In a dynamic shift within the world of artificial intelligence, Chinese startups are increasingly turning to the GeForce RTX 4090 graphics card, a gaming GPU, in response to restrictions on high-end data center NVIDIA GPUs. This trend comes as NVIDIA’s alternative offerings, including the H20 AI accelerators, have proven less appealing to smaller companies in terms of cost and value.
At the heart of this pivot lies the US sanctions enacted on China, which prevent sales of NVIDIA’s top AI chips, such as the A100 and H100, within the country. Faced with limited options, many Chinese AI firms are now choosing the GeForce RTX 4090 as a more economical and readily available solution for their computational needs.
Despite being a gaming card, the RTX 4090 has been favored due to its strong performance and relative benefits in both availability and pricing over the H20 AI GPU, which only began mass production in the second quarter of 2024. The H20, a scaled-down version of the H100, struggles to compete as it offers substantially lower TF16 FLOPS of theoretical performance and does not stand out in power efficiency. Its higher cost makes it an impractical choice for budget-conscious startups that are concentrating on optimizing their investments.
In an interesting development, these gaming GPUs are not just accessible but also modifiable for the AI sector, with a China-specific variant — the RTX 4090D — being equipped with double the standard memory to better support large language models.
While the H20 accelerators demand premium pricing and are mainly utilized by large-scale enterprises like Alibaba and Tencent, smaller AI startups are strategically installing more RTX 4090 units. For instance, one company reports plans to incorporate an additional 100 RTX 4090 GPUs in the coming month, highlighting the significant role this gaming GPU is playing in the Chinese AI market.
Interestingly, the reliance on the RTX 4090 isn’t a new phenomenon, as NVIDIA’s H20 had not been an available alternative until earlier this year. Consequently, startups had little choice but to seek out the best available technology to meet their needs.
Despite these market dynamics, large-scale deployments of H20 accelerators to major data centers suggest that NVIDIA stands to make considerable revenue from these high-end units, even as smaller startups opt for the gaming GPUs. Given the current state of technology and market restrictions, the trend of startups favoring gaming GPUs for AI applications is likely to continue until more suitable and affordable AI-specific alternatives emerge, whether from NVIDIA or its competitors.
The narrative unfolding in the Chinese AI sector underscores the adaptability and innovation of startups in response to regulatory challenges and market forces. By leveraging gaming GPUs like the GeForce RTX 4090, these companies continue to drive the AI frontier forward under complex and rapidly evolving conditions.
