Since the Biden administration opened a new front in the technology war with China in September 2022, escalating restrictions to prevent China from acquiring and capitalising on American innovations have triggered an action-reaction dynamic whose outcome is impossible to predict. 

The intensifying US-China technology rivalry came up in a call between Presidents Joe Biden and Xi Jinping this week. Protesting the US “trade and technology suppression measures, Xi told his counterpart; “This is not ‘de-risking’ but creating risk.”

What is known as the Sullivan doctrine has at its heart the proposition that the US must establish and maintain the biggest lead possible over strategic rivals in technologies critical for national security.

These include advanced semiconductors, AI, advanced computing, biotechnology and green technologies. Ironically, these restrictions have forced a pace of indigenisation in China that might otherwise have occurred much more slowly. 

In the short term, the restrictions are a barrier to China; in the long term they may well provide an impetus to achieve.

US semiconductor restrictions deepen

The most important restrictions are limits on China’s access to advanced semiconductors and the equipment needed to make them. The latter includes Electronic Design Automation (EDA) tools where the US enjoys a virtual monopoly, and Deep and Extreme Ultraviolet Lithography (DUV and EUV) tools, where the Dutch company ASML has a similar virtual monopoly.

The most immediate objective of the US restrictions is to prevent China from making progress in the production of non-planar semiconductors – FinFET and Gate All Around - which offer higher density, greater speed, lower energy consumption and generally better performance than planar semiconductors are able to provide.

Since the initial restrictions announced by the US Department of Commerce in October 2022 – see Enodo Untangled: No Hiding Place in US-China Tech War for details – efforts have been made both to tighten up the restrictions and to press US allies into at least partially applying them.

A year after the initial restrictions were announced, additional measures were introduced. There are almost certainly more to come.

These include additional checks on Chinese firms attempting to evade export restrictions by routing shipments through other nations, and adding Chinese chip design firms to a trade restriction list that will require overseas manufacturers to gain a US license to fill orders from those companies.

Measures have also been taken to prevent US tech companies such as Nvidia from continuing to develop lower-powered Graphics Processing Units (GPUs) with data transmission speeds below 600 gigabits per second for the China market.

More generally, US tech companies planning to sell to China semiconductors whose performance is just below the levels at which restrictions apply will be required to notify the Department of Commerce, with the possibility that they may be required to apply for licences. US GPU manufacturer AMD has already fallen foul of this provision.

Sales to Chinese subsidiary companies outside China will also be restricted and licences will be required for the sale of technologies to third countries that might serve as conduits to China.

Meanwhile, US tech companies have tightened some of their own processes. Nvidia has disallowed the use of its CUDA software on third-party platforms, a move that will significantly complicate the efforts of Chinese tech companies to develop indigenous LLMs.

China’s response

China continues to decry US efforts to constrain its technological development. In March, Foreign Minister Wang Yi stated that the lengthening US sanctions lists had reached bewildering levels of absurdity: “if it (the US) only wants itself to prosper but denies other countries legitimate development, where is international fairness? If it persistently monopolizes the high end of the value chain and keeps China at the low end, where is fairness and competition?”

Xi Jinping, for his part, told Biden on Tuesday that China would prefer to cooperate with the US, warning: “If the U.S. insists on suppressing China's high-tech development, depriving China of its legitimate right to development, we won’t stand by indifferently”, according to a readout of the call by the Xinhua news agency.

But behind the scenes China is engaged in a massive effort to reduce its dependence on US technology and promote a new model for advancing China’s indigenous technology sector.

The watchword for these efforts is New Quality Productive Forces (新质生产力), a term first used by Xi Jinping in September 2023 during an inspection tour of the rustbelt province of Heilongjiang. He elaborated on the concept in a speech to a study session of the Politburo in January 2024.

The idea is to promote innovation focusing on high technology, high efficiency and high quality through the encouragement of revolutionary technology breakthroughs, the innovative allocation of production factors and the in-depth transformation and upgrading of China’s industries.

