AI chips have always wanted to be more and more powerful. TSMC has just pointed out the real limit: efficiency

TSMC said chip customers now want more AI performance without a big jump in power use.

Its upcoming A14 node is expected to deliver over 20% more performance while cutting power by up to 30% versus N2.

The story says the industry is also relying on advanced packaging, chip stacking, photonics, and architectural changes to keep improving.

Huawei’s ideas, including Tau Scaling Law and LogicFolding, show how rivals are also chasing efficiency gains.

The broader shift is that power, heat, and cost are becoming as important as raw speed for next-gen chips.