Microsoft and Nvidia used a San Francisco Surface event on October 7, 2026 to stake out new territory in personal computing, pairing Nvidia’s RTX Spark hardware platform with a new Windows software layer built for autonomous AI agents. The companies are calling the resulting machines “Builder PCs,” a label PCMag reported the day after the event as shorthand for a Windows PC category purpose-built to run large language models, image generators and background AI agents locally rather than in the cloud. Satya Nadella, Jensen Huang and Windows and Surface chief Pavan Davuluri shared the stage, according to The Verge’s coverage of the announcement.
The framing matters because it is the clearest signal yet that both companies see the next phase of the PC market not as a faster laptop or a bigger screen, but as a machine that runs its own software agents around the clock. Preorders for RTX Spark-based Builder PCs opened the same day, with broad retail availability set for October 16, 2026.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Microsoft and Nvidia Actually Announced
The October 7 event built on groundwork the two companies laid months earlier. Nvidia first showed the RTX Spark hardware family at COMPUTEX 2026 in Taipei on June 1 and 2, describing it as a new class of Windows laptops and compact desktops designed for personal AI agents. On May 31, Microsoft’s Windows Experience Blog and Nvidia’s own newsroom published matching announcements framing the partnership, with Microsoft’s blog post describing it as a new chapter for Windows PCs accelerated by Nvidia silicon.
What changed in October is the software story. Rather than simply restating RTX Spark’s chip specs, Microsoft used the Surface event to introduce Microsoft Execution Containers, or MXC, as generally available infrastructure for running AI agents persistently in the background under operating-system control. That is the piece that turns an RTX Spark machine from “a laptop with a strong GPU” into what Microsoft and Nvidia want people to call a Builder PC: a computer whose operating system treats agents as first-class citizens, with sandboxing, identity attribution and security controls baked into Windows itself rather than bolted on by a third-party app.
Our earlier coverage of the RTX Spark launch detailed the first wave of machines and their pricing. The October event reframes that hardware as the first half of a two-part platform, with MXC as the software half.
Inside the RTX Spark Superchip: Specs That Matter
RTX Spark is built around a Blackwell-generation RTX GPU paired with a 20-core Nvidia Grace CPU built on the Arm architecture, connected through Nvidia’s NVLink-C2C chip-to-chip interconnect rather than a conventional PCIe link. That interconnect is what allows RTX Spark systems to share up to 128GB of unified memory between the CPU and GPU, a design choice aimed squarely at running large local AI models that would otherwise be split awkwardly across separate memory pools.
The GPU itself carries up to 6,144 CUDA cores and fifth-generation Tensor Cores with support for FP4 precision, a lower-precision numerical format that trades some accuracy for a meaningful jump in inference throughput and power efficiency. Nvidia, which announced the 1 petaflop rating in June 2026, and Microsoft put the combined platform’s AI performance at up to 1 petaflop of FP4 compute, a figure the companies use to argue RTX Spark machines can run chatbots, coding assistants, image generators and autonomous agents locally instead of routing every request to a cloud data center.
Nvidia also says RTX Spark laptops are designed to hold full AI and graphics performance while unplugged, with all-day battery life, addressing a long-standing complaint about GPU-heavy laptops throttling hard the moment they leave the charger. That claim has not yet been independently benchmarked by reviewers at the time of writing.
| Component | RTX Spark Specification |
|---|---|
| GPU architecture | Blackwell RTX, up to 6,144 CUDA cores |
| CPU | 20-core Nvidia Grace (Arm-based) |
| Chip interconnect | NVLink-C2C |
| Tensor Cores | 5th-generation, FP4 precision support |
| Unified memory | Up to 128GB shared CPU/GPU pool |
| Claimed AI compute | Up to 1 petaflop |
| Form factors | Laptops and compact desktops |
| Operating system | Windows 11, with Microsoft Execution Containers |
| Preorder date | October 7, 2026 |
| Retail availability | October 16, 2026 |
What “Builder PCs” Actually Means
Builder PC is not a formal product name stamped on a box the way Copilot+ PC is. It is the descriptive term PCMag used to capture what Microsoft and Nvidia were pitching on stage: a PC built for people who build things with AI, whether that means a developer tuning a local model, a studio artist running an image generator without a subscription meter running, or a small team wiring together autonomous agents to handle repetitive work. The Verge’s reporting draws the same distinction, noting that RTX Spark targets AI developers and creators specifically, positioning it as separate from the consumer-focused Copilot+ PC initiative Microsoft has pushed since 2024.
