Welcome to a/symmetric, our weekly newsletter. Each week, we bring you news and analysis on the global industrial contest, where production is power and competition is (often) asymmetric. To receive issues over email, subscribe here.

This week:

• China bets on industrial AI: Industry leaders and government players see industrial applications of AI as a way to gain an edge over foreign competitors.
• Weekly Links Round-Up: Apple edges closer to China, the EU raids Chinese airport scanner giant Nuctech, and Germany arrests suspected Chinese spies. Plus: could the US ban LNG exports to China?

China bets on industrial AI

A few months ago, we posed the question of how China might leverage asymmetries to gain a competitive edge in AI.

“Even supposing China lags its most advanced competitors in AI research,” we wrote, “could it gain an edge in developing and deploying industrial applications of AI (say, autonomous driving), in turn winning new leverage in global markets?”

Recent statements by industry and government players suggest that China is doubling down on industrial AI. The immediate upshot of that strategy is twofold: one, it could help China’s AI sector sidestep its current dependence on US technology; and two, it could amplify China’s existing competitive advantages in industrial production.

Going small and vertical

Asked by the moderator to compare Chinese and American LLMs, Zhou said the technology gap is “inevitable;” the question is how to catch up. Closing the gap on LLMs is “much less difficult than doing the same for lithography machines and chips, because it is software after all,” he said.

Then there’s artificial general intelligence, considered to be something of a holy grail — and requiring immense computing power. But might there be something more readily achievable than AGI, and that could confer equal or greater advantages? Zhou continues:

“…if you make the LLM smaller and smaller, and take the road of verticalization, industrialization, enterprization, and scenarioization,1 you may not need trillions or hundreds of billions of parameters, but only tens of billions of parameters, plus some private core data. With blessing, it can exceed GPT4 in a vertical unit.

This reduces the computing power requirements to very low, and maybe ten consumer-grade graphics cards can be used in one scenario. This is very suitable for our country, and this is also a scenario for us to overtake on the corners.

We have so many industrial categories, and the country attaches so much importance to the digital transformation of manufacturing. If all Chinese companies use this small-scale LLM and integrate LLM with business in their own scenarios, it will be as meaningful as us catching up with foreign countries in super AGI.”

A national strategy

Zhou’s wager is that marrying China’s industrial might with strategic uses of AI will help close the AI gap with the US — and then some.

State media accounts echo his assessment, suggesting a broader alignment between industry and government players on China’s competitive AI strategy. Economic Daily wrote this month:

“To develop LLMs, we must take the actual route of differentiation…In the face of the dominant advantages of other countries, blindly following is not conducive to real innovation…We should be application-oriented, build on existing advantages, and deeply integrate with vertical industries, which can not only stimulate new industrial momentum, but also forge new technological advantages.”

The Chinese government is already directing resources to industrial AI. In recent months, numerous local governments – including Beijing, Shanghai and Guangzhou – have rolled out shortlists of “application scenarios” to support and test out various industrial AI deployments, helmed by leading tech firms.

Meanwhile, the China Academy of Industrial Internet, run by the Ministry of Industry and Information Technology, this month called for direct state support to build a nationwide industrial dataset to power AI applications:

“Through measures such as capital subsidies, tax reduction and exemption, and policy tilt, we will guide market entities of key industrial industries…to open up their industrial data to public platforms and form a data resource pool covering key domestic industrial fields.”

Will it work?

China is of course not alone in pursuing industrial AI.

And it knows as much: a report published last month by the Tencent Research Institute on industrial LLM applications namechecks several foreign industrial AI initiatives, including Microsoft and Symphony AI’s Industrial LLM; Retrocausal’s Kaizen Copilot, a generative AI tool focused on optimizing manufacturing; and Nvidia’s ChipNeMo LLM for chip design.

The differentiating factor, as is often the case, is China’s centralized government direction and resource allocation. Through this, Beijing can direct domestic companies to pursue longer-term strategic objectives, aided by state support, including non-market incentives.

The end result could go both ways. Centralization can stymie innovation. It can also enable China to steamroll competitors that are more constrained by short-term profit demands. The latter dynamic has played out in other industries. There’s no reason why it can’t play out again in industrial AI.

Weekly Links Round-Up

️ Apple deepens its China presence. The iPhone maker counted 52 Chinese suppliers in its supply chain last year, up from 48 in 2022. That’s according to Apple’s latest annual list of suppliers, covering 187 companies. Even as the company expands its China footprint, sales of iPhones plunged 19% in the first quarter. Huawei’s phone sales surged 70% over the same period. (Nikkei Asia, CNBC)

️ The EU raids Nuctech. Authorities searched two European offices of the Chinese airport security scanner and demanded access to certain data, reportedly looking for evidence that the company had received state financial support from Beijing.

The raid suggests that Brussels is positioning to aggressively deploy its new tool designed to tackle market-distorting foreign subsidies. Sam Lowe has a good explaineron the differences between the EU’s anti-subsidy trade defence tool, and the newer foreign subsidies regulation tool.

And a side note: evidence of extensive Chinese state aid for Nuctech dates back to at least 2017, when Beijing prominently named the company a “single champion” to be supported in its pursuit of global market share, as our research into the Chinese industrial program shows. (SCMP, Most Favoured Nation)

️ Is the US considering banning LNG exports to China? A “temporary pause” by the Department of Energy in reviewing applications to export LNG may nudge it in that direction, write Gabriel Collins and Steven Miles. But they argue that any such ban would be high risk and lead to unintended negative consequences, including inviting retaliatory bans of critical materials and pushing Beijing ever closer to Moscow. (Foreign Policy)

️ Germany arrests suspected Chinese spies. Conveniently timed after chancellor Olaf Scholz’s confab with Xi Jinping last week, Berlin authorities on Monday arrested three German citizen suspected of spying for China. On Tuesday, police also detained an aide to a top European lawmaker on suspicions that he was working for Chinese intelligence. And on Wednesday, the head of Germany’s domestic spy agency said of China’s risks to German businesses: “We have numerous examples where a maybe-too-optimistic and positive attitude towards China trade led to these companies practically being dissolved.” (AP, Politico, Reuters)

And in case you missed it, here’s our take on Scholz’s China strategy from last week.

(Photo by mikemacmarketing/Wikimedia Commons)