China's AI infrastructure build-out is rapidly lifting demand for domestic accelerators, giving Huawei, Alibaba's T-Head Semiconductor, Cambricon and Hygon a larger role in a market once dominated by Nvidia.
Participation in the Global Pavilion at the upcoming SEMICON Taiwan 2026 has reached a new high, with the number of participating countries climbing to 18 and the booth count surging from 62 in 2023 to 220 in 2026, an increase of more than 250%. Both metrics set new records, reflecting that amid rising artificial intelligence (AI) investment and the regionalization of supply chains, countries are moving faster to deepen cooperation with Taiwan's semiconductor industry.
China's research and academic community appears to be drifting away from international collaboration, even as the UK, Germany, and the US have steadily expanded cross-border partnerships in recent years. The contrast has drawn fresh attention after Moonshot AI's flagship 2026 model Kimi K3 and other Chinese technology products, including Unitree Robotics's H1 robot, prompted questions over how much of China's advanced innovation is truly homegrown.
Lightelligence is positioning optical interconnects, optical switching, and optical computing as the next foundation of AI infrastructure, arguing that the industry's biggest challenge has shifted from faster chips to moving data efficiently across increasingly larger GPU clusters.
The AI industry may be riding a wave of momentum, but two clouds have recently unsettled the outlook: growing public opposition to data center construction across the United States, and a dramatic leap in the rankings by Chinese-developed model Kimi K3, which has renewed fears of AI overinvestment. NVIDIA CEO Jensen Huang addressed both concerns directly on July 21.
Nvidia's newly detailed Vera Rubin NVL72 is being pitched less as a faster chip and more as a cheaper unit of AI output. This shift matters most for the power-constrained data centers now defining the ceiling of the industry's growth. In its July 21 announcement, Nvidia said the rack-scale system delivers roughly 10 times more tokens per megawatt than its current Grace Blackwell NVL72, and about one-tenth the cost per million tokens of the GB200 NVL72. This framing positions electricity, not silicon, as the scarce resource.
Intel and Fortinet have expanded their long-standing partnership to develop the Fortinet Security Processor 6 (SP6), a move expected to shape how cybersecurity hardware is designed, produced, and distributed globally. The effort is also intended to make the supply chain for critical security chips more resilient, potentially benefiting enterprises, governments, and consumers worldwide.
China is considering tighter export controls on artificial intelligence and semiconductor technologies amid intensifying competition with the US in frontier AI. The proposed measures, reported by Financial Times, would also target overseas acquisitions of advanced technology startups and could appear in the next revision of Beijing's export control catalogue.
Google is reportedly developing a new chip that embeds information from its Gemini AI models directly onto the silicon to boost efficiency. If the anticipated performance gains materialize, this new design would mark a deeper integration of the tech giant's hardware and AI capabilities.
Chinese AI developer Z.ai has completed a giant data center running exclusively on domestically made chips, a milestone in Beijing's drive to replace restricted Nvidia silicon for future AI development.
