NVIDIA's Jensen Huang tells US media China should not be feared; Beijing rolls out a 'Token economy' AI inference (running already trained models) compute subsidy billed by token, by autonomous AI agent, and by completed task; China's market
NVIDIA CEO Jensen Huang said China's open-source AI models are 'excellent' and the US 'should not fear' them. The endorsement, carried in a CCTV Finance secondary chain without the original US outlet named, arrived July 22 — followed the next day by a Beijing 'Token economy' policy and SAMR's Q2 'involution' review: three moves across two days, in China's AI competition posture.
Beijing's Development and Reform Commission issued 京发改〔2026〕1185号 on July 23, the city's first formal 'Token economy' (词元经济) package. It subsidizes inference compute — chips that run already-trained models, not chips used to train them — and backs three new commercial models: TaaS (Token-as-a-Service), AaaS (Agent-as-a-Service), and RaaS (Result-as-a-Service). The pivot moves Beijing's AI support from training-side to inference-side subsidies.
Same day, SAMR (State Administration for Market Regulation) held its Q2 2026 review and called for deeper action against 'involution' (内卷式) — destructive cutthroat competition — plus continued rider-party-building for delivery drivers. The 'involution' line extends a 2025-2026 policy frame; this deployment reads as enforcement under existing rules, not a fresh turn.
Also on the wire: Meta priced $12 billion in bonds for a roughly 1-gigawatt Texas data center at talks above 7%; UniCredit raised its Commerzbank stake to 48% with a Q4 close targeted; Mobileye's Amnon Shashua will step down as CEO once a successor is named; spot gold fell 1.96% to $4,049/oz; a Chinese team ran the first cross-region, 1,000-person simultaneous brain-signal collection; ByteDance registered a Zhongwei subsidiary with 2.4 billion yuan (~$335M at mid-July 2026 rates) in registered capital.