本地 AI 模型已能媲美云端前沿模型，数据中心将走向个人化

But for much of everyday knowledge work, there is no reason to send the query to the data center at all. The combination of local models plus a router is sufficient for the supermajority of work.⁷

Cloud remains essential for long multi-step reasoning, the hardest technical domains, & workloads where scale & parallelization matter. Cloud hardware still holds an edge over local hardware in many use cases. Cloud inference delivers a 40% energy efficiency gain relative to local models.⁶ Datacenters batch queries, a trick local hardware serving one user at a time cannot use, yet.

It also cuts energy 80%, compute 77%, & cost 74% against an all-cloud baseline.

Tracking computing efficiency over time is not new. Koomey’s law found that computing power per watt doubled roughly every 1.5 years for decades, a trend that shrank the power of a mainframe into a laptop’s chassis.²³

The impact is no less impressive : the best local model’s win/tie rate against a frontier model, rose from 23.2% in 2023 to 71.3% in 2025, adding roughly 20 percentage points a year. In 2026, a local model selected for the task by another local AI bumps that number to nearly 90%.⁵

Efficiency improved alongside the trend line : intelligence-per-watt rose 5.3x over the same period, split into a 3.1x gain from better models & a 1.7x gain from better chips. Computers are faster, models are smarter. And the end user benefits.

That means we’re generating more intelligence per watt of electricity : more work from the same number of electrons.¹

GPUs follow a more languid curve today, doubling efficiency roughly every 2.7 years over the last 15 years, not every 1.5 years.⁴

Just as performance-per-watt guided the mainframe-to-PC transition, intelligence-per-watt will guide AI’s transition to the edge.

— Jon Saad-Falcon, Avanika Narayan, et al., “Intelligence per Watt”

5

1. Jon Saad-Falcon, Avanika Narayan, et al., “Intelligence per Watt: Measuring Intelligence Efficiency of Local AI,” Stanford University & Together AI, November 2025. arXiv:2511.07885. See also the Stanford Hazy Research overview. ↩︎ ↩︎

2. Koomey’s law, Wikipedia. ↩︎

3. Watts measure power, the rate energy is drawn, & joules measure the energy itself; Koomey’s original metric was computations per joule, but the underlying trend is the same one intelligence per watt now tracks for AI models. ↩︎

4. Anson Ho, Ege Erdil & Tamay Besiroglu, “Limits to the Energy Efficiency of CMOS Microprocessors,” 2023. arXiv:2312.08595. ↩︎

5. 71.3% & 89% are two different measurements from the same 2025 data, not the same number at different times. 71.3% is the best single local model’s win/tie rate against a frontier model, the trend line this chart plots (23.2% in 2023, 48.7% in 2024, 71.3% in 2025). 89% is a separate, higher ceiling: routing each query to whichever of the 20+ local models tested handles it best beats any single model by 16.3 to 28.8 percentage points. That gain is a selection effect: the paper notes local routing draws from 20+ diverse models versus three frontier cloud models, so on some benchmarks the best-of-local ensemble even surpasses best-of-cloud. More candidates to choose from, not just smarter routing, is what raises accuracy. Both figures come from the same study & neither supersedes the other. ↩︎ ↩︎

6. Jon Saad-Falcon, Avanika Narayan, et al., “Intelligence per Watt: Measuring Intelligence Efficiency of Local AI,” Stanford University & Together AI, November 2025. Cloud accelerators deliver at least 1.4x higher intelligence-per-watt than local chips running the same models, roughly a 40% efficiency premium. arXiv:2511.07885. ↩︎

7. Most AI Work Can Wait, tomtunguz.com. ↩︎

Local AI models can already answer 89% of everyday chat & reasoning queries as well as a frontier cloud model, a result that holds broadly across more than a million real queries & 20+ local models tested.¹

Mainframes became personal. So will your data center.