Microsoft unveiled its most aggressive AI PC lineup yet during the first week of June 2026, spanning both Computex in Taipei and its own Build developer conference in San Francisco. The twin events saw the launch of NVIDIA’s new RTX Spark silicon, a Surface Laptop Ultra that pushes performance boundaries, and a surprise Surface RTX Spark device that puts a dedicated local AI processor directly into a productivity tablet. The message was unambiguous: agentic local AI is now the centerpiece of Windows.

At Computex, NVIDIA CEO Jensen Huang announced the RTX Spark, a scaled-down GPU architecture purpose-built for thin-and-light laptops and tablets. Unlike prior mobile GPUs that leaned heavily on discrete memory and large thermal envelopes, RTX Spark integrates a 48 TOPS neural processing engine alongside 8GB of dedicated high-bandwidth memory, all on a single die that sips just 15W of power. Huang called it “the first genuinely post-Copilot AI engine,” explicitly referencing Microsoft’s earlier attempts with Qualcomm’s Snapdragon X Elite and Intel’s Lunar Lake. The new silicon is not intended to replace a laptop’s main GPU for gaming or rendering; rather, it is a dedicated local inference accelerator that enables always-on agentic behavior without killing battery life.

The immediate beneficiary is Microsoft’s Surface Laptop Ultra, a 15-inch magnesium-alloy clamshell that pairs an Intel Lunar Lake CPU with the RTX Spark accelerator. During his Computex keynote, Microsoft Windows chief Pavan Davuluri demonstrated the Ultra running a local version of “Windows Agent Studio,” a new tool that lets users build custom agent pipelines that run entirely on-device. In one scenario, an agent monitored a user’s email, calendar, and OneDrive files to automatically draft a quarterly report, pull data from a protected SharePoint library, and even generate charts using Excel, all while the laptop was in standby mode. The demo drew applause, but also sparked immediate questions about privacy and resource consumption. Microsoft later confirmed that all agent execution on the Surface Laptop Ultra is sandboxed at the hardware level via the RTX Spark’s secure enclave, and that user data never leaves the device unless explicitly authorized.

The Surface Laptop Ultra goes on sale June 18, starting at $2,199. That price includes 16GB of LPDDR6 RAM, 512GB of NVMe storage, and the RTX Spark as a standard component. A higher-end configuration with 32GB RAM, 1TB SSD, and a 120Hz OLED panel hits $2,799. Microsoft claims up to 16 hours of battery life in “agent-aware mode,” where the RTX Spark dynamically scales power based on pending background agent tasks. Early reviewers have not yet tested those figures, but the hardware giant is clearly betting that users will pay a premium for local AI that does not require a cloud subscription.

Alongside the Ultra, Microsoft unveiled what many had speculated about: a Surface device with an integrated RTX Spark. Called the Surface RTX Spark, it is a detachable 13-inch tablet with an advanced cooling system that allows the 15W accelerator to run at full tilt without throttling. The device includes a new Surface Slim Pen 4 with haptic feedback and an AI-driven context engine that predicts the next stroke based on the user’s writing style. More importantly, the Surface RTX Spark is the first Windows on Arm device with a dedicated AI accelerator that matches x86 AI performance. The tablet runs on a Snapdragon X2 Elite processor, but the NVIDIA silicon handles all agentic workloads. It ships June 25, starting at $1,899.

Davuluri explained the strategic logic: “We’ve learned that developers need a consistent AI target. Today, whether you code for an NPU on a Qualcomm chip, an Intel processor, or an AMD APU, you have to adjust your models and quantization. RTX Spark gives us a single, predictable, powerful accelerator that any Windows developer can target, regardless of the CPU architecture underneath.” That promise of uniformity is key to Microsoft’s goal of making agentic AI a first-class feature in Windows 11’s next major update, code-named “Hudson Valley,” due later this year.

At Build in San Francisco, Microsoft’s developer story came into sharper focus. The company released the Windows AI Toolkit 2.0, which includes a new ONNX runtime optimized for RTX Spark, pre-trained agent models for common tasks, and a local agent emulator that lets developers test multi-agent workflows without deploying to a physical device. Another announcement was “Copilot Local,” a fully on-device variant of Microsoft’s assistant that can handle natural language queries, file management, and even offline coding assistance. Unlike the cloud-based Copilot, Copilot Local requires an RTX Spark or equivalent dedicated AI accelerator. Microsoft said that over 50 third-party laptops from Dell, HP, Lenovo, and ASUS with RTX Spark will ship in the second half of 2026.

