How AI Is Reshaping Modern PC DesignsNext-Generation Thin-and-Light Laptops Require More Memory Bandwidth and Better Approaches to Thermal Management

As agentic workloads drive token volumes and cloud costs higher, power users are buying their own local inference hardware, and the PC industry is racing to keep up. New independent research from Moor Insights & Strategy examines why thin-and-light notebooks are colliding with hard physical limits as they try to deliver on-device AI.

Tokens per second is the metric that matters for local inference, and it scales with memory bandwidth. Most laptops today are capped at roughly 150 GB/s by 128-bit LPDDR5 memory buses – nowhere near enough to run frontier-quality models locally. Chip vendors are responding with wider buses and the coming shift to LPDDR6, but wider buses mean memory chips must sit closer to the SoC, in exactly the space traditionally reserved for fans and heat pipes.

Moor Insights & Strategy’s analysis is clear: fan-based cooling will not scale with the denser memory architectures AI-ready notebooks require. Fans will have to shrink (and get louder) or disappear, making solid-state cooling the most credible path forward for OEMs that don’t have the luxury of an on-package unified memory approach.

This white paper provides an independent, system-level analysis of the forces reshaping premium notebook design, including:

• Why memory bandwidth, not raw compute, is now the bottleneck for local AI inference in thin-and-light laptops
• A silicon architecture assessment of Apple, Qualcomm, AMD, Intel, and NVIDIA’s approaches to AI-ready PC design
• The geometric and thermal constraints wider memory buses impose on chassis design
• Why solid-state cooling is emerging as the most credible alternative to traditional fans
• Analyst recommendations for OEMs and silicon vendors racing to lead the on-device AI era