Amid market rumors that the Beijing authority has been advising local companies not to use NVIDIA’s H20, which is tailored for the Chinese market, tech conglomerate Huawei is said to initiate sampling of its latest AI accelerator, Ascend 910C, to Chinese customers, according to reports by the South China Morning Post and Tom’s Hardware.

According to the South China Morning Post, large Chinese server companies and internet firms have received the samples of the Ascend 910C, which is regarded as an upgraded version of the Ascend 910B.

Sources cited by a previous report of the Wall Street Journal noted that TikTok’s parent company ByteDance, search giant Baidu, and state-owned telecom operator China Mobile are in preliminary talks with Huawei to secure the Ascend 910C chip.

Citing remarks from Eric Xu, Huawei’s Rotating Chairman, the South China Morning Post indicates that as the AI chips embargo launched by the U.S. is unlikely to be lifted soon, the scenario gives Huawei an opportunity to step in.

Huawei’s Ascend 910B, which the company claims to rival NVIDIA’s A100, has been popular among multiple industries across the country for AI model training. According to the report by the South China Morning Post, the tech giant’s Ascend solutions were used to train roughly half of more than 70 of China’s top large language models as of last year.

According to the report, NVIDIA is projected to ship over 1 million H20 GPUs to China this year, generating around USD 12 billion in revenue. Initially, demand for the H20 was sluggish, but sales have steadily gained momentum in recent months.

In terms of the upcoming Ascend 910C, Tom’s Hardware notes that it may reportedly sell for roughly USD 2 billion together this year, as the launch time would probably fall in the fourth quarter.

However, it is worth noting that Huawei also stands to gain by locking Chinese companies into its Ascend 910C hardware-software ecosystem, the South China Morning Post suggests. Citing a server company employee, the report states that if a firm buys Huawei’s AI chips, it may also need to purchase other offerings, like network and storage solutions, which might cause some hesitation.

Read more

• [News] Huawei Rumored to Launch New High-End AI Chip, Potentially Rivaling NVIDIA’s H100
• [News] U.S. Reportedly Enhancing Chip Export Restrictions to China, while NVIDIA’s H20 Might Get Banned

(Photo credit: Huawei)

Rumors have been circulating that NVIDIA has stopped taking orders for its H20 chips customized for China since August. Now, according to the latest report by Bloomberg, regulators in China have been advising companies against buying H20, as part of the country’s strategy to bolster its semiconductor industry and respond to further US sanctions.

As the initiative aims to boost the market share of domestic Chinese AI chip manufacturers, Huawei and Cambricon Technologies, which are leading AI processor makers in China, may turn out to be the major beneficiaries, Bloomberg suggests.

Beijing’s approach has been more of a guideline than a strict prohibition, as the authority still hopes to support its own AI startups, the report notes.

However, it is indicated that in recent months, several Chinese regulators, including the Ministry of Industry and Information Technology, did issue the so-called “window guidance”—informal instructions that lack legal authority—to minimize the use of NVIDIA.

It is worth noting that China has a thriving AI sector amid US restrictions. Major tech player like ByteDance and Alibaba are making significant investments, while numerous startups are vying for dominance. According to an earlier report by The Information, it is rumored that ByteDance has ordered over 200,000 NVIDIA H20 chips this year for AI model training, costing it over USD 2 billion.

In addition, there are six rising stars in the country’s development of large language models, which are crucial for generative AI, including 01.AI, Baichuan, Moonshot, MiniMax, Stepfun, and Zhipu, Bloomberg notes.

According to Bloomberg, some companies are disregarding the Chinese directive to avoid H20 chips, hastily acquiring more before a potential US sanction by the end of the year. However, they are also purchasing domestic Huawei chips to appease Beijing.

As early as in 2022, the US government prohibited NVIDIA from selling its most advanced AI processors to Chinese clients to curb Beijing’s technological progress. In response, the AI chip giant launched a series of AI chips tailored for the Chinese market, including H20, L20 and L2. According to a previous report by Wccftech, H20 GPU has 41% fewer Cores and 28% lower performance versus H100.

NVIDIA declined to comment to Bloomberg’s report, neither did China’s Ministry of Commerce, Ministry of Information and Technology, and Cyberspace Administration respond, Bloomberg notes.

In a separate statement, NVIDIA CEO Jensen Huang noted in an interview with Bloomberg Television that he is focused on serving customers in China while adhering to US government restrictions.

Read more

• [News] Nvidia’s Custom H20 Chip for China Reportedly Halted

(Photo credit: NVIDIA)

Nvidia has stopped taking orders for its H20 chips since August, according to Chinese media outlet CLS, citing distributors. While there has been no official notice, sources suggest distributors are “waiting on the latest updates”. One industry insider said Nvidia began accepting H20 chip orders in February, shipping them in April, with bulk shipments following in May. However, several distributors recently reported that they are no longer accepting H20 orders, though some still have stock available.

