Intel believes AI PCs are the ‘biggest change in personal computing in 20 years.’ The fading chipmaker wants to sell 100 million of the things by 2025 – and it’s not alone. Nearly every PC and chip manufacturer is touting the hardware as the next revolution in computing, with AMD, Dell, HP, Lenovo, Intel, and NVIDIA all announcing AI PC innovations at CES this year.
AI PCs purport to house more capable hardware that supports AI technologies on the device itself instead of the cloud, offering better security, battery life, personalisation and improved performance. But if they are a reformation akin to the smartphone or Wi-Fi, right now it’s one defined more by the marketing literature than sales numbers. According to Gartner data, AI PC shipments will reach 43m units in 2024. If that sounds a lot, that would only be equivalent to 17% of the total number of PCs – some 241.89m – shipped last year, and an impressively large total at that considering that it would mean a jump in sales of almost 100%.
“This has been very much a supply-side push,” says Ranjit Atwal, a research director in Gartner’s Quantitative Innovation team. “It’s the manufacturers pushing this, as opposed to businesses wanting it – which explains the muted response.”
Manufacturers are hoping businesses will pay a premium for AI PCs – the average cost of one is 5-15% higher than traditional models, according to some estimates – but enterprises are averse. The value proposition, explains Atwal, simply isn’t clear when most normal PCs can run AI easily enough in the cloud. “Right now, this hardware is interesting but not interesting enough to find the budget for the technology,” he says. In this sense, says Atwal, AI PCs arrived ‘12 months too early’ when the technology is still in the ‘piloting phase’ among many businesses.
What, then, is the point of an AI PC? Forrester’s Andrew Hewitt is asking the same question. “Most customers today are trying to figure out what the benefits are for their organisation,” says Hewitt. “What types of users and roles are going to benefit most from an AI PC? How is this going to impact overall costs? We’re very much at the beginning stages.”
What are AI PCs useful for?
Exacerbating the uncertainty is that no one can agree on a firm definition for an AI PC. Is it the combination of a Neural Processing Unit (NPU), CPU and GPU, or just the NPU?
The answer depends on who you ask. Kieren Jessop, an analyst at Canalys, prefers an NPU-centric definition (so do Gartner and Forrester.) “From a marketing perspective, the term ‘AI PC’ can be ambiguous, potentially leading to consumer and organisational confusion,” he says.
For their part, Microsoft and Intel say an AI PC includes a CPU, GPU, and an NPU, integrated with shiny new AI-focused software like Microsoft Copilot. Others, like Google, consider any PC capable of performing AI tasks—whether through CPUs, GPUs, or NPUs—as an AI PC, he explains. Interestingly, Apple’s Mac’s have an NPU, labelled a ‘Neural Engine,’ but Apple doesn’t market them as an AI PC.
But this confusion speaks to another problem dragging enthusiasm for AI PCs: what do these features actually allow the user to accomplish? Actually, many things, Intel’s general manager for client AI Robert Hallock tells Tech Monitor, citing the AI-powered features of the latest version of Adobe Photoshop as one example of an AI PC’s transformative potential. More importantly, adds Hallock, AI PCs allow their users avoid something that PCs simply cannot.
“Currently people are running all their AI capabilities in the cloud,” he says. “It’s our intent that if there’s an [AI] model or technique that you like, it will become available on the PC itself. This can eventually save users money on cloud subscriptions.”
That could mean yearly efficiencies in the hundreds of thousands of dollars, according to a recent report from Forrester. Take the gaming startup Latitude: its cloud costs skyrocketed to nearly $200,000 per month when its ‘AI Dungeon’ game went viral (sometimes for the wrong reasons.) An AI PC running a LLM, meanwhile, might only cost ‘pennies’ to run. Perhaps more interestingly, users won’t necessarily even know whether their AI-powered application is running on the PC or in the Cloud.
Improving performance
Make no mistake, says Hallock: it’ll be a long time before AI PCs go mainstream. But it’s an inevitability, he argues. “What I think some pundits get wrong is that this is not a technological speed bump that everybody’s going to collectively forget about,” says Hallock. “We fully believe AI acceleration in 24-36 months becomes an indispensable part of what it means to be a processor or graphics device that is both high performance and high efficiency.”
Or, put another way, he says, Intel expect the software industry to “so broadly transition to AI-based features” that if you don’t have high quality, mature, stable AI acceleration, then ‘“you don’t win performance and you don’t win energy efficiency.”
It’s much more efficient in performance per watt to run the world’s AI workloads on a local device, he says. He points to Intel’s new FastDraft technology which doubles the performance of a LLM and halves the expensive memory bandwidth needed.
Revolution or evolution?
Analysts agree the AI PC is more of an evolution than the revolution manufacturers would have the market believe. As such, Atwal expects their price to fall to a more “natural increase” between 5-10% and sales to soon start to accelerate, with NPUs eventually becoming a standard feature for PC vendors.
Indeed, Gartner is forecasting that while AI PCs will represent 43% of all PC shipments by 2025, demand for AI laptops will be even higher, with shipments expected to account for 51% in 2025. By 2026, AI laptops will be the only choice available to large businesses, the market research firm argues, up from less than 5% in 2023.
Analysts expect this rally will, in part, be driven by Microsoft pulling the plug on Windows 10 by October next year, forcing a stock refresh, as well as a general uptick in PC procurement expected after a recent lull. But, even so, Hewitt says the question most enterprises ask as they weigh up their options is: are AI PCs really ready for prime time? The analyst remains undecided.
“To be frank, today, while you get a better experience…you’re not really fundamentally changing things [for the customer] unless there’s some new capability that you never had before,” Hewitt argues.
But this is expected to change, with new AI PC-specific applications set to emerge. For example, Canalys’ Jessop says leading cybersecurity firms are developing solutions that leverage NPUs for increased threat detection rates. Mobile Device Management solutions are also harnessing NPUs for quicker and more agile deployments.
Is now, then, the right time for businesses to take the leap and opt for an AI PC? Atwal thinks so. “You should be buying it because you need to future-proof,” he says. “There’s no point buying a non-AI PC that won’t be replaced for another five years.”
Whether or not their target market will agree is another question. “At the end of the day,” says Atwal, “this will be determined by enterprises budgets.”
Read more: Tech firms are shunning the London Stock Exchange. That could soon change.
