Nvidia and Microsoft Announce Arm-Blackwell Chip and Windows Agentic to 'Reinvent the PC' 40 Years Later

Jensen Huang opened the keynote at GTC Taipei on Sunday, 31 May, and used the phrase that defines Nvidia's focus for the next twelve months outside the datacentre: "Microsoft and Nvidia will reinvent the PC. This is the first line of completely reengineered and reinvented PCs in 40 years." The product that supports this claim is called RTX Spark, with the N1X SoC at its core. The architecture is Arm v9.2 with 20 cores, 10 for performance and 10 for efficiency, fabricated in 3 nm by TSMC, in a 2.5D package that links the CPU die designed by MediaTek to Nvidia's Blackwell GPU die via NVLink-C2C at 300 GB/s.

The GPU features 6,144 CUDA cores across 48 SMs, the same number as the desktop RTX 5070. The system comes with 128 GB of LPDDR5X in unified memory and delivers around 1,000 TOPS in NVFP4 precision, equivalent to 1 petaflop of AI capacity. The announced thermal envelope ranges between 45 and 80 watts, lower than a typical mobile RTX 5070, opening opportunities for 14- and 16-inch chassis with 14 mm thickness and approximately 1.4 kg, according to the official release. The expected market arrival is in the second half of 2026.

The Choreography with Redmond

Microsoft joined the choreography with its own release on the Windows Blog, also dated 31 May. Satya Nadella summarised the aim in one phrase: "Our goal is to deliver unmeasured intelligence to every home and every desk with Windows." The company will detail the Windows agentic platform with identity governance, containment, and manageability at the operating system level at Build 2026 in June, which changes the risk topology for the corporate security team. The name of Nvidia's orchestration framework for the Windows environment is OpenShell.

The list of confirmed OEMs on stage includes Dell, HP, ASUS, Lenovo, MSI, and Microsoft's Surface team. The manufacturers are preparing over 30 laptop models and around 10 desktops for the second half of the year, according to Reuters. Nvidia described the PC CPU market as "a $200 billion industry," the battleground where Intel, AMD, Qualcomm, and Apple compete for volume.

What Changes in the Corporate IT Ecosystem

The computation for the CIO isn’t trivial. An RTX Spark notebook can run Llama, Mistral, or Phi class models locally, without consuming tokens on a cloud provider. Legal teams in magic circle law firms in London, research areas in Wall Street banks, and compliance teams in Brazil that currently pay for Copilot, ChatGPT Enterprise, and Claude for Work concurrently will now have on-device alternatives in most cases involving sensitive data. The question shifts from "which API do I contract" to "what fraction of the agentic workload runs at the edge."

For the chip makers that support the corporate market, the impact is uneven. Apple, with the M5, maintains the efficiency frontier per watt in the MacBook Pro. Qualcomm, which with Microsoft has pushed the Snapdragon X Elite as the Arm route for Windows over the last eighteen months, sees Nvidia entering with an AI TOPS proposal five times higher. Intel and AMD need to respond at Computex 2026, risk losing the corporate refresh cycle in 2027.

The memory chain is another vector. Nvidia revealed in the same keynote that the Vera Rubin datacentre platform, now in full production, depends on HBM from Samsung Electronics, SK hynix, and Micron. The LPDDR5X of the RTX Spark further increases the demand for high-performance memory and ties Korea to the company’s roadmap until 2028. Jensen Huang is travelling to Seoul on 5 June for meetings with SK, Hyundai Motor, LG, and Naver, according to the Seoul Economic Daily.

The Necessary Interpretation

Nadella's promise of "unmeasured" intelligence on the PC contrasts with the announcement of GitHub Copilot, which began charging per token from this Monday. What Microsoft is signalling is a division of labour. Lightweight, everyday inference migrates to local silicon. Heavier agentic sessions, involving retrieval from corporate corpora, remain in the cloud and consume AI Credits.

The discomfort lies elsewhere. Nvidia has yet to disclose pricing for the RTX Spark notebooks, and the press covering the keynote classifies the segment as premium. For delivery centres in India, the Philippines, and Brazil, the Q3 discussion isn’t about whether AI fits in the developer's chip but whether it fits in the client's fleet refresh CAPEX.