Baidu has just placed a large order with Huawei for its Ascend AI chips, with the deal happening in China because of US sanctions and control of what Chinese companies can buy from the likes of AMD and, especially, NVIDIA.
![Baidu turns from NVIDIA AI GPUs, begins ordering Huawei's AI chips after US sanctions 605 Baidu turns from NVIDIA AI GPUs, begins ordering Huawei's AI chips after US sanctions 605](https://static.tweaktown.com/news/9/4/94257_605_baidu-turns-from-nvidia-ai-gpus-begins-ordering-huaweis-chips-after-us-sanctions.jpg)
The cloud giant purchased 1600 of Huawei's new Ascend 910B AI chips, which the Chinese company developed as an alternative to NVIDIA's A100 AI GPU. Reuters' sources report that Baudi will use the 1600 new Huawei AI chips for 200 servers and that by the end of last month, the company had delivered over 60% of the order -- or around 1000 chips.
Baidu's AI chip order with Huawei is said to cost around 450 million yuan ($61.83 million USD), and Huawei will deliver each and every one of the 1600 x 910B Ascend AI chips by the end of the year. The order of $61 million worth of AI GPUs isn't much when you consider that companies are buying hundreds if not thousands of high-end NVIDIA AI GPUs for tens of thousands of dollars each... but it signals something more important: Baidu moving -- even if it's slowly -- away from NVIDIA AI GPUs and into using Huawei AI GPUs.
One of the sources said: "They were ordering 910B chips to prepare for a future where they may no longer be able to purchase from NVIDIA".
- Read more: AI platform floats in ocean as 'sovereign nation-state' with 10,000 x NVIDIA GPUs
- Read more: US gov restricts NVIDIA AI GPUs to China, more 'vigorous' sanctions
- Read more: NVIDIA's new cut-down A800 GPU created because of US export controls to China
- Read more: US restricts NVIDIA from exporting H100, A100 chips to China, Russia
This makes sense, given that US sanctions from the Biden administration are continuing to tighten, with China and other countries heavily affected by not being able to secure AI chips from the best in the business. They're having to resort to using lower-end chips just because they simply can't buy higher-end AI chip designs.