Industry analysts greeted the moves with cautious optimism. “Microsoft is trying to solve two problems at once,” said Carolina Milanesi, president of Creative Strategies. “First, they want to justify higher PC prices just as the traditional upgrade cycle slows. Second, they need to differentiate Windows from macOS and ChromeOS in an AI context that Apple and Google are also targeting. The RTX Spark gives them a hardware moat.” However, Milanesi warned that the strategy could fragment the Windows PC market if older devices cannot run the new agentic features. Microsoft insisted that most agent capabilities will gracefully fall back to cloud processing on non-RTX Spark machines, though with higher latency and potential privacy trade-offs.

Privacy advocates were quick to weigh in. The Electronic Frontier Foundation released a statement within hours of the Surface RTX Spark unveiling, praising the on-device processing but cautioning that “hardware sandboxes are only as good as their firmware.” Microsoft later committed to releasing the RTX Spark secure enclave firmware as open source by the end of 2026, a move that surprised many. “We want independent verification that your data stays yours,” Davuluri said on stage at Build, drawing applause from the developer-heavy audience.

The RTX Spark architecture itself is an interesting departure from NVIDIA’s usual playbook. Instead of using the same Tensor cores found in its GeForce RTX cards, the RTX Spark employs a new type of integer-optimized core that excels at transformer model inference. At 15W, it delivers roughly the same AI throughput as a last-generation RTX 4060 running at 45W. The silicon also supports a technique NVIDIA calls “Agent Threading,” which allows multiple agent processes to run concurrently with minimal context-switching overhead. Huang claimed that an RTX Spark laptop could run 15 to 20 simultaneous agent tasks—email triage, meeting transcription, code refactoring, document comparison, image generation—without any noticeable system slowdown.

Microsoft’s own agentic platform, Windows Agent Studio, is built on top of a new local orchestrator that manages these parallel workflows. Developers can define agents using a YAML-based script that specifies triggers, tools, and output actions. The studio includes a visual debugger that shows how agents pass messages to one another. At Build, a third-party developer demonstrated an agent that monitors a GitHub repository, runs a test suite on commit, and deploys to Azure while logging all steps locally—all triggered within seconds of a code push. The entire workflow ran on a Surface Laptop Ultra, with the RTX Spark handling the model inference for code analysis and the Intel CPU managing compilation and deployment scripts.

Battery life remains a critical concern. NVIDIA claims the RTX Spark’s power envelope is so small that even under sustained agent workload, it adds less than 2W to system draw compared to idle. Microsoft’s own telemetry, gathered from insider builds of Hudson Valley, suggests that an RTX Spark laptop will lose only about 5% battery per hour when running a high-density agent workload. That would translate to roughly 20 hours of continuous local AI processing on the Surface Laptop Ultra’s 68Wh battery, though real-world usage will vary.

The Surface Laptop Ultra and Surface RTX Spark also introduce new physical design cues that signal their AI readiness. Both feature an “AI Active” LED strip on the keyboard deck that glows a soft blue when local agent processing is active. Users can disable it, but Microsoft says it was added as a privacy indicator analogous to the camera light. The Surface RTX Spark tablet, when attached to its keyboard cover, reveals a dedicated “Agent Button” that summons Copilot Local and shows a task manager for all running agents.

Pricing and availability extend beyond Microsoft’s own hardware. Dell announced an XPS 15 RTX Spark Edition, Lenovo previewed a ThinkPad X1 Carbon with the accelerator, and ASUS showed a ProArt Studiobook with a 16-inch 4K OLED and RTX Spark as standard. All these machines carry a “Windows AI PC” badge that Microsoft says guarantees certain baseline capabilities: support for Windows Agent Studio, Copilot Local, and a minimum of 45 AI TOPS across CPU, GPU, and RTX Spark combined. That badge will be critical for consumers trying to navigate the increasingly complex AI PC landscape.

Looking ahead, the Windows AI PC push seems to be coalescing around three pillars: hardware acceleration from NVIDIA’s RTX Spark, a unified developer platform via the AI Toolkit and Agent Studio, and a tiered experience that prioritizes local execution but doesn’t abandon cloud fallback. Whether that’s enough to move the needle on PC sales remains to be seen. But by marrying a dedicated NVIDIA silicon with Surface hardware and a revamped Windows AI stack, Microsoft has drawn a clear line in the sand: the AI PC is not a nebulous concept; it is a device with specific, measurable capabilities that now have a price and a date.