That separation is a notable strategic choice. Instead of folding RTX Spark into the existing Copilot+ branding, Microsoft and Nvidia are drawing a line between PCs that run lightweight, NPU-accelerated consumer features (Copilot+ PCs) and PCs built to host full local models and persistent agents (Builder PCs, powered by RTX Spark). The practical effect is a tiered AI PC market: entry-level machines with modest on-device AI for everyday tasks, and a new higher tier aimed at people who need real local compute.
Why Draw a New Category At All
Part of the answer is pricing and expectations management. RTX Spark machines carry workstation-class components, and Dell’s RTX Spark-equipped XPS 16, detailed in our report on the Dell XPS 16 RTX Spark Creator Edition, launched at a price point well above a typical Copilot+ laptop. Giving these machines their own identity lets Microsoft and its OEM partners justify that premium with a distinct value proposition, local AI horsepower, rather than competing head-to-head with cheaper Copilot+ machines on the same shelf.
Microsoft Execution Containers: The Software Half of the Equation
Microsoft Execution Containers are the operating-system mechanism that lets an AI agent keep running in the background, persistently, without a user babysitting an open browser tab or terminal window. According to reporting on the October 7 event, MXC handles three things Windows has not historically had to manage for third-party software: sandboxing so an agent cannot reach outside its assigned scope, identity attribution so actions taken by an agent are traceable to a specific authorization, and security controls that sit at the OS level rather than inside whatever app launched the agent.
This is a direct response to a problem that has dogged the early wave of AI agents on both desktop and cloud platforms: agents that quietly keep running, consume resources, or take actions a user did not explicitly approve. By making containment a Windows primitive instead of leaving it to individual AI vendors, Microsoft is effectively proposing itself as the trust layer for agentic software on the PC, similar to the role it has long played for driver signing and app sandboxing.
Who’s Building RTX Spark Machines
Nvidia’s OEM list for the first RTX Spark wave includes Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI, with Acer and Gigabyte expected to follow with their own models. That is a broader simultaneous launch roster than Nvidia used for DGX Spark, which shipped mainly through Nvidia’s own channel and a smaller set of workstation partners.
Several of these OEM machines have already surfaced individually in recent weeks. HP’s entry was detailed in a price leak pointing to a $2,999 configuration, while Microsoft’s own Surface Laptop Ultra variant appeared in leaked specs showing the full 128GB RAM, 20-core configuration. Reading those individual leaks against the October 7 framing, it’s clear they were never isolated product stories — they were pieces of a coordinated, multi-OEM launch that Microsoft and Nvidia only tied together publicly with the Builder PC name.
| OEM Partner | Status at Launch |
|---|---|
| Microsoft Surface | First-wave partner, Surface Laptop Ultra variant |
| ASUS | First-wave partner |
| Dell | First-wave partner, XPS 16 Creator Edition |
| HP | First-wave partner, leaked at $2,999 |
| Lenovo | First-wave partner |
| MSI | First-wave partner |
| Acer | Expected to follow |
| Gigabyte | Expected to follow |
RTX Spark vs DGX Spark vs DGX Station: Picking Apart Nvidia’s AI PC Lineup
Nvidia now has three distinct products carrying the Spark or DGX naming that solve overlapping but different problems, and the overlap has caused genuine confusion among buyers trying to figure out which one they need. RTX Spark is the consumer- and creator-facing Windows platform discussed above. DGX Spark, which we covered in detail when the 64GB configuration launched at $4,999, is a deskside developer appliance aimed at people doing heavier local AI development work outside the standard Windows laptop form factor. DGX Station, meanwhile, is Nvidia’s deskside AI supercomputer, which our earlier reporting showed scaling up to 748GB of memory and 20 petaflops of compute for Windows-based AI supercomputing workloads.
The simplest way to read the lineup: RTX Spark is the everyday machine you’d actually buy and carry, DGX Spark is the developer’s deskside box for serious local model work, and DGX Station is the high-end workstation for teams pushing the largest local models Nvidia supports on Windows. All three share the same underlying philosophy that Jensen Huang has been pushing publicly — that AI compute should live closer to the user rather than exclusively in a remote data center.
| Platform | Target User | Form Factor | Positioning |
|---|---|---|---|
| RTX Spark | Creators, developers, gamers | Laptop / compact desktop | Windows PC with agentic AI via Builder PC branding |
| DGX Spark | AI developers | Deskside appliance | Dedicated local AI development system |
| DGX Station | AI researchers, power teams | Deskside workstation | Enterprise AI supercomputing at larger scale |
| Copilot+ PC | Mainstream consumers | Laptop | NPU-based everyday AI features, lower cost |
Where Copilot+ PCs Still Fit
Copilot+ PCs are not going away. They remain Microsoft’s volume play for mainstream laptops with modest on-device AI acceleration through an NPU, covering tasks like live captions, image search and lightweight generative features. RTX Spark and Builder PCs sit a tier above that, aimed at people whose workloads actually need the unified memory and raw compute RTX Spark provides. Microsoft’s own Surface lineup now straddles both tiers, as shown by the Snapdragon-based Surface Pro covered in our report on the Snapdragon X2 Plus launch sitting alongside the RTX Spark-equipped Surface Laptop Ultra.
