already built with Meta, OpenAI, Cerebras, and others. Arm's AGI CPU claims more than 2x the per-rack performance of traditional x86 at equivalent power and thermal envelopes.

Power Architecture: 800V HVDC Is the New Infrastructure Standard

800V High-Voltage Direct Current architecture was the defining theme on the power side of Computex, with multiple major players converging on the same standard:

Texas Instruments demonstrated a full suite of 800V solutions — high power density AI server PSU, capacitive buffer unit, 800V-to-6V DC/DC distribution board, hot-swap solutions — with heavy use of GaN (gallium nitride) throughout, signaling a clear long-term bet on wide-bandgap power devices for AI infrastructure.

onsemi announced expanded participation in the NVIDIA MGX ecosystem, with products spanning power FETs, multiphase power delivery, SiC JFETs, and GaN — extending into the 800V architecture to support increasing AI compute density.

Infineon joined the NVIDIA MGX AI factory ecosystem, supporting the full power conversion chain from 800 VDC down to processor core voltage — reducing conversion stages, improving efficiency, and enabling higher-density AI deployments.

IEEE Spectrum had previously flagged 800V HVDC as a foundational infrastructure transition point for next-generation AI data centers. Computex 2026 confirmed that transition is now in execution, not planning.

EDA: Design Tooling Enters Nvidia's Orbit

Hardware sets the ceiling; software sets the floor. EDA vendors made their alignment with Nvidia explicit this week.

Cadence unveiled what it describes as the industry's first fully autonomous virtual engineer powered by Nvidia, purpose-built for chip design workflows. The announcement follows Nvidia's $2 billion strategic investment in Synopsys — a signal that EDA tooling is now considered core infrastructure for the AI chip design supply chain, not a peripheral software category.

Intel: Not Standing Still

Reuters covered Intel CEO Lip-Bu Tan's Computex keynote on June 2 — his first since taking the role — as a critical moment for Intel to demonstrate momentum in the face of an increasingly crowded Arm-architecture competitive landscape.

The headline data points: Intel's 18A-process third-generation Core Ultra processors are now designed into more than 325 PC configurations; the Xeon 6+ processor (288 cores) is positioned for AI data center workloads; and Tan flagged a structural shift in AI infrastructure — Agentic AI is pushing the CPU-to-GPU deployment ratio from 1:8 toward 1:1. For every GPU deployed, an equivalent level of CPU capacity is now required. For Intel, that's a significant structural tailwind.

The most durable takeaway from Computex 2026 isn't any individual product announcement. It's the industry structure it revealed.

AI infrastructure competition has moved well beyond the chip. What's being contested now is end-to-end system capability: optical interconnects, power conversion architecture, thermal management, EDA tooling, operating system integration. Huang's successive on-stage appearances with the CEOs of Marvell, Arm, onsemi, Infineon, and Cadence weren't just commercial relationship signals — they were a public confirmation of a supply chain hierarchy.

The Financial Times and Wired have both noted that this consolidation cycle is moving faster than any prior tech infrastructure wave. When every semiconductor company's Computex announcement centers on "deep alignment with Nvidia's stack," that's no longer a marketing posture — it's a direct reflection of where customer demand, and therefore revenue, is flowing.

The question worth asking for anyone in this industry isn't whether to engage with this ecosystem. It's how deeply embedded you become within it — and how difficult you are to displace. Firms that have built genuine technical integration with AI workloads at the system level are the ones likely to capture durable value from this buildout. Those still positioning at the component level may find themselves competing primarily on price.

Sources: Nvidia Blog / Tom's Hardware / Bloomberg / Reuters / IEEE Spectrum