On February 28, Hangzhou convened an AI improvement summit on the Hangzhou Civic Center aimed toward positioning town as China’s main synthetic intelligence (AI) hub. According to East Money and Xinhua, 12 AI initiatives had been signed, with a complete funding of CNY25.5 billion (US$3.71 billion).
The solely challenge devoted to AI inference GPUs was Sunrise’s “High-Performance GPU and Inference Chip R&D Project.” The Zhejiang-based firm signed the challenge in Hangzhou as a part of town’s computing infrastructure initiative.
Sunrise Co-CEO Wang Zhan mentioned computing energy, not mannequin functionality alone, will decide future industrial improvement, including that Hangzhou’s AI strategy affords a supportive surroundings for the corporate’s growth.
In an announcement, Sunrise mentioned the settlement marks a brand new stage of its Hangzhou growth and can help town’s AI computing infrastructure and broader innovation framework.
Sunrise was beforehand the large-chip unit of SenseTime and was spun off on the finish of 2024. It develops high-performance GPUs and multimodal inference chips for industrial deployment. The firm cites a number of generations of validated merchandise, although statements comparable to “ten-thousand-card-scale delivery” originate from company and native disclosures quite than unbiased third-party verification.
China’s AI chip deployment expands throughout a number of distributors
Sunrise’s challenge displays native governments’ strategic help for home AI computing suppliers, however it’s not the one pressure in China’s AI chip deployment. Reports from worldwide and home media point out that China’s AI computing provide and substitution strategy has developed right into a multi-player construction.
Huawei Ascend Series: A unified platform for coaching and inference
Huawei’s Ascend collection, together with the 910C, has been cited by main media as certainly one of China’s major home platforms for large-model coaching and inference. Reports point out the corporate is getting ready for large-scale shipments to satisfy demand for alternate options to Nvidia, though manufacturing capability and coverage constraints proceed to restrict provide.
US officers have indicated that home manufacturing might stay restricted, with one estimate suggesting 2025 shipments may very well be capped at about 200,000 models, highlighting ongoing provide chain and know-how constraints.
Multiple home AI accelerator suppliers
Beyond Huawei, one other main supply of AI chip provide in China comes from AI accelerators and inference-focused processors.
Cambricon is ceaselessly cited as a consultant home AI accelerator vendor. Its MLU collection has appeared on authorities and enterprise procurement lists. The Financial Times reported that Huawei and Cambricon chips had been included in official home procurement packages to help indigenous know-how improvement.
Reuters reported that the China Unicom data center in Xining, Qinghai, deployed multiple domestic AI chips, together with merchandise from Alibaba’s T-Head, MetaX, and Biren. The challenge illustrates real-world deployment of Chinese AI processors in large-scale infrastructure.
The Reuters report additionally said that Moore Threads and Enflame Technology are anticipated to produce chips for later phases of the information heart and associated cloud initiatives.
Chinese media have reported that Enflame is advancing a STAR Market IPO, indicating continued capital market exercise alongside product improvement.
Credit: AFP
Domestic CPU alternative efforts broaden
At the CPU degree, distributors comparable to Loongson and Phytium are positioned as home alternate options in authorities, state-owned enterprise, and industrial server environments, although they don’t instantly change AI coaching accelerators.
China is progressively limiting the usage of Intel and AMD chips in government-funded techniques to advertise domestically developed alternate options, offering coverage help for distributors together with Loongson and Phytium.
China’s AI chip ecosystem enters a multi-vendor section
Sunrise’s GPU and inference chip challenge in Hangzhou represents one instance of collaboration between native governments and rising chipmakers. However, broader deployment tendencies reported by main world media and first-hand challenge circumstances present:
● Huawei Ascend is without doubt one of the few home computing platforms with concrete shipments and deployments in large-model coaching.
● Alibaba T-Head, MetaX, Biren, and Cambricon chips have been deployed or deliberate in main knowledge heart initiatives comparable to China Unicom’s amenities.
● Moore Threads and Enflame are rising as extra provide sources.
● Chinese CPUs, together with Loongson and Phytium, are advancing substitution in server and authorities purposes below coverage help.
China’s AI chip ecosystem is shifting from reliance on a single provider to a multi-vendor construction. However, gaps stay in computing scale, manufacturing capability, and world ecosystem integration in contrast with Nvidia and established worldwide provide chains.
Article edited by Joseph Tsai
