Back to list

AI Business36氪

Tech Giants Wake Up: Selling Chips Beats Racing on Models

36氪2 min read
Read original
Tech Giants Wake Up: Selling Chips Beats Racing on Models

AI Summary

According to IDC, China's data center AI accelerator shipments reached 2.008 million units in the first half of 2026, with non-GPU accelerators (1.016 million units) surpassing GPUs for the first time. Major tech companies such as Alibaba, Baidu, and Tencent are turning to chip sales to generate revenue as the model race grows costly.

AI Insights

China's AI chip market is shifting as non-GPU accelerators now ship more units than GPUs, driven by domestic players like Huawei Ascend, Alibaba's T-Head, and Baidu's Kunlun. Large tech firms are monetizing self-developed chips to offset the high cost of model competition, with inference demand from agents and enterprise deployments becoming the key growth engine. The strategic implication is that full-stack AI giants can turn internal validation into external revenue, while customer-supplier-investor relationships increasingly blur competitive boundaries across the ecosystem.

Analysis written by AI from the original article. Check it against the source.

Translated text

Two changes are accelerating in China's AI chip market. Nvidia's share is being eroded by domestic vendors, and shipments of non-GPU accelerators are rising. IDC data released recently confirms this: in the first half of 2026, China's data center AI accelerator shipments totaled 2.008 million units, with non-GPU accelerators at 1.016 million, surpassing GPUs for the first time. Nvidia remains first, but its gap with Huawei Ascend is only 6 percentage points. Alibaba CEO Eddie Wu disclosed in March that the annualized revenue of its T-Head chip unit had reached the billion-yuan level. At the September Yunqi Conference, Alibaba unveiled the Zhenwu V900, planned for mass production and commercialization in Q1 2027. Baidu is pushing to spin off Kunlunxin, and its Tianchi supernode is fully domestic. Large companies that have been loudly promoting full-stack AI, spanning chips, cloud, models, and applications, are finding that selling more chips is a practical choice amid intense model competition. While model companies still search for profitability, training and inference costs have become profit for chip giants. IDC reports that in H1 2026 Nvidia's China data center AI chip shipments fell 44% year over year, giving it a 34% share; Huawei held 28%, while T-Head and Kunlunxin held 7% and 4%, together exceeding Nvidia. Gartner-style forecasts from Deloitte suggest inference will account for about two-thirds of AI compute in 2026. Tencent, ByteDance, and even DeepSeek are reportedly developing their own chips. Big companies can validate chips with their own models, cloud, and high-traffic apps before selling to outside customers. Baidu's Kunlunxin P800 was fully verified internally before being pushed to external industries. Amazon has followed a similar path with Trainium, with its chip business exceeding a $25 billion annualized revenue run rate. Chinese giants pursue two monetization routes: selling hardware such as supernode systems, as Huawei does, and spinning off chip units for separate valuation and financing, as Baidu does. Kunlunxin filed confidentially for a Hong Kong listing in January. In China's AI chip ecosystem, today's major customer may be tomorrow's competitor. Tencent is the largest shareholder of Enflame, holding 20.26%, and 83.79% of Enflame's 2025 revenue came from Tencent.

Related News

3