Nvidia's Dual-Pronged Strategy: AI in TSMC's Fabs and a PC Chip Challenge to Intel and AMD

Nvidia unveils RTX Spark superchip for Windows laptops and supplies AI tools to TSMC fabs, driving shares up 2% but still 8% below 52-week high.

Nvidia is making a bid to become indispensable in two distinct corners of the technology world. One is the highly specialised domain of semiconductor manufacturing, where TSMC is now deploying the company's accelerated-computing libraries and vision AI inside its most advanced fabs. The other is the fiercely competitive consumer PC market, where Jensen Huang unveiled the first Nvidia-designed processor for Windows laptops at his GTC Taipei keynote. The dual offensive has caught the attention of investors, sending shares to €185.32 on the day – a gain of roughly 2% that still leaves the stock about 8% below its 52-week high of €201.05.

The PC gambit: RTX Spark and a desktop supercomputer

The biggest surprise to come out of Taipei was the RTX Spark, a superchip that marries a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor cores to a 20-core Grace CPU built on the Arm architecture. Connected via Nvidia's NVLink-C2C interconnect, the chip delivers up to 1 petaflop of AI performance and 128 GB of unified memory – enough to run large language models entirely on-device without cloud access. Microsoft, Dell, HP, ASUS, Lenovo and MSI have all signed on to launch laptops carrying the chip from autumn 2026. The move marks Nvidia's first direct assault on a turf long dominated by Intel, AMD, Qualcomm and Apple.

For corporate customers, Huang also unveiled the DGX Station for Windows, a deskside AI supercomputer powered by the GB300 Grace Blackwell Ultra chip. With up to 748 GB of coherent memory and 20 petaflops of FP4 compute, it can handle frontier models of up to a trillion parameters locally. The system is due in the fourth quarter of 2026, with system partners including ASUS, Dell, GIGABYTE, HP, MSI and Supermicro.

Deeper into chipmaking: TSMC's fab-floor AI

Only days before the keynote, Nvidia announced that TSMC had started using its cuLitho GPU-accelerated library for computational lithography, along with a suite of other tools for design, simulation and production control. The move pushes Nvidia deeper into the semiconductor value chain: instead of merely selling AI accelerators for data centres, it is now supplying the infrastructure that underpins the fabrication of the very chips that power the AI boom. TSMC is also deploying CUDA-X libraries, the cuEST chemistry-simulation library – which Nvidia claims can speed up materials simulations by a factor of 50 – and the Metropolis platform with the TAO Toolkit for defect inspection on wafers. Vision AI is being used to classify nanometre-scale defects, reducing the need for repeated labelling and retraining.

The economic logic is clear. Better defect detection cuts scrap, tighter process control improves yields, and faster simulation shortens development cycles. Nvidia also highlighted Omniverse as a potential tool for building digital twins of fabs, allowing TSMC to test layout changes and investment decisions before committing physical resources. The financial impact of the TSMC deal is not yet quantifiable – Nvidia does not disclose contract terms – but the strategic positioning is significant: if Nvidia's platform becomes the standard compute layer for chip manufacturing, its addressable market extends well beyond AI servers.

Robotaxis, world models and a Taiwan mega-campus

Huang's keynote also fleshed out Nvidia's ambitions in autonomous mobility. The DRIVE Hyperion platform for Level-4 robotaxis has added Foxconn, which plans to launch services in Taiwan in 2028 before expanding across Asia; VinFast, targeting Southeast Asia; Uber, with a pilot in Munich; and Humain, bringing the platform to Saudi Arabia. On the software side, Cosmos 3 was introduced as an open AI world model that unifies image understanding, world generation and action prediction. The new Vera CPU, meanwhile, is already in full production, with OpenAI, Anthropic and SpaceX among its first customers.

Nvidia's deepening ties to Taiwan were underscored by Huang's disclosure that annual spending in the country has risen from $10–15 billion to $100 billion. A new campus dubbed "Constellation" is planned to open around 2030 as Nvidia's overseas headquarters, housing roughly 4,000 employees.

Record revenue and generous capital returns

All this product activity is backed by a strong financial foundation. For the full fiscal year 2026, Nvidia posted revenue of $215.9 billion, up 65% year on year, with an operating margin of around 60%. In the first quarter of fiscal 2027, revenue reached $81.6 billion – an 85% increase from the prior year – with the data-centre segment alone contributing $75.2 billion, up 92% year on year and 21% sequentially. Guidance for the second quarter stands at $91.0 billion, plus or minus 2%, though that forecast excludes data-centre compute sales to China, underlining the geopolitical constraints on the growth story.

Shareholders are being rewarded generously. The board authorised an additional $80 billion in share buybacks on 18 May, and the quarterly dividend has been raised from $0.01 to $0.25 per share, payable on 26 June to holders of record on 4 June. In the first quarter alone, Nvidia returned roughly $20 billion through buybacks and dividends. With the stock still trading below its recent peak, the market is now weighing whether the Taipei product blitz and the TSMC partnership can together ignite the next leg of the rally.