China expands domestic AI compute capacity and green-infrastructure support
China’s AI compute ecosystem saw new policy measures, chip and foundry investments, local-AI hardware launches, and expanded commercialization of foundation models.
• China’s cyberspace, energy and other authorities issued a 2026-2030 digital-green transition plan calling for higher computing-facility energy efficiency and greater renewable-power use, while separately promoting green design, liquid cooling, waste-heat recovery and low-power chips.
• MIIT’s AI SME Startup Support Plan targets more than 10,000 technology-oriented SMEs and over 2,000 specialized “little giant” companies by 2028, with pooled and elastic access to lower-cost computing resources through the China Computing Power Platform.
• Huahong Grace Semiconductor plans to add a 12-inch specialty-process wafer line in Wuxi with monthly capacity of about 55,000 wafers, backed by a capital increase involving Huahong Semiconductor and other institutions.
• Houmo AI and Guoke Yidao launched the fully domestic FN008 AI laptop, combining a Phytium D3000M processor, Houmo’s 160-TOPS M50 dNPU, KylinOS and support for running 35-billion-parameter models locally.
• Moore Threads said its Lushan graphics chip, based on the company’s Huagang architecture, is expected by the end of 2026 and is designed to add domestic hardware ray tracing alongside projected gains in game rendering and AI performance.
• Nubia and CXMT said the NaviX Ultra AI-agent phone will use domestically produced LPDDR5X memory rated at 10,667Mbps, while the device will include ByteDance’s Doubao Phone Assistant.
• Saudi AI company Humain launched Humain-M3, an Arabic model commissioned from a Chinese AI company and built on MiniMax’s M3 model with 428 billion parameters and more than 1 trillion Arabic-content tokens used for further pre-training.
• Zhipu joined Alibaba’s Tmall token top-up center alongside Kimi and MiniMax, offering model services through a purchase flow compared with phone-credit top-ups.