A Microsoft patent application for training AI models on spare GPU capacity in devices such as gaming computers has 20 claims, and by our count none of them mentions paying, crediting or compensating the people who own those devices. The application, US 2026/0349820 A1, was published by the U.S. Patent and Trademark Office on 5 November 2026; the applicant is Microsoft Technology Licensing, LLC.
Payment appears in one place: paragraph [0019] of the specification, one of 188 numbered paragraphs, which says the system “may compensate the owners of the edge devices”. The application (No. 19/195,047) was filed on 30 April 2025, 554 days before it was published, by our arithmetic. A published application is not a granted patent and not a product.
Business Pill · GPUs
A short explainer of why AI runs on graphics processors, the chips first built for video games. It teaches the general idea only and does not discuss Microsoft’s filing.
The key insight: By our count, none of the 20 claims in Microsoft’s application US 2026/0349820 A1 mentions paying, crediting or compensating device owners; payment appears only in paragraph [0019] of the description, as something the system “may” do “in some examples”.
The filing names one inventor, Aditya Vasanth Challapally of San Francisco, and is titled “Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization”. Its abstract describes “a system and method for leveraging spare GPU capacity on edge devices through opportunistic micro-training of machine learning models”.
The specification starts from supply. It says the demand for GPUs “has led to shortages”, and that “gaming computers often feature powerful GPUs that are largely unused when not playing games”. It defines edge devices as network-connected devices with GPUs, including smartphones, laptops and Internet of Things devices, which are “typically not owned or controlled by the entity seeking to utilize their computational capabilities”.
What each of the 20 claims in Microsoft’s application US 2026/0349820 A1 covers, grouped by the main feature each claim adds, by our count. Payment to device owners appears in no claim, only in description paragraph [0019]. Source: USPTO.
What the 20 Claims Cover
The claims are what a granted patent would protect. Claims 1, 8 and 15 are the independent claims, written as a method, a computing device and a machine-readable medium. Each covers splitting a model’s training into “independently trainable micro-tasks” using a dependency graph, scoring candidate split points on communicationcost, memory overhead and data locality, sending a micro-task to a device, and weighting the returned result by a validation score, a staleness factor or a device reliability score.
By our count, the 17 dependent claims group into four features. Nine (claims 2-4, 9-11 and 16-18) cover monitoring devices and predicting when they are free, using GPU utilization, memory usage, device temperature and power consumption. Three cover how the dependency graph is built, three cover sizing a micro-task to each device, and two cover a reliability score based on an exponentially weighted moving average.
Where the Payment Idea Sits
The compensation language is in the description, in full: “In some examples, the system may track the amount of computations each edge device has provided and may compensate the owners of the edge devices. For example, each computation may result in a credit being applied to an owner’s account. The credits may be exchanged for free software, model access credits allow the user to access the model being trained, money, or the like.” The same paragraph says the devices may be third-party devices whose owners “have given their express authorization”.
As we read the filing, payment is described as one possible example rather than claimed as part of the invention. The filing does not set a rate, a payout method or a programme.
As we read the claims, what Microsoft seeks to protect is the engineering of using unreliable spare capacity: how to cut training into pieces, when a device is free, and how much to trust what comes back. Nine of the 17 dependent claims, by our count, are about monitoring devices and predicting availability.
As we read the timeline, the application was filed on 30 April 2025 and published on 5 November 2026, 554 days later by our arithmetic; publication makes the filing public, and it does not mean the claims have been examined or allowed.
US 2026/0349820 A1, paragraph [0013]
“gaming computers often feature powerful GPUs that are largely unused when not playing games”
Three Implications
FOR GPU OWNERS The filing describes payment as credits that may be exchanged for free software, model access credits or money, in some examples; no claim covers it, by our count.
FOR AI COMPUTE SUPPLY The specification says GPU demand has led to shortages and points to idle GPUs in gaming computers as unused capacity.
FOR THE PATENT ITSELF The application, No. 19/195,047, was published on 5 November 2026 and has not been granted.
The framework says: “The structural bottleneck isn’t computing, it’s electrons.”
As we read it, Microsoft’s filing frames the shortage as one of GPUs, while the framework places the limit in power. Spreading training across household devices would spread the power draw too; the claims monitor power consumption, and the filing does not discuss electricity costs. The framework was published in December 2025 and does not discuss this filing.
What Is Not Established
The application has not been granted, and claims often change during examination. The filing does not say whether Microsoft will build or launch such a system, how much owners would be paid, or how electricity costs would be handled. On payment, the filing goes no further than saying the system “may compensate” owners, “in some examples”.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Microsoft’s idle-GPU application, published 554 days after it was filed, claims a method for splitting AI training across spare consumer GPUs. None of its 20 claims mentions paying device owners, by our count; that idea sits in one paragraph of the description.
A note on sourcing. All facts are from the USPTO’s published application US 2026/0349820 A1 (Appl. 19/195,047), read in full on 10 November 2026: its front page, abstract, specification and 20 claims. The 554-day gap and the grouping of claims are our arithmetic and our count. Microsoft was not contacted for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
