
China gives nod to ByteDance, Alibaba and Tencent to buy Nvidia's H200 chips - sources
Jan 28 (Reuters) - China has given the green light to three of its largest tech companies to buy Nvidia's H200 artificial intelligence chips, four people familiar with the matter told Reuters, marking a shift in position as Beijing seeks to balance its AI needs against spurring domestic development.
ByteDance, Alibaba and Tencent have been approved to purchase more than 400,000 H200 chips in total, with other firms now joining a queue for subsequent approvals, the sources said, speaking on condition of anonymity.
The Chinese government is only granting approvals with conditions and the sources said they were still being decided upon.
A fifth source said that the licenses were too restrictive and customers were not yet converting the approvals to purchase orders.
China's industry and commerce ministries as well as Nvidia had not yet responded to requests for comment at the time of publication. ByteDance, Alibaba, and Tencent had not responded either.
Iluvatar CoreX targets Nvidia’s Rubin with GPU road map amid China chip push
Chinese chip designer Shanghai Iluvatar CoreX Semiconductor has unveiled a multi-year graphics processing unit (GPU) architecture road map, aiming to surpass Nvidia’s next-generation Rubin platform within two years
The company said its Tianshu architecture had outperformed Nvidia’s Hopper platform last year, while a second architecture, Tianxuan, would be built to benchmark against the US firm’s Blackwell platform. A third, Tianji, was designed to overtake Blackwell in 2026, while a fourth, Tianquan, was projected to surpass Rubin by 2027, after which Iluvatar CoreX planned to move towards what it described as a “breakthrough” architectural design.
The company highlighted efficiency gains from Tianshu, claiming more than 90 per cent effective utilisation of computing power. Innovations included features that reduce redundant memory access to boost effective bandwidth and cut power use, and ease resource contention through dynamic task allocation. It said Tianshu delivered about 20 per cent higher average performance than Hopper on DeepSeek’s V3 model.
Iluvatar CoreX also unveiled four products under its Tongyang, or TY, series for edge computing, covering performance from 100 tera-operations per second to 300 tera-operations per second. The company said its TY1000 outperformed Nvidia’s AGX Orin in various testing scenarios.
Founded in 2015, Iluvatar CoreX launched its first general-purpose GPU – TG Gen1 – which it described as China’s first domestically mass-produced GPU for AI training. A second generation followed in 2023, with TG Gen3 scheduled for mass production in 2026. A separate inference-focused product line, the ZK series, was launched in 2022.
Iluvatar CoreX listed in Hong Kong on January 8 – following peers including Moore Threads Technology and MetaX Integrated Circuit, which listed in Shanghai, and Shanghai Biren Intelligent Technology, which listed in Hong Kong – amid strong capital interest in Chinese chipmakers. Centurium Capital was a key external shareholder, holding nearly 23 per cent of the company before the IPO.
Iluvatar CoreX’s shares closed at HK$182 on Tuesday, giving the company a market capitalisation of HK$46.3 billion (US$6 billion), compared with Moore Threads at 284.6 billion yuan (US$40 billion), MetaX at 229 billion yuan, and Biren at HK$83 billion.
According to its prospectus, Iluvatar CoreX recorded revenue of 189 million yuan, 289 million yuan and 540 million yuan between 2022 and 2024, with a compound annual growth rate of 68.8 per cent, while net losses widened from 554 million yuan to 892 million yuan over the period.
In the first half of 2025, revenue reached 324 million yuan, with a net loss of 609 million yuan. The company said it had delivered more than 52,000 general-purpose GPUs by June 2025.