Nvidia’s Taipei Power Play: Record Cash, Desktop AI Supercomputers, and a Bold Laptop Chip Ambush

At Computex, Nvidia unveils RTX Spark laptop chip, DGX Station deskside supercomputer, and Nemotron 3 Ultra model, aiming to turn AI from cost to revenue driver.

Jensen Huang used his Computex keynote in Taipei to redraw the boundaries of Nvidia’s ambition. No longer content to be seen as a chip supplier, the chief executive cast his company as an end-to-end infrastructure provider for artificial intelligence — one that helps customers turn AI from a cost centre into a revenue driver. The message landed with force, and the market responded in kind.

A New Hardware Front: RTX Spark Takes on the PC Establishment

The most striking hardware reveal was the RTX Spark, Nvidia’s first processor designed specifically for Windows laptops. The superchip marries a Blackwell RTX GPU packing 6,144 CUDA cores and fifth-generation tensor cores with a 20-core Grace CPU built on Arm architecture, linked by NVLink-C2C. The result is up to 1 petaflop of AI compute and as much as 128 GB of unified memory — enough to run large language models entirely on-device without a cloud connection.

Laptop makers including Microsoft, Dell, HP, ASUS, Lenovo and MSI are slated to ship machines with the chip from autumn 2026. With this move, Nvidia steps into a battlefield long dominated by Intel, AMD, Qualcomm and Apple.

DGX Station: Billion-Parameter Models on the Desk

For enterprise users, Huang unveiled the DGX Station for Windows, billed as the world’s most powerful deskside AI supercomputer. Powered by the GB300 Grace Blackwell Ultra Desktop Superchip, the system offers up to 748 GB of coherent memory and delivers 20 petaflops of FP4 performance — sufficient to run frontier models with as many as one trillion parameters locally.

The device can serve a single developer or act as a local node for a team, with workloads able to scale into the GB300 data-centre infrastructure or the cloud as needed. ASUS, Dell Technologies, GIGABYTE, HP, MSI and Supermicro are confirmed as system partners, and the launch is set for the fourth quarter of 2026.

Software Ambitions Run Deeper

Nvidia also pushed further into the software layer with Nemotron 3 Ultra, an open-weights model built on a mixture-of-experts architecture with 550 billion parameters. The company claims up to five times faster inference and costs roughly 30% below leading open models. It scored 48 on the US Intelligence Index for open models and can generate more than 300 tokens per second.

Early adopters read like a who’s who of enterprise AI: Accenture, Cadence, CrowdStrike, Cursor, Deloitte, EY, Oracle Cloud Infrastructure, Palantir, Perplexity, ServiceNow, Siemens, Synopsys and Zoom are all integrating models from the Nemotron family. The breadth of that list underscores Nvidia’s strategy of owning not just the silicon but the software stack around it.

On the hardware side, Huang confirmed that Vera, Nvidia’s next-generation CPU, is now in full production. He argued that AI agents will become the largest consumers of compute, and Vera is the first chip purpose-built for that future. Nvidia says it handles tasks 80% faster than x86 systems, with OpenAI, Anthropic and SpaceX among the early users.

Record Financials Fuel the Narrative

The product blitz is backed by numbers that few companies can match. For the first quarter of fiscal 2027, Nvidia posted revenue of $81.6 billion, up 85% year on year and above the $78.9 billion analysts had pencilled in. Diluted earnings per share came in at $1.87, beating the consensus estimate of $1.77 and representing a 140% jump from a year earlier.

The data-centre segment continued to dominate, delivering a record $75.2 billion in revenue — 92% higher than the prior year and 21% ahead of the previous quarter. For the current quarter, management guided for $91 billion in sales, well above the $86.8 billion consensus. The forecast assumes no revenue from China in data-centre compute.

Over the longer term, the financial picture is equally robust. For the full fiscal year 2026, Nvidia generated revenue of $215.9 billion, a 65% increase, with an operating margin near 60%. The board authorised an additional $80 billion in share buybacks in May, and the quarterly dividend was lifted from $0.01 to $0.25 per share, payable on 26 June 2026.

Capital Commitment in Taiwan

Huang also highlighted Nvidia’s deepening roots in the island. Annual spending in Taiwan has ballooned from $10-15 billion to $100 billion, and a new campus dubbed “Constellation” is planned to open around 2030 as the company’s overseas headquarters, housing roughly 4,000 employees.

Beyond the PC and data-centre push, the keynote touched on automotive and robotics. The DRIVE Hyperion platform for Level-4 robotaxis gained new partners: Foxconn plans to launch in Taiwan in 2028 before expanding across Asia, VinFast is targeting Southeast Asia, Uber is working on a programme in Munich, and Humain will bring the platform to Saudi Arabia. On the software side, Cosmos 3 — an open AI world model — was introduced, combining image understanding, world generation and action prediction.

Stock Takes Its Cue

The flurry of announcements had an immediate effect on the share price. On the day of the keynote, the stock rose roughly 2% to €185.32 in Frankfurt. The following session saw further gains, with the shares climbing 4.17% to €188.96. That leaves the equity up 17.29% since the start of the year and 57.36% over twelve months, though still about 8% below its 52-week high of €201.05.

With Computex running until 4 June and further presentations on robotics and networking still to come, the market will be watching to see if Huang’s vision of Nvidia as the entire AI infrastructure stack — not just the chip inside it — continues to gain traction.