The report also quoted a distributor saying, “Large firms have mostly stockpiled, while smaller ones might still have some demand.” Orders are still being taken, but interest in the H20 chip is low due to its underwhelming performance and relatively high price, with some even describing it as a “chicken rib”—something of little value. Nvidia declined to comment on these claims.

According to Commercial Times, the U.S. had imposed export restrictions on chips to China, prompting Nvidia to develop three custom chips for the Chinese market based on its H100 series, with the H20 being the most powerful. The H20, built on the Hopper architecture, features 96GB of HBM3 and offers 4.0TB/s memory bandwidth. However, it has 41% fewer GPU cores and 28% lower performance than the H100.

Earlier, Sina Finance reported that Chinese firms, including ByteDance, have been stockpiling Nvidia chips, with ByteDance reportedly holding over 100,000 units. H20 orders for 2024 are expected to reach between 320,000 and 330,000 units, with 140,000 to 150,000 already delivered. Additionally, ByteDance plans to purchase more than 30,000 H100 chips from overseas, with potential follow-up orders for B200 and H200 chips.

(Photo credit: Nvidia)

ByteDance, the parent company of TikTok, is said to be collaborating with TSMC, eyeing for the mass production of two self-developed AI chips by 2026, according to reports by Economic Daily News and The Information.

ByteDance’s AI chips are expected to be made with TSMC’s 5nm node, which would be one generation behind the foundry giant’s most advanced process, the reports suggest, making the move comply with the U.S. export regulations to China. The chips are similar to NVIDIA’s next-generation flagship AI chip, Blackwell, which are manufactured with TSMC’s 4NP node.

Citing sources familiar with the matter, the reports note that the tech giant in China aims to reduce its reliance on NVIDIA for AI model development. Though the chips are still in the design phase and the plan is subject to change, ByteDance’s self-designed chips could save billions of dollars compared to purchasing NVIDIA’s products, according to the reports.

The Information estimates that ByteDance’s spending on developing generative AI models has been increasing, and it is rumored that the company has ordered over 200,000 NVIDIA H20 chips this year, costing it over USD 2 billion, with some orders still pending delivery.

In response to US export bans, NVIDIA launched AI chip H20, L20 and L2, specially designed for the Chinese market earlier this year. According to a previous report by Wccftech, H20 GPU has 41% fewer Cores and 28% lower performance versus H100. Still, the product is reportedly seeing strong demand for AI servers among Chinese Cloud Service Providers (CSPs) and enterprises, including Huawei and Tencent.

However, due to its lower computing power, Chinese companies need to purchase more H20 chips to build clusters with equivalent computing capacity, which raises costs, Economic Daily News notes.

According to TSMC’s financial report in the second quarter, North American clients contributed 65% of its total revenue. While China, the second-largest market, contributed 16% of its quarterly revenue, with a significant jump from 9% in the first quarter and 12% during the same period last year.

Read more

• [News] China Plans to Set at Least 50 AI Standards by 2026, from Language Models to Semiconductors
• [News] Broadcom Reportedly Developing 5nm AI Chip with ByteDance, Manufactured by TSMC

(Photo credit: ByteDance)

According to a report from wccftech, AMD CEO Lisa Su has indicated that she believes the AI Supercycle has just started, and the company has accelerated its product development plans to meet the substantial market demand.

In addition to NVIDIA, AMD is a significant player in the AI market as well. This is not only due to its market impact but also because the company has significantly expanded its AI product portfolio over the past few quarters, attracting attention from major clients like Microsoft and Amazon.

While AMD has not yet replicated NVIDIA’s success in the market, the company remains optimistic about the future, which is why it believes the AI boom has only just begun.

A few months ago, AMD outlined its AI chip development roadmap for the next year. The “Advancing AI” event in this October will showcase the next-generation Instinct MI325X AI chip.

The flagship Instinct MI350 AI chip is scheduled for release in 2025, followed by the Instinct MI400 AI chip in 2026. Despite AMD’s advancements, there remains a generational gap, as competitor NVIDIA is poised to launch its Blackwell architecture AI chips in the coming months.

This move is particularly noteworthy as AMD is focusing on competing with NVIDIA’s CUDA on the software front.

Read more

• [News] AMD Unifies RDNA and CDNA into UDNA Architecture, Aiming to Compete with NVIDIA’s CUDA
• [News] AMD Could Emerge as a Prospective Buyer for Intel’s Potential Sale of Altera

(Photo credit: AMD)