Historical Context: How We Got to the Agentic PC
The agentic PC pitch did not appear out of nowhere. Microsoft’s Copilot+ PC push in 2024 introduced the idea of an NPU as a standard PC component, not an optional add-on, and trained the market to expect some baseline of on-device AI acceleration in a new laptop. Nvidia, meanwhile, had spent the prior two years building out its DGX line for data-center-style AI development, then progressively miniaturizing it: first into DGX Spark as a deskside appliance, then into DGX Station as a Windows-compatible supercomputer, and now into RTX Spark as something small enough and cheap enough to sell through mainstream OEM channels.
What’s new with the October 7 event is the explicit coupling of that hardware trajectory with an operating-system commitment. Windows has shipped security and sandboxing features before, User Account Control, SmartScreen, Windows Defender Application Control, but none of them were built specifically around the idea that a piece of software might act autonomously on a user’s behalf for hours or days without direct supervision. Microsoft Execution Containers is the first Windows feature explicitly designed around that scenario, which is why Microsoft and Nvidia are treating this announcement as a turning point rather than another hardware refresh.
Expert Voices: What Huang, Nadella and Fisher Are Saying
Jensen Huang framed the moment in personal terms during the event, saying the personal computer is the ultimate tool, it’s my ultimate tool, and for a whole generation of people, it’s our ultimate tool, according to Nvidia’s own blog recap of the Microsoft event. He went further on what agents mean for that tool, asking what happens in the era of agents, when the agent is on your computer, adding that it is now your personal assistant, a line that captures Nvidia’s pitch for why local compute matters more as agents proliferate.
Jeff Fisher, Nvidia’s senior vice president of personal computing, put the partnership in more direct commercial terms in Microsoft’s own announcement, stating that Nvidia and Microsoft share a vision that agents are the future of personal computing, as published on Microsoft’s Windows Experience Blog. Fisher also described RTX Spark as combining Nvidia’s full technology stack with Microsoft Windows, calling it purpose-built for creators, gamers and AI developers in the personal AI era, in the same post.
Satya Nadella’s framing leaned less on hardware and more on access, saying the goal is to deliver unmetered intelligence to every home and every desk with Windows, a line that appeared in Nvidia’s investor relations press release announcing the partnership. That word choice, unmetered, is doing real work: it signals Microsoft wants to position local AI compute as an alternative to the per-token cloud billing that has defined most consumer AI products since ChatGPT’s debut.
Market Impact: Why Both Companies Need This to Work
For Nvidia, RTX Spark extends its AI dominance beyond data centers into consumer and prosumer hardware at a moment when GPU demand for training and inference clusters is already stretched thin. Selling Blackwell-derived silicon into millions of PCs, rather than thousands of data center racks, diversifies Nvidia’s revenue base and gets its architecture directly into developers’ hands, which tends to pay off later in cloud-platform loyalty.
For Microsoft, the stakes are arguably higher. Windows has faced years of questions about its long-term relevance as computing shifted toward phones, browsers and cloud services. Positioning Windows as the only operating system with native, OS-level agent containment, rather than leaving every AI vendor to build its own ad hoc sandboxing, is a bid to make Windows indispensable again, this time as the trust and execution layer for autonomous software rather than just a desktop shell. If MXC becomes the default way agents run safely on a PC, Microsoft gains a platform lock-in point that doesn’t depend on browser market share or Office subscriptions.
The OEM dynamics matter too. A six-to-eight-partner simultaneous launch is unusually coordinated for a brand-new hardware category, suggesting both companies pushed hard to avoid the slow, staggered rollout that characterized early Copilot+ PC availability in 2024.
Competitive Landscape: Apple, Qualcomm and the Edge-AI PC Race
RTX Spark and Builder PCs do not exist in a vacuum. Apple has spent several generations building its own unified-memory architecture across the M-series chips, and its approach to on-device AI through Apple Intelligence remains tightly integrated with macOS rather than exposed as a general-purpose developer platform the way Nvidia and Microsoft are pitching RTX Spark. Qualcomm’s Snapdragon X-series chips, which already power Copilot+ PCs and the Snapdragon-based Surface Pro noted above, compete on power efficiency and battery life rather than raw local-model throughput, leaving a gap at the top end of the market that RTX Spark is explicitly designed to fill.
