Could now finally be a good time to buy an AI PC? This report says so

• AI PCs are emerging as a viable option to run local AI without unpredictable costs
• One-time PC cost alleviates the need to fork out for cloud token fees
• Broader research reaffirms rising popularity of smaller models

New Gartner data has claimed now could actually be a good time to buy AI PCs, as cloud computing faces numerous challenges in a rapidly-changing business world.

Data center construction is slipping behind demand as supple chains strain and local communities oppose new projects, meaning that metered compute could end up costing some companies more than they'd bargained for.

By shifting some of their AI processing locally, companies could be able to avoid some of those extra monthly costs with a one-time purchase of a more powerful PC as part of their regular refresh cycles.

AI PCs present an ideal hybrid compute model

While AI PC adoption started pretty slow with companies struggling to understand the benefits, they're now being seen as a cloud fallback option rather than a primary benefit in their own right.

With AI usage increasingly sharply and unpredictable token consumption hitting companies hard, forecasting monthly costs is a new major challenge that many face.

Small language and reasoning models, including specially trained models for individual business use cases, ultimately need fewer resources than leading frontier models, allowing them to be run locally as part of a broader hybrid approach.

Gartner predicts that speech and chat, text generation, image and audio generation and more could soon shift to workers' PCs, with only the most intensive tasks routed via hyperscaler data centers.

By as soon as 2029, the company's researchers anticipate that around one-third (30%) of enterprises could use AI PCs to reduce cloud AI token costs. By 2030, 70% of corporate PCs could be able to run some GenAI tasks locally.

Omdia researchers also noticed a shift in AI model usage, with smaller and medium models proving popular, with domain-specific tasks not needing the full breadth of compute.

"Older GPUs are retaining value and remaining in service, as they continue to offer a cost-effective option for small and midsized model inference and disaggregation," Senior Principal Analyst for Advanced Computing Alexander Harrowell said.

Via The Register