Physix Frontier · News Briefing Card (enbrief · Sep 28, 2026)

Compute-Power Synergy Needs End-to-End Optimization; No Short-Term Edge for US or China

KEY FACTS

In Q2 2026, China's average daily token calls reached 500 trillion, 16 times the same period last year.
The National Energy Administration projects that computing power electricity consumption will reach 800 billion kWh by 2030, accounting for about 6% of total social electricity consumption.
The national plan targets AI computing power of 9,800 EFLOPS by 2030, more than a 5-fold increase over 2025.
NVIDIA's VR200 GPU in 2026 consumes about 2,200W, and that could double in the future.
Google's core AI response cost fell by more than 30% in Q1 2026.

KEY DATA

500 trillionChina daily average token calls

800 billion kWh2030 computing power electricity consumption

9,800 EFLOPS2030 AI computing power target

2,200WNVIDIA VR200 power consumption

PHYSIX OBSERVATION

Compute-power synergy is shifting from an enterprise technology upgrade to a national systems engineering effort, and electricity has become the physical precondition for unleashing computing power. China and the US have different resource endowments, but the focus of competition has shifted to value per watt. Grid expansion cycles are far slower than chip iteration, and this time mismatch is the biggest risk; whoever can coordinate faster will seize the early advantage in the agent economy.

Source: enbrief original report ↗