HN Works - stock.adobe.com
Enterprises warming to AI PCs amid growing cloud costs
While global memory shortages will pose a threat to the broader PC market, AI PCs are gaining momentum across Asia as companies look to cut cloud costs, boost productivity and secure sensitive information
As enterprises double down on the use of artificial intelligence (AI), more are warming up to AI-enabled personal computers to cut cloud computing costs, improve productivity and protect sensitive corporate data.
That was according to Ketan Patel, president of HP’s personal systems business, who revealed that AI PCs made up 35% of the company’s total PC shipments in its most recent fiscal first quarter, up from 30% in the previous quarter.
In a recent interview with Computer Weekly in Singapore, Patel noted that procurement of AI PCs in the first half of 2025 was mostly about future-proofing, as enterprise buyers fear missing out on new technologies over a standard three-to-four-year hardware refresh cycle.
But in the second half of the year, adoption accelerated as businesses began to see more returns on investment (ROI) from AI PCs, driven largely by the growing cost of using cloud-based AI services, said Patel.
“When customers are accessing AI through the cloud, the number of tokens they are consuming and the associated costs have become a discussion point,” he explained. “If you’re able to deliver that kind of experience on a device, then it’s a strong ROI.”
Another driver of AI PC adoption is data privacy. With the advent of highly efficient, small language models, enterprise users can analyse local documents and data on AI PCs while maintaining data privacy and meeting other compliance requirements.
Patel added that latency-sensitive applications, such as live translation and retail environments using edge-based ambient AI, are also pushing inferencing away from the cloud onto AI PCs equipped with neural processing units (NPUs).
Digital assistants
The biggest driver of demand for AI PCs may well be the growing use of digital assistants that help users improve productivity in day-to-day tasks. Microsoft’s research found that the most efficient users of Copilot saved 10 hours a month, while the figure for the average person was nearly five hours.
More recently, some users of the Claude Cowork agentic AI tool that works with local files and apps to help individuals achieve specific tasks have reported productivity savings of at least 10 hours per week. Anthropic, the company behind Claude, has recently released plugins for Cowork that can be used to perform specific job functions like sales, legal and financial analysis.
Further taking advantage of AI PCs are use cases related to endpoint security, where suppliers such as ESET are using NPUs to run some AI models for threat detection, improving scan performance and power efficiency.
Asia catching up
Patel noted that the adoption of AI PCs is ramping up in Asia, which initially lagged behind North America in adopting the prerequisite Windows 11 operating system.
“Windows 11 refresh was faster in the US last year, and now Asia is catching up big time,” he said, noting that growth in the Asian market over the past two quarters has been “pretty strong”.
The region’s dynamic software industry helps, too. “There are many companies in the software ecosystem that want to use the latest technology,” said Patel, adding that HP’s Garage 2.0 programme is actively inviting startups and independent software vendors (ISVs) in the region to co-develop local and hybrid AI applications.
His optimism aligns with broader industry projections. Research firm Gartner predicted that AI PCs would account for 31% of the global PC market by the end of 2025, before hitting 55% – or 143 million units – in 2026. By 2029, Gartner expects AI PCs to become the industry norm.
Despite growing adoption, the AI PC market faces macroeconomic and supply chain headwinds. Analyst IDC recently warned that the deepening global memory shortage crisis could cause the overall PC market to contract by around 5-9% in 2026.
Read more about end-user computing
- End-user computing has shifted from desktop support to a financial control plane, exposing SaaS sprawl, labor inefficiency and cost drift at the endpoint.
- AI, identity-first security, Windows 10 end of support and VMware's rebranding as Omnissa are reshaping EUC as CIOs make critical decisions for 2026 and beyond.
- As Windows 10 support ends, organisations must plan Windows 11 migrations carefully. Assess apps, data and device configurations to avoid disruptions during the upgrade.
- PC manufacturers know the migration to Windows 11 is slowing down and that the focus must shift to enticing users with features and AI capabilities.
With memory manufacturers pivoting production towards high-bandwidth memory to feed the demand of AI datacentres, PC suppliers are facing tight supply constraints for standard RAM components. Because AI PCs require more memory to run local models efficiently, these shortages could lead to price hikes and suppress overall unit growth next year.
To maintain its sales momentum with AI PCs, HP is leaning into security by doubling down on investments in its Wolf Security stack to differentiate its hardware and help customers fend off increasingly sophisticated cyber threats and AI-enabled attacks.
With AI PCs, buying conversations are also expanding beyond traditional IT procurement teams to line-of-business leaders, such as customer support directors, who are demanding AI-first workflows.
“Our sales approach is changing to a more business-leader-centric value proposition and communication, while still ensuring we go through the regular IT and procurement process,” said Patel.
Meanwhile, the growing influence of business decision-makers could alter procurement models. While HP’s device as a service (DaaS) offerings do not yet feature outcome-based pricing specifically tied to AI capabilities, he noted that such pricing models are being considered.
“We definitely see a path where our DaaS will be able to deliver a certain outcome in specific areas, with a subscription model tied to it,” said Patel. “It’s premature at this stage because we are developing the concept, but once we are ready, we’ll share more.”