AMD is the more direct silicon competitor here, with its Ryzen AI Max+ series already appearing in high-memory handhelds and mini PCs aimed at local AI workloads. What Nvidia and Microsoft have that AMD currently lacks is the combination of a dedicated Windows software layer (MXC) built jointly with the OS vendor, plus an eight-OEM simultaneous hardware rollout. AMD’s response, if one comes, would likely need a comparable Microsoft partnership of its own rather than a silicon-only play.
The Preorder Window and What Early Buyers Should Know
Preorders for RTX Spark Builder PCs opened on October 7, 2026, with broad retail availability beginning October 16. That’s an unusually short nine-day gap between preorder and shipping, suggesting Microsoft and Nvidia had inventory largely staged ahead of the announcement rather than taking orders against future production, a sign of confidence in supply that stands in contrast to the GPU shortages that have plagued other recent Nvidia launches.
Buyers evaluating a Builder PC purchase should weigh it against the narrower, workstation-style DGX Spark and DGX Station if their primary need is heavy local model development rather than a daily-driver laptop. For most creators and developers, the appeal of RTX Spark is that it doubles as a normal Windows PC — something neither DGX Spark nor DGX Station is designed to do.
Risks and Open Questions
Several questions remain unresolved even after the October 7 reveal. Microsoft and Nvidia have not published independent, reviewer-verified battery life figures to back the “full performance unplugged” claim, and early hands-on testing will determine whether that holds up under sustained local-model workloads rather than short demo sessions. It’s also unclear how Microsoft Execution Containers will handle agents built by third-party AI vendors rather than Microsoft’s own Copilot ecosystem, which will shape whether MXC becomes a genuinely open standard or a soft lock-in mechanism favoring Microsoft’s own agent products.
There’s also a pricing question the current wave of leaks has only partly answered. Individual OEM price points reported so far, Dell’s XPS 16 RTX Spark Creator Edition, HP’s leaked $2,999 configuration, and the standalone RTX Spark desktop, sit well above typical consumer laptop pricing, and it remains to be seen whether that premium narrows as production scales through 2027.
5 Predictions for the Agentic PC Era
- Expect Acer and Gigabyte to confirm their own RTX Spark machines within weeks, rounding the OEM roster out to eight brands by early 2027.
- Microsoft Execution Containers will likely expand beyond RTX Spark hardware to standard Copilot+ PCs in a scaled-down form, since the sandboxing and identity-attribution logic doesn’t strictly require RTX Spark’s unified memory.
- AMD and Qualcomm will each need to announce a comparable OS-level agent framework with Microsoft within the next one to two product cycles or risk ceding the “serious local AI” segment entirely to Nvidia.
- Pricing on second-generation RTX Spark machines should compress as more OEMs compete on the same Nvidia reference design, mirroring how early Copilot+ PC pricing softened through 2025.
- Expect more scrutiny of MXC’s security model from independent researchers once Builder PCs reach real households, given how much trust Microsoft is asking users to place in background agents running with persistent system access.
Frequently Asked Questions
What is a Builder PC?
Builder PC is the term used to describe Windows machines built around Nvidia’s RTX Spark hardware platform combined with Microsoft’s new Execution Containers software, aimed at developers, creators and anyone running AI agents locally rather than in the cloud.
Is RTX Spark the same as DGX Spark?
No. RTX Spark is a Windows laptop and desktop platform aimed at creators, developers and gamers, while DGX Spark is a separate deskside appliance from Nvidia built specifically for AI development work outside the standard laptop form factor.
When can I buy an RTX Spark Builder PC?
Preorders opened October 7, 2026, with broad retail availability beginning October 16, 2026, across partners including Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI.
What is Microsoft Execution Containers (MXC)?
MXC is Microsoft’s new Windows infrastructure for running AI agents persistently in the background under OS-level control, including sandboxing, identity attribution and security controls.
How much unified memory does RTX Spark support?
Up to 128GB, shared between the CPU and GPU through Nvidia’s NVLink-C2C interconnect.
Is this different from a Copilot+ PC?
Yes. Copilot+ PCs target mainstream consumers with NPU-based everyday AI features, while RTX Spark Builder PCs target creators and developers who need to run larger local AI models and persistent agents.
Which companies are making RTX Spark machines?
Microsoft Surface, ASUS, Dell, HP, Lenovo and MSI are confirmed first-wave partners, with Acer and Gigabyte expected to follow.
Does RTX Spark replace DGX Station?
No. DGX Station remains Nvidia’s larger deskside AI supercomputer for enterprise workloads, scaling to far more memory than RTX Spark for teams running the biggest local models Nvidia supports.
