How Huawei's expected new AI chip could disrupt the market

Details of the Ascend 910D processor remain scarce but it could be competitively priced

Huawei Technologies has long been producing chips, having established its HiSilicon processor arm in 1991. Reuters

China's Huawei Technologies is reportedly working on a new artificial intelligence-powered chip designed to take on Nvidia.

Details remain sparse on the processor named the Ascend 910D, which is still in its early stages of development, The Wall Street Journal reported this week. Huawei has not commented on the report.

Nevertheless, hype is building, along with interest on what this will mean for the industry, performance benchmarks and the wallets of users.

Chip history

Through the years, Huawei has used third-party processors on its devices, including those from US-based Qualcomm and Taiwan's MediaTek.

That does not mean Huawei is new to the game: back in 1991, the company founded its chip arm HiSilicon, which would eventually collaborate with Britain's ARM Holdings and create its in-house chip, Kirin, similar to Samsung Electronics' Exynos and Apple's A line of processors.

At present, Huawei has a unit called Ascend Computing, dealing with AI infrastructure. Ascend shares the same name of a line-up of smartphones Huawei began selling in the early 2010s, even spilling over to the first Mate devices.

GPU v CPU

A CPU, or central processing unit, has origins dating back to the 1950s and has been part of tech speak since. It basically acts as the brain of a computer, handling all processes.

On the other hand, a GPU – graphics processing unit – is a more specialised chip designed to support CPUs, specifically built to handle more complex tasks such as image rendering, video manipulation, AI workloads and everything gaming, its first major utility when GPUs popped up in arcades in the 1970s.

And while each processor is distinct, both are needed to meet varied computing demands, according to Intel.

Where do Huawei GPUs rank?

According to AIMultiple, an Estonia-based IT consultancy and research firm, Huawei GPUs ranked ninth for data centre workloads, based on multiple scenarios.

Unsurprisingly, Huawei chips are popular in China, especially as AI labs in the country could not buy the newest and best processors from US companies, AIMultiple noted.

In terms of value, it would be difficult to ascertain the value of Huawei's chip division, since it is a privately held company. California-based Nvidia has a market capitalisation of about $2.66 trillion, as of Thursday.

How much will it cost?

There are clues. Last month, Reuters reported Huawei was preparing to ship out its 910C chips, with further reports suggesting the company had received orders for about 70,000 units worth an estimated $2 billion.

That implies each processor would cost less than $29,000 – competitive considering Nvidia's H100 AI chips are believed to cost anywhere between $27,000 and $40,000. Bulk orders typically lower total costs.

Huawei's products – smartphones, laptops, wearables and others – have traditionally been priced lower than Samsung and Apple equivalents, allowing it to be more competitive. Before US sanctions hit hard, Huawei also ranked top for smartphones in 2018 and 2020 as the world's biggest manufacturer.

The success of Ascend 910D will depend on how the new chip performs and how widely it is accepted, but price tends to play a big role.

Memories of DeepSeek and Qwen

What is DeepSeek and why is it disrupting the AI sector?

01:15

When China-based DeepSeek launched its AI platform in January – it was virtually free and hyped to be even better that ChatGPT and the rest of the field, upending the entire AI world.

Then in March, DeepSeek itself was usurped by the launch of Qwen, the open-source generative AI service from Alibaba.

All share three common denominators – China-made, lower-priced and improving, if not already better, in quality.

It remains to be seen if big AI chip players will be affected by Huawei's new launch. But if DeepSeek taught us anything, it is that any new platform can be disruptive, costly and may cause a shift in perception on US tech, the argument being it is possible to create something good for cheaper. Just ask OpenAI and Nvidia.