Several manufacturers, including HP, Intel, MSI, and Lenovo, are now offering PCs with dedicated AI hardware, touting them as the next big thing at conferences like CES—with Intel’s new line of Meteor Lake processors packaging a dedicated AI component called a neural processing unit (NPU). Promised features include everything from enhanced threat detection and productivity to AI integrations with apps like Microsoft Copilot.
Yet executives hesitant to dip their toes into the AI PC market may have good reason. Buying NPU-boosted enterprise hardware so soon could backfire, experts told IT Brew.
No killer, AI filler. A major concern for many organizations? Runaway costs—an argument in favor of AI PCs, which could help slash increasing cloud bills by running on dedicated, local hardware. Running large language models (LLMs) on a PC also eliminates the need to transmit potentially sensitive data over the internet or to service providers.
According to Forrester Research’s Andrew Hewitt, a principal analyst focused on infrastructure and operations, AI PCs remain a solution in search of a purpose. Hewitt recently issued a report advising enterprises should hold off until at least next year to start stocking AI PCs, arguing they don’t have any killer apps in the business segment yet.
“Much of the story around the benefit from AI PCs is focused on end user experience,” he told IT Brew. “And so what’s missing from that conversation is what benefit is there for IT.”
NPUs currently mostly support user experience enhancements like automatic framing or picture enhancement on video calls, Hewitt said. In the meantime, he sees IT teams more concerned about issues like the upcoming Windows 10 end-of-life deadline, cost management, and security.
IT teams also won’t pay attention until game-changing capabilities like running LLMs locally start showing up, according to Hewitt.
“When I think about IT purchasing, and specifically around devices and procurement teams that I work with today, they’re usually a few years behind the vendor announcements,” Hewitt told IT Brew. “So, what that means is they typically tend to prioritize other things beyond user experience.”
Top insights for IT pros
From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.
No common standards. Another factor which might turn off enterprise buyers is the lack of a common framework for NPUs. That means compatibility between devices and services is not guaranteed, unlike graphics processing units (GPUs), which operate off shared DirectX application programming interfaces.
“I don’t believe that all NPUs are created equally,” Michael Cherry, the Windows analyst at IT information and advisory service Directions on Microsoft, told IT Brew.
He said that while most PCs with dedicated AI hardware use Intel NPUs, competitor AMD has its own NPU-like offerings, Microsoft is developing its own processors, and unexpected wild card entrants could include companies like ARM. It’s not clear if any given manufacturer’s NPUs will be able to run every AI model.
“We don’t know if those [Microsoft] processors are going to be limited to just what they use in their Azure workloads, or whether they’re going to make it into desktop-type devices,” Cherry said. “It is a really bad time to buy a PC.”
Hewitt said that while he expects “pockets” of unsupported capabilities, the bigger issue is that many customers may not want to pay what he estimated to be a 15% premium for an NPU to run models capable of running perfectly well on decent CPUs and GPUs.
“Likely, you’re not gonna see a massive benefit in terms of the overall user experience,” he said. “So, until that gap is increased and I’m getting much, much more performance out of the AI chipset, it’s not gonna be convincing for IT.”