Gartner says AI PCs may find an enterprise role cutting token costs

Gartner is giving corporate buyers a more concrete reason to consider AI PCs: reducing cloud AI token costs by shifting some workloads onto employee devices. In a Strategic Roadmap for Agentic AI PCs published Monday, research vice president Steve Kleynhans said enterprise-grade desktop AI tools have been slow to arrive, but rising usage bills could change procurement math.

The argument is cost displacement, not a sudden breakthrough in employee demand. Kleynhans wrote that on-device AI has not reached mainstream adoption and is still mostly used by developers and enthusiasts. He said enterprises are paying closer attention to the economics of cloud AI, including the uneven way providers define and price tokens.

That is a familiar problem for CIOs now trying to budget AI usage beyond pilots. Cloud model pricing can vary by provider, model, input, output and time of use, making forecasts hard to defend. Gartner’s view is that some routine workloads may be cheaper to run locally as small models improve, although Kleynhans also said there is no agreed measure yet for how large the cost benefit will be.

Small models are the hinge

Gartner’s case rests on advances in small language models, small reasoning models and domain-specific models. Kleynhans said some of these models should run on current AI PCs, which include neural processing units rated at 50 TOPS or more. The firm did not say that most enterprise software is ready for broad deployment on those machines.

Kleynhans pointed to OpenClaw and on-device AI tools including Claude Cowork, Microsoft Scout and OpenAI Codex as examples that could show buyers what is possible. He predicted local models will handle speech, chat, image, audio and text generation, along with application and model orchestration. He also expects personal agents to execute routine tasks locally while coordinating with applications, models and cloud services.

Those are forecasts, not deployed enterprise patterns at scale. The gap between a capable NPU and a managed corporate workflow remains the part PC vendors and software companies still need to close. Gartner’s report effectively shifts the AI PC sales pitch from novelty features to infrastructure economics.

Gartner’s adoption forecast

Kleynhans offered two dated predictions. By 2029, Gartner expects 30 percent of enterprises to use AI PCs to reduce cloud AI token costs. By 2030, the firm expects 70 percent of the corporate PC installed base to be able to run some local generative AI workloads.

Gartner does not argue that endpoints replace cloud AI. Kleynhans said cloud platforms will continue to run the most demanding workloads, while optimized mature models move to PCs where appropriate. He also expects AI PCs to become ten times more powerful by 2031.

For buyers, Gartner’s recommendation is to treat AI PCs as part of IT infrastructure rather than as a premium hardware refresh. Kleynhans advised organizations to build ROI models around token cost displacement, with developers as the initial fit, and to include local execution in discussions about new AI deployments for employees.

He also suggested more serious planning once third-generation AI PCs arrive in 2027, and said companies should begin experimenting with small language and reasoning models. The near-term signal for the industry is straightforward: AI PC demand may depend less on branded copilots and more on whether local inference can make AI spending less volatile.

This story draws on original reporting from The Register.