The focus is very much on developing the “real economy,” Jin Zhuanglong, Minister for Industry and Information Technology, emphasised in a boosterish interview given to the People’s Daily in January.

Jin claimed that China uniquely possessed all categories in the United Nations industrial classification index and that China’s industrial added value in 2022 amounted to 33.3% of GDP. He also stated that China’s investment in R&D constituted 25% of the world total, and that China had established 27 national and 260 provincial manufacturing innovation centres. But he did not dwell on the real challenges faced by China in key “choke-point” areas, in particular advanced semiconductor production.

China’s determination to move up the technology value chain and to indigenise advanced technology production is not new, and is driven by both economic and geo-political considerations. The latter are arguably the more salient.

A combination of economic slow-down, Western decoupling/derisking and an impending demographic cliff edge could condemn China to a middle-income trap unless it can capture a larger share of advanced product value and significantly increase efficiency, productivity and competitiveness.

If China can establish a dominant role in advanced technologies, especially in terms of international standard-setting, that will confer both economic and geo-political advantage as the world becomes increasingly dependent on Chinese technology systems.

China has for many years been seeking to reduce dependence on US technologies which it perceives as constituting a strategic vulnerability. It has successfully decoupled from the US in terms of hardware and Internet technology.

It is now targeting enterprise software with a requirement for Chinese government departments and SOEs to replace foreign enterprise software by 2027, thereby threatening an area of major profitability for US companies such as Oracle and Microsoft.

The next front in this war will be an effort to indigenise semiconductor production, the lack of which remains a key vulnerability.

Semiconductors

By far the most important of the “choke-point” technologies, semiconductors best illustrate China’s new model.

China’s institutional response to US technology controls represents an intriguing combination of dirigisme with incentivising collaborative efforts within the private sector. 

In July 2023 a Central Science and Technology Commission was established to focus effort on innovation, and in 2024 the national S&T budget was increased by 10%.

The 2024 annual legislative session announced efforts to reduce bureaucracy in the S&T sector with a view to enabling China’s scientists to spend more time focusing on innovative research. Experience suggests this will not be easy to achieve.

A Semiconductor Leading Small Group headed by Politburo Standing Committee member Ding Xuexiang has been set up to integrate work in the sector and focus effort on resolving key challenges. 

Solving these challenges will involve a new approach to public-private collaboration by facilitating the transfer of state-backed R&D to designated conglomerates, consisting of a major SOE that provides funding for a constellation of private sector companies incentivised to work together to overcome technical challenges.

This approach is not limited to semiconductor manufacture but also to design automation tools, advanced materials, advanced packaging and new systems engineering approaches. Notable beneficiaries of this new approach are telecommunications national champion Huawei, semiconductor manufacturers SMIC and YMTC, and toolmakers Naura and AMEC.

US technology restrictions have inadvertently created the beginnings of a virtuous circle in which companies such as national champions SMIC and YMTC provide feedback to local suppliers who are then able to improve their designs.

The irony is that greater uncertainty over continued access to critical US capabilities such as Electronic Design Automation (EDA) tools and advanced photolithography systems has pushed China’s semiconductor manufacturers to rely on local alternatives.

Gauging the exact effect of this new approach has been difficult, and there have been some false starts.

China’s top photolithography company SMEE announced in December 2023 that its own Deep Ultraviolet Lithography (DUV) machines could produce semiconductors at the 28 nm node - but that announcement has since been rescinded, and there is no sign of the 28 nm chips. And while in 2023 Huawei and SMIC stunned the US by producing a 7 nm semiconductor for Huawei’s Mate 60 Pro mobile phone, this involved a process of repeated (DUV) immersions that is unlikely to be scalable or commercially competitive and, critically, relied on equipment from US companies Applied Materials Inc. and LAM Research Corporation.

The challenge for China in moving to advanced semiconductor production is not just the development of advanced ultraviolet photolithography equipment but also critical inputs such as photoresists, the production of which continues to lag behind western competitors. 

