Nvidia and Microsoft stole the spotlight at Computex 2026 last week, unveiling hardware that promises to bring powerful AI processing out of the cloud and onto local Windows machines. The Nvidia RTX Spark GPU and the Surface Laptop Ultra, both officially announced during packed keynotes in Taipei, represent a new class of AI-specialized hardware designed for the next generation of Windows AI PCs.

Computex 2026 ran from June 2 to June 5 at the Taipei Nangang Exhibition Center, drawing over 50,000 attendees. The show floor buzzed with announcements from every major PC maker, but the twin launches from Nvidia and Microsoft signaled a definitive shift toward on-device AI capabilities.

Nvidia RTX Spark: A Dedicated AI Coprocessor for Desktops and Laptops

Nvidia CEO Jensen Huang took the stage on June 3 to introduce the RTX Spark, a discrete AI accelerator that slots into a standard PCIe 4.0 x4 slot or comes soldered onto laptop motherboards. The Spark is not a traditional GPU; it's a dedicated neural processing unit (NPU) built on a custom 4nm process with 96 Tensor cores and 16GB of LPDDR6 memory. Nvidia claims it delivers 200 TOPS of INT8 performance, quadrupling the AI capability of current top-tier laptop NPUs.

The RTX Spark runs Nvidia's CUDA and TensorRT software stacks, but critically, it integrates directly with Windows Copilot Runtime and DirectML. Developers can call Spark-accelerated models through the Windows AI API without rewriting code. During the demo, a prototype workstation running a 20-billion-parameter large language model generated responses at over 60 tokens per second entirely locally, with no internet connection.

"The Spark brings the power of a data center to your desk, and it does so while consuming only 45 watts," Huang said. "This is for the researcher who wants to run Llama 4 locally, the video editor who needs real-time generative fill, and every developer targeting Windows AI PCs."

Gigabyte, Asus, and MSI all showcased pre-built desktops and motherboards with the RTX Spark integrated. Asus announced the ProArt B660-Spark motherboard, which includes the accelerator as a permanently attached chip. MSI's Creator Z16 laptop was the first ultrabook shown with a Spark module, achieving a thickness of just 17mm despite the coprocessor. Nvidia says Spark add-in cards will be available starting July 2026 for $299, with OEM systems shipping in August.

Surface Laptop Ultra: Microsoft's Vision for AI-Centric Hardware

Microsoft countered with its own hardware statement: the Surface Laptop Ultra. Priced at $2,499 and available for pre-order immediately, the Ultra is a 14-inch notebook with a custom Qualcomm Snapdragon X Elite Gen 2 processor that includes an integrated NPU rated at 85 TOPS, supplemented by an optional RTX Spark module in a higher-tier configuration.

What sets the Surface Laptop Ultra apart is its software. Windows 11 2026 Edition, codenamed "Hudson Valley," introduces a new "Copilot Everywhere" feature that leverages the local AI hardware to provide always-available assistance. Using a dedicated haptic touchpad zone, users can summon Copilot contextually without typing or speaking. The laptop's 1080p IR camera supports continuous presence detection, allowing the device to proactively suggest actions based on what you're doing.

Panos Panay, now head of Windows and Devices, demonstrated how Copilot can automatically summarize a spreadsheet while you scroll, translate a video call in real time with on-screen subtitles, and organize files upon recognizing a project folder. All processing stays on-device, Microsoft promised, with no data sent to the cloud unless explicitly requested.

"The Surface Laptop Ultra is the first PC that truly understands you and your work, without compromising your privacy," Panay said. "It's the physical manifestation of what AI can do when it's personal, fast, and secure."

With a 3K 120Hz OLED display, 32GB of LPDDR5x RAM, and a 1TB SSD, the Ultra also offers 20 hours of battery life in typical use. The base model without the RTX Spark costs $1,999, but that version relies on the Qualcomm NPU for AI tasks, providing a slightly reduced Copilot experience.

Intel's Lunar Lake-U and Arrow Lake-HX Push AI Across the Lineup

Intel used Computex to detail its latest mobile processors, which include third-generation NPUs designed for Windows Studio Effects and small models. The Core Ultra 300 "Lunar Lake-U" chips feature a 4-core NPU delivering 40 TOPS, double the previous generation. Meanwhile, the enthusiast "Arrow Lake-HX" series for gaming laptops offers integrated AI engines that can boost game AI and power background tasks like auto-highlight recording.

Dell and Alienware showed new laptops with these processors. The Alienware m18 R7, powered by Arrow Lake-HX and an optional RTX Spark, targets developers who want to train models on the go. Dell's Latitude 16 2026 for enterprises includes a physical shutter for the NPU, a nod to security-conscious businesses.

Asus, MSI, and Others Expand the AI Ecosystem

Asus launched the Zenbook AI series with ultra-thin chassis and a dedicated Copilot key that glows when on-device processing is active. MSI's Creator line added an "AI dial" on the keyboard deck for quick access to generative tools. Gigabyte demonstrated an AI Topology feature in its BIOS, allowing users to allocate Spark resources to specific applications.

BenQ surprised the audience with the DesignVue PD2730A, a 5K monitor with a built-in neural chip that adjusts color accuracy and brightness based on ambient lighting and content. The monitor can communicate with the RTX Spark to optimize the display for AI-generated imagery in real time.

Corsair introduced the first liquid cooler with an embedded AI module that monitors CPU and Spark temperatures to dynamically adjust fan curves and pump speeds, promising near-silent operation during light AI tasks.

Local AI on Windows: A Unified Platform

Underpinning all this hardware is a coordinated software effort. Windows 11 2026 Edition provides a standardized NPU and Spark driver stack, a unified AI compiler, and a growing library of optimized models in the Microsoft Store. Third-party developers like Adobe, Blackmagic Design, and Unity committed to updating their applications to leverage local AI acceleration by the end of 2026.

During the show, Microsoft announced a partnership with Stability AI to bring native Stable Diffusion 4 inference to Windows, accelerated by both NPUs and the RTX Spark. Early tests show image generation in under one second on a Surface Laptop Ultra with Spark, compared to 15 seconds on cloud services.

What It Means for Users and Developers

The shift to local AI addresses long-standing concerns about latency, cost, and privacy. Enterprise customers, in particular, can deploy AI features without worrying about data leaks to public cloud endpoints. Developers gain a clear target: a minimum of 40 TOPS on Intel and Qualcomm platforms, with the Spark offering an upgrade path for heavy workloads.

Prices for AI PCs are falling rapidly. Basic models with sufficient NPU performance start at $799, while Spark-equipped workstations remain a premium offering. IDC analysts at the show predicted that by 2027, 80% of new Windows PCs will ship with AI-capable NPUs, up from 35% in 2025.

Computex 2026 made one thing clear: the age of the AI PC is no longer a distant promise. With Nvidia and Microsoft leading the charge, local AI on Windows is here, finally ready to reshape how we work and create.