Electronics profits jump 96.9% in first half; ChangXin Technology hits 3.28 trillion yuan on debut
China’s electronics industry profits soared 96.9% year-on-year in the first half of 2026, boosting overall industrial profit growth by 8.5 percentage points, with integrated circuit manufacturing up 2,579.5%. Memory chip maker ChangXin Technology listed on the STAR Market, ending its first day 466% higher with a 3.28 trillion yuan market value, surpassing ICBC as the top-valued A-share stock. A conference showcased a 100,000-card AI computing cluster, and executives discussed volatile server pricing, evolving CPU-to-GPU ratios, and the trend toward cloud-based AI delivered via mobile apps.
Electronics industry profits among industrial enterprises above designated size surged 96.9% year-on-year in the first half of 2026, contributing 8.5 percentage points to the total profit growth of those enterprises. Within the sector, integrated circuit manufacturing profits skyrocketed 2,579.5%. Computer whole-machine manufacturing profit rose 689.3%, computer peripheral equipment manufacturing profit increased 305.8%, semiconductor discrete device manufacturing profit grew 31.2%, electronic special materials manufacturing profit climbed 209.7%, and electronic circuit manufacturing profit was up 26.9%.
On July 27, ChangXin Technology, a leading memory chip maker, made its debut on the STAR Market. The shares opened 471% higher and closed with a 465.82% gain, giving the company a market capitalisation of 3.28 trillion yuan. It surpassed Industrial and Commercial Bank of China to become the most valuable company in the A-share market.
At the ‘Photosynthesis Organization 2026 Intelligent Computing Application Conference’ in Zhengzhou, an exhibitor showcased what was described as China’s first AI computing cluster equipped with 100,000 cards. The venue was packed, with server component and system manufacturers actively engaging with AI application developers. A person from a GPU firm noted that server prices fluctuated so sharply that contract price validity was shortened to a week, requiring re-pricing thereafter.
Liao Guangshan from Inspur Computer’s AI product department explained that when ordinary users run large models, cloud-based inference draws on cloud computing resources. For local deployment to tackle privacy concerns, the main cost is purchasing servers because the models are open-source and free to download. Resource demands differ markedly by task: running DeepSeek needs a baseline level, fine-tuning requires four to eight times that, and inference training needs over 20 times. If the goal is merely training stability — for instance, maintaining character consistency in AI-generated short dramas — companies can buy tenant isolation services from large model providers without any local deployment.
Li Bin, senior vice-president at Sugon, said computing power is moving closer to everyday life. The Sugon 8000 (Dengfeng) system in Zhengzhou, which formerly served mainly research institutes, now dedicates a large share of its resources to the back-end of mobile apps, offering token services. Zhang Panyong, general manager of Haiguang Information’s server product division, noted that with the arrival of the AI agent era, CPUs are taking on a larger share of workloads including task orchestration, tool calling and memory management, while GPUs remain focused on inference. In data centres, the CPU-to-GPU ratio has shifted from 1:4 or 1:8 to about 1:1.
On the outlook for computing, Li Bin said the broader direction is cloudification, with edge devices carrying lightweight computing for real-time response, so individuals will not need local equipment but can access AI capabilities through apps. Ying Zhiwei, vice-president at Haiguang Information, remarked that computing will become as accessible as utilities, with all businesses deeply integrated with AI. Future chip design will not be limited to large AI processors in data centres; phones, wristbands and smart glasses will all incorporate AI processing, while massive computing centres will handle top-tier tasks, making application scenarios ever more diverse.