Even if China makes progress in these areas, it will struggle to replace the expertise of US engineers who had been providing support for the advanced equipment used by Chinese semiconductor manufacturers and who have been withdrawn following the imposition of US restrictions.

On a more positive note for China, Huawei’s indigenously produced Ascend 910B graphics processing unit (GPU) supposedly has 80% of the performance of Nvidia’s top-of-the range A100 GPU, though it is unclear whether it can yet be produced on a significant scale. Huawei has also claimed a breakthrough in the production of indigenous EDA software, probably working in conjunction with Empyrean, China’s EDA national champion.

Huawei has become a target for additional US action by virtue of its covert acquisition of a number of Chinese semiconductor manufacturers. These include DRAM chip manufacturers Swaysure and Fujian Jinhua, logic chip manufacturer Shenzhen Pensun Technology Co, image sensor and radio frequency (RF) chip manufacturer PXW and power management chips and microcontroller manufacturer Qingdao Si’en. All of these companies now look set to be hit by new US sanctions.

China is in a much stronger position when it comes to the production of legacy semiconductors. And the announcement by US Commerce Secretary Gina Raimondo that funds from the CHIPS and Science Act would be devoted to the production in the US of legacy semiconductors has galvanised China’s efforts to establish a leading role. China now plans to produce one million more wafers per month than in 2023 despite a decline in global usage rates from 100% in 2020 to between 65 and 75% at present. China is expected to increase its global share of legacy semiconductor manufacture from 31% in 2023 to 39% by 2027.

Data and artificial intelligence (AI)

In 2020 the State Council defined data as a new factor of production alongside land, labour, capital and technology, and hence critical for China’s economic development and national security.

In 2021, the government consolidated this position through a series of legislative and administrative measures designed to create a data-driven ecosystem, including Data Security and Personal Information Protection Laws, alongside anti-monopoly regulations intended to curb monopolistic behaviour by major digital platforms.

In terms of practice, China has established a national network of 48 data exchanges at which individuals and organisations can buy and sell data. Thus far, these exchanges have struggled to gain traction with many private sector companies, as they are reluctant to incur the expense involved in cleaning up their data to the required standard. Much data brokerage is therefore still undertaken by private data brokers.

The National Data Administration has further established a Data Element X Plan (数据要素X) covering the period 2024-2026, working with sixteen government departments to achieve a 20% annual increase in the growth of the data industry. The plan aims to unlock the multiplier effect of data across twelve industries with some 300 application scenarios. It also sets out three objectives: to enhance levels of data supply, optimise the data circulation environment and strengthen data security guarantees.

For AI, it is hard to determine China’s progress relative to the US, although it appears the US retains an overall lead in most areas.

During a March 2024 visit to Beijing’s Academy of Artificial Intelligence, Premier Li Qiang was told that China lagged significantly behind the US due to “a serious lack of self-sufficiency in building its own generative AI.”

As is the case in the US, established technology majors such as Baidu and Alibaba occupy a dominant role by virtue of their access to large volumes of data, computing power and capital. The Chinese state has approved over 40 LLMs, but there is estimated to be over 200 in the country as a whole. Many are expected to drop out in the face of significant challenges.

China’s LLMs are trained on Meta’s Llama open-source system, but these models suffer from a lack of access to US GPUs – there has been significant stockpiling since the US tech controls were introduced but GPUs have a shelf- life of between one and two weeks – and locally produced alternatives are much less powerful. Moreover, while Chat GPT has been trained on 175 billion parameters, the average Chinese LLM is trained on between just 6 and 13 billion.

A major consideration for China’s generative AI is the CCP’s determination to keep control. In the summer of 2023, the Chinese Party-state issued regulations for the development of generative AI which stated that these systems must “align with core values of socialism” and must not undermine state authority, harm national unity or spread false information.

The video surveillance company iFlytek quickly fell foul of these regulations when one of the company’s AI tablets, aiding students in homework, generated an essay containing information critical of Mao Zedong. Several staff members were punished.

During Li’s visit to the Academy of Artificial Intelligence he was told that a major issue for LLM training was a lack of “quality content in line with the facts” – a euphemism for content aligned with CCP values. Li implicitly acknowledged this issue when he stated that henceforth the Party-state would seek to create a more relaxed environment for the development of the AI industry.

China may be lagging on foundational research, but AI systems have from the outset been made widely available to the public. Real-world applications have therefore seen a dramatic uptake, with a focus on schools, the workplace and e-commerce, where standards of customer service and feedback are being transformed.

Premier Li last month introduced the AI + initiative with the aim of “actively developing the digital industry, transforming traditional industries with digital technologies, and fully integrating digital technology into the real economy.”

A key element of China’s AI strategy is to develop intelligent computing – the optimisation of computing via autonomous perception, information gathering, analysis and reasoning. A working group comprising Alibaba Cloud, TenCent Cloud, Hikvision, Dahua, Uniview and the Wuxi National Supercomputing Centre will set standards. China aims to complete a national integrated computing power network by 2025 to provide the essential infrastructure for such activities.

A comparison between the US and China in AI remains difficult because the world’s only major AI powers are pursuing different objectives in different ways.

The US effort is led by the private sector, focused primarily on foundational science and subject to no meaningful regulation. China by contrast seeks to achieve a balance between innovation and regulation while focusing primarily on practical application for economic benefit.

It is reasonable to assume that China will play a waiting game, looking at how AI develops within the US before making critical decisions on how best to utilise this technology.

US starts to restrict Biotech 

Initially there was little evidence of US activity to restrict China’s advances in the biotechnology sphere. This was at least in part due to lobbying by US pharmacological groups highly dependent on China for manufacturing and research. This has now begun to change.

In February 2024, President Biden issued an Executive Order on Preventing Access to Americans’ Bulk Sensitive Personal Data and United States Government-Related Data by Countries of Concern. For “Countries of Concern”, read “China” which is engaged in collecting population-level datasets on a global scale by whatever means, including espionage.

The purposes of such collection are unclear. But the US government believes them to include espionage, commercial advantage, cognitive warfare and the development of biological weapons based on gene technology that can be targeted against specific nationalities. (Muslim Uighurs, in particular, are known to be targeted for biodata collection).

It remains to be seen whether this latter contingency is realistic given that susceptibility to pathogens is determined not by specific genes, but rather a complex interaction between multiple genes that may be hard to identify even with the aid of AI. Moreover, genetic diversity between major population groups is neither particularly significant nor consistent.

In the face of credible evidence that the PLA is pursuing research into bio-weapons, the US government is not disposed to take risks.

The US Congress has also begun its own actions to contain China in the sphere of biotechnology. A draft Biosecure act will seek to restrict US companies in receipt of federal funding from contracting with “foreign adversary biotech groups of concern deemed to be a threat to national security.”

This draft, which has cleared the committee stage in the Senate, is targeted specifically at the biopharma group Wu Xi App Tech and the Chinese genomics group BGI. Wu Xi is a major supplier of critical inputs for many of the world’s largest pharmaceutical companies including Eli Lilly and Pfizer. In 2023, Wu Xi’s earnings from sales to the US totalled RMB 23 billion, almost 60% of its total revenue. It remains to be seen how the draft will fare in Congress, but senior US officials have signalled to Enodo that the US government is keen to clip the wings of these and other Chinese biotech companies whom they believe to have close links to the PLA.

Conclusion

US determination to slow China’s progress in advanced technologies will almost certainly translate into further restrictions. The fenced yard will get bigger and the walls higher.

The two countries are now caught up in an action/reaction dynamic. The outcome defies easy forecasts, because the evolution of advanced technologies is so inherently unpredictable.

It is safe to say that China is developing a critical mass of advanced technology capabilities to the point where absolute advantage may not be that significant.

In the long term, US efforts to constrain China will therefore fail. Whether this leads to a total decoupling is hard to predict. But significant divergence seems inevitable.