The author is a Reuters Breakingviews columnist. The opinions expressed are her own.

By Robyn Mak

HONG KONG, Jan 24 (Reuters Breakingviews) - In 2023, smartphones-to-silicon conglomerate Huawei quietly released its flagship Mate 60 Pro handset. The launch, while muted, was worth celebrating in the People’s Republic: the device featured a made-in-China chip that had previously seemed out of reach amid crippling U.S. sanctions. Late last month, Hangzhou-based startup DeepSeek added the latest technological surprise by claiming to have trained a large language model on a par with the offerings of OpenAI and Meta Platforms META.O at a fraction of the cost.

Those accomplishments underscore the ways in which the People’s Republic is finding workarounds as the West tries to curb its progress on artificial intelligence. At stake is a potential economic boost that could be worth $6 trillion a year, using McKinsey estimates. President Xi Jinping has also prioritised making the People’s Liberation Army a world-class military force, including by developing AI capabilities in areas like unmanned weapons and data processing. It’s still early days, but the evidence so far suggests China could just about keep pace with the West.

Consecutive U.S. administrations, along with the country’s allies, have imposed investment and trade restrictions on the grounds that chip and AI advances in the People’s Republic pose a security risk. The latest blow came earlier this month, when the U.S. government restricted the flow of advanced graphics processing units (GPUs) for AI data centres, in a move that looked partly intended to stop China getting its hands on prized Nvidia NVDA.O kit through third countries.

Beijing has responded to the years-long onslaught by funnelling billions of dollars – through subsidies, research grants, tax breaks, cheap bank loans and more – to domestic champions. Shenzhen-based Huawei is leading the charge with advanced silicon, including a 2019 processor that matched Nvidia’s best at the time, and the now-infamous 2023 Mate 60 smartphone chip. The privately held group is a formidable innovator, spending almost one-quarter of its $100 billion revenue in 2023 on research and development, compared with 11% to 19% for Alphabet GOOGL.O, Meta, Amazon.com AMZN.O and Microsoft MSFT.O.

It’s far from alone, however. As Nvidia’s top-of-the-line GPUs grow scarcer in China, other local outfits like $40 billion Cambricon Technologies 688256.SS are vying to fill the gap. Citigroup analysts estimate that Jensen Huang’s company will see its market share in the People’s Republic fall as low as 50% in 2025, compared with 80% in 2024. Shanghai-listed Cambricon is on track to more than double its top line this year, to 4.3 billion yuan ($587 million), reckon analysts at Bernstein. The hope is that, with sufficient money and state support, Chinese chip designers can break Nvidia’s bottleneck.

Another key battleground is chip manufacturing, where Western-aligned players like Taiwan Semiconductor Manufacturing (TSMC) 2330.TW and Dutch kit maker ASML ASML.AS have dominant positions. China’s $60 billion Semiconductor Manufacturing International (SMIC) 0981.HK has racked up some important wins in this arena, including mass-producing the 2023 Huawei smartphone processor, which was 7 nanometres using the chip industry’s key measure of sophistication. Progress has admittedly slowed since then, partly because of limited access to ASML’s newest circuit-printing equipment, which $890 billion TSMC is using to make cutting-edge 2-nanometre chips.

Faced with that constraint, Chinese companies are adapting. Huawei, for instance, spent more than four years redesigning its AI processor to match SMIC’s manufacturing capabilities. Other workarounds include clustering, which involves grouping less-powerful processors together, and training models on smaller datasets. Software developers are also tweaking algorithms to maximise power efficiency, while companies are deploying trained models quickly to generate answers at an earlier stage.

The results have been impressive, showing little evidence of a huge gap between Chinese and U.S. AI offerings. E-commerce group Alibaba’s 9988.HK Qwen large language model is among the world’s top-ranked open-source products, while social media behemoth Tencent’s 0700.HK software has on some metrics outperformed Meta’s Llama 3.1. Local upstart DeepSeek’s globally competitive model also seems to be one of the most cost-effective: the company claims to have trained it in two months for just $6 million.

In adoption, too, China seems to be doing well. A survey last year of 1,600 worldwide industrial decision-makers by U.S. analytics software company SAS and Coleman Parkes Research found that 83% of Chinese respondents said they used generative AI, the highest among 16 other countries and regions including the United States. TikTok owner ByteDance, search engine operator Baidu 9888.HK, fintech group Ant and others have all rolled out popular AI-powered personal assistants that can write emails, generate videos, hail taxis and more. Ruth Porat, Alphabet’s chief investment officer, said in an interview this week that China is “on par and may even be a bit ahead on ... diffusion of basic capabilities”, referring to the ability to spread AI processes across the economy. That’s a particularly promising assessment for a country where many businesses have been slow to digitise.

In the vital manufacturing sector, the People’s Republic is already the largest market for industrial robots, accounting for more than half of the world’s installations, according to the Information Technology and Innovation Foundation. For large manufacturers like carmaker BYD 002594.SZ, injecting AI into highly automated factory lines could deliver a further production boost.

China’s progress on military AI is harder to judge. Some security analysts argue that its weapons like precision-guided missiles are less sophisticated than U.S. equivalents. Integrating subpar hardware with AI systems would, in theory, have a minimal impact. Yet the weaponry gap could close: the U.S. Department of Defense’s latest report on China’s military flagged rapid advancements in unmanned aerial systems, known as drones, which had “matched U.S. standards”. China’s forces are also making progress developing AI applications, using in-house models or open-source ones like Meta’s, Reuters reported citing academic papers and analysts.

Over the long term, the real prize would come from SMIC making another 2023-style manufacturing breakthrough, allowing it to produce smaller and more powerful chips. That would loosen the West’s chokehold on sophisticated hardware. It’s hard to bet against that outcome, given the resources that Beijing could in theory throw at the problem. Joe Biden’s U.S. Commerce Secretary Gina Raimondo in December described efforts to hold back China’s semiconductor progress as a “fool’s errand”. By extension, the same may be true for AI.

Follow @mak_robyn on X

(Editing by Liam Proud and Oliver Taslic)

((For previous columns by the author, Reuters customers can click on MAK/ robyn.mak@thomsonreuters.com))