Chinese tech companies aiming to train their own generative AI models are scrambling to cope with a U.S. ban on advanced artificial intelligence chips, leveraging their stockpiled chips and turning to domestically-produced ones. They are also awaiting the arrival of modified chips that comply with the new export control measures.
Semiconductor companies like NVIDIA, Intel, and AMD have said that they are intensifying efforts to develop revised AI chips that they can sell to China under the ban. NVIDIA plans to launch series of chips including HGX H20 as substitutes for banned advanced AI chips. Meanwhile, AMD’s MI 210, still available for sale in China, is considered competitive during the fine-tuning stage of large models.
Due to escalating demand and shrinking supply, prices for currently available computing power have surged. Market observers note that rental prices for NVIDIA’s A100 chip computing services have experienced dramatic increases, with some rising by as much as 100%. But more commonly, computing resources in high demand for training AI are out of stock.
Some companies are turning to domestically produced chips for their AI model training needs. Huawei’s Ascend 910B has already been commercially deployed in iFlytek’s data centers. Test results show that its performance is comparable to NVIDIA’s chips in some aspects, but the chip is significantly limited by manufacturing capacity.
According to the latest IDC report, Chinese cloud AI accelerator chip manufacturers are growing rapidly with the support of government policies. In the first half of 2023, Chinese AI servers used 500,000 locally sourced/developed AI accelerator chips, accounting for 10% of the entire server market in China. This number is expected to grow further in 2024.
Other companies are leveraging stockpiled chips. Baidu CEO Robin Li said that Baidu had stockpiled a significant number of AI chips, ensuring the continuous update of its AI models for the next one to two years. Tencent’s senior management also said that they have a high level of inventory for chips like the H800, which will support the updating of several generations of Tencent’s AI models.
Due to heightened demand, NVIDIA experienced a surge in revenue and profits in its third quarter. For the third quarter of the 2024 fiscal year, NVIDIA saw revenue of $18.12 billion, a 205.5% increase year-over-year. Operating profit was $10.42 billion, a staggering increase of 1633.7% year-over-year, with an operating profit margin of 57.5%.