Semigods: winners in the US vs China tech race

Convention dictates that as geopolitical rivalries intensify, cross-border trading relationships break down, growth suffers and investment potential diminishes. But this is not always true. Strategic moves by the US and China to defend their national interests underpin a new order, reshaping where opportunity can be found.

A worldwide battle over nanometres

Starting with bans on Huawei technologies in 2017, US protectionism has expanded across both Republican and Democratic administrations to cover advanced and clean-energy technologies. It is a key feature of the nation’s political agenda and the global macroeconomic landscape.

Crucially, the US is blocking exports of advanced semiconductors to China, stymying the country’s development of generative artificial intelligence (gen AI) technologies. In response, China is restricting exports of rare earth metals gallium and germanium, which are widely used in the semiconductor industry.

Similar measures could expand to other high-tech breakthroughs: Republicans are keen to block China from accessing quantum computing and sensing technologies, on the basis they may deliver compelling advantages in the future. And China is already narrowing access to antimony, an input for military and green-energy applications. 

But the market’s focus on gen AI has put advanced semiconductors in the spotlight. Is the battle over chips measured in billionths of metres disrupting the investment potential of this breakthrough technology? Not in our view, but it is influencing where opportunities are found. 

Intelligence unlocked: the gen AI supercycle

Gen AI is a step-change technology: its ability to identify patterns in unstructured data is opening possibilities that are only beginning to be exploited. Whereas traditional data processing can only make use of the 20% of information that exists as structured data, gen AI can harness the other 80% with the promise of unlocking huge efficiency gains.

Unstructured data can be used by retrieval-augmented generation (RAG), which augments the output of large language models (LLMs) with relevant information from external sources. This is intended to improve accuracy and reduce ‘hallucinations’, whereby a model perceives patterns that are either non-existent or irrelevant, generating outputs that may be incorrect or nonsensical.

Gen AI’s predictive capabilities mean that aside from turbocharging process automation, it can automate a wide range of tasks that previously necessitated human intervention. Applications range from sophisticated chatbots for customer service to digital assistants for legal research and medical diagnoses, industrial applications and agricultural management.

For example, models can trawl for past rulings associated with a legal case and help develop a litigation strategy by profiling past decisions made by a specific judge. In the healthcare industry, gen AI can aid drug discovery through compound design, manufacturing, dose optimisation, and data collection and analysis.

Currently, we are in the first phase of gen AI adoption, characterised largely by server upgrades and conversational outputs, and in which the main beneficiaries are technology infrastructure companies. In time, we expect gen AI to permeate the entire economy – beyond chatbots to agricultural management, assistant humanoids and robotic surgeons, benefitting different types of businesses and catalysing investment opportunities beyond hardware (see Figure 1).

FIG 1. The four key phases of generative AI adoption and their main beneficiaries¹

While installing gen AI servers costs up to six times more than traditional server upgrades, productivity gains mean revenue is four to five times higher, and chip maker TSMC² claims a gen AI server can hold 20 times more content.³ Furthermore, rapid technological advances have brought the innovation cycle down from two to three years for Intel-designed servers to one year for Nvidia’s solutions.²

The hyperscalers – companies like Microsoft, Google and Amazon² – are leading adopters, vying to make intelligent automation key to business productivity and efficiency. Even as the hyperscalers progress and other firms follow, we estimate that only 12% of servers globally will be AI-powered by the end of 2025³, leaving a long pathway for capex and associated growth opportunities in the coming years.

Semiconductors: at the frontline of the US-China rivalry

Gen AI requires highly advanced two to three nanometre (nm) semiconductor chips. By controlling the market in these chips, it is therefore possible to control development (the nm count is essentially an indicator of chip density – the lower the number