The potential cloud savings and security proposition of local AI could have appeal, but CIOs warn of lock-in and lack of incremental improvements, instead advising a trial fleet before diving in.
As Windows 10 nears its end of support, some IT leaders, preparing for PC upgrade cycles, are evaluating the possible cloud cost savings and enhanced security of running AI workloads directly on desktop PCs or laptops.
With no security updates for Windows 10 available after Oct. 14, 2025, many CIOs find themselves at a crossroads, with some considering Copilot+ certified AI PCs as part of their enterprise fleet plans for Windows 11.
Some industry experts believe doing so can give CIOs a jump on the inevitable.
“Within two years, it will be virtually impossible to buy a PC, tablet, laptop, or mobile phone without AI,” John-David Lovelock, distinguished vice president analyst at Gartner, recently told CIO.com. “Whether you want it or not, you’re going to get it sold to you.”
Still, the era of AI PCs is in its infancy. The first batch of AI PCs with neural processing units (NPUs) were launched in the second half of 2024, equipped with sub-40 TOPS (trillions of operations per second) chips. These dedicated NPUs — accelerators that run AI workloads more efficiently and save on battery life — are supplied by Intel, AMD, Apple, and Qualcomm, supplementing the core CPUs and GPUs that power PCs and laptops.
Second-generation AI PCs just hitting the market in 2025 will offer a substantial upgrade in performance, providing Silicon-on-Chip (SOC) NPUs that exceed 40 TOPS and 16GB of RAM — requirements for the Copilot+ PC branding designated by Microsoft. OEMs that have shipped or are readying AI PCs for 2025 or 2026 include Dell, Acer, Asus, HP, Lenovo, Samsung, and Microsoft.
Most AI PCs can run LLMs locally but for inferencing only. Some vendors claim their AI PCs can be used to train models.
Dell’s Pro Max, unveiled earlier this month at CES, is a high-end AI PC Workstation that incorporates the 40+ TOPS NPU accelerators and discrete GPU necessary to support AI inferencing and finetuning LLMs, as well as to train them, Dell claims. Dell also launched Dell Pro AI Studio, a toolkit that enables developers to make use of NPUs’ horsepower and the massive memory required for AI models.
End-users can perform basic generative AI capabilities offered by Copilot+ on the cloud and Copilot+ AI PCs on the market today, such as document summarization and content creation. The NPUs offering 40 TOPS being integrated into PCs today will provide both users and developers the ability to run more AI workloads and smaller, specialized AI models.
The enterprise’s local AI alternative
Analysts point out that a PC refresh cycle will begin in earnest later this year when Windows 11 — which offers initial support for NPUs — replaces Windows 10.
“We’ll see a lot of companies refreshing their PCs in 2025 and our expectation is that some percentage of them will go all the way up the stack [AI PC Copilot+], plus, a large percentage of them will go one down and buy those AI PCs with sub-40 TOPS, and of course a sizable percentage in 2025 will still be non-NPU,” says Tom Mainelli, a group vice president at IDC.
“The appeal to CIOs will be the ability to run LLMs locally, driving potential cost, performance, privacy, and security benefits,” he adds.
Mainelli acknowledges that most of the practical use cases on AI PCs will be many of the same things people use AI in the cloud for today — content creation, content editing, text summarization, language translation, automation of repetitive tasks, prototyping, personalization, predictive insights, and virtual assistants — but they will run locally on the device making it faster, cheaper, more private, and more secure.
Allocating some AI workloads to PCs offers CIOs other benefits, he says, noting that Microsoft will continue to make its Copilot+ applications available in the cloud.
“The vision around AI PCs is that, over time, more of the models, starting with small language models, and then quantized large language models … more of those workloads will happen locally, faster, with lower latency, and you won’t need to be connected to the internet and it should be less expensive,” the IDC analyst adds. “You’ll pay a bit more for an AI PC but [the AI workload is] not on the cloud and then arguably there’s more profit and it’s more secure.”
The installed base of such PCs and laptops is small, but more certified AI PCs will be purchased this year and much larger investment is likely in 2026, Mainelli says. According to IDC, nearly 80% of businesses are planning to refresh their PC portfolio by the end of 2025, and 70% of consumers will refresh their PCs in the next two years. IDC also sees an “onslaught” of AI PCs over the long term, as NPUs are integrated into lower-tier hardware.
CIO pros and cons
Despite the looming end date for Windows 10, few CIOs contacted were interested in talking about Copilot+ PCs.
In the meantime, AI PCs and laptops will become more important as agentic AI capabilities start to proliferate, says Nate Melby, vice president and CIO of Dairyland Power Cooperative.
“Larger companies tend to take a lifecycle approach to technology like this, where they would have refreshes on an annual basis. One of my favorite strategies is to refresh one-third of the PCs/laptops each year so that over a three-year cycle, everything stays current,” Melby says. “My approach to this would be to evaluate AI PCs for power users and specific use cases where it makes sense and then start to work them into the next year’s procurement cycle.”
Melby points out there are numerous benefits and claims there is potential for AI PCs to disrupt some SaaS markets.
“It’s smart for CIOs to consider some early deployments of these to bring the AI closer to the employees and processes,” Melby says. “A side benefit is that it keeps the compute local and reduces cyber risk to a degree. But it takes a strategic view and precision targeting. The costs of AI PCs/laptops are at a premium right now, so we really need a compelling business case, and the potential for reduced cloud costs could help break loose those kinds of justifications.”
Not all IT leaders are on board with running AI on PCs and laptops.
“Unfortunately, there are many downsides to this approach, including being locked into the solution, upgrades becoming more difficult, and not being able to benefit from any incremental improvements,” says Tony Marron, managing director of Liberty IT at Liberty Mutual.
Developer ecosystem in its infancy
Aside from collaborating with silicon partners, Microsoft is working with independent software vendors to develop unique AI applications that exploit NPU acceleration.
At CES, Intel trotted out a number of ISVs working on AI capabilities that exploit NPUs and AI PCs.
McAfee, for example, has created a smart AI engine and is working on features for AI PC security, including scam detector technologies to protect the enterprise fleet. Zoom also discussed at CES its new AI companion to enhance its collaboration tool for teams on Copilot+ AI PCs.
As another example, video-editing app Capcut is offering real-time editing, advanced effects, and AI-based filters enhanced by the AI PC’s NPU for smoother performance and quicker rendering times. DaVinci Resolve is another video editing application optimized for NPU capabilities and exploits the computing power of Copilot+ PCs to create efficient workflows for filmmakers, colorists, and editors, the company says.
But these are early days, IDC’s Mainelli says.
“We’re going to have a chicken-and-egg situation where there are few applications that can utilize an NPU,” he says. “All the app developers are working on it; AMD, Intel, and Qualcomm are working on it; and Microsoft is working with software vendors that enable AI functionality at the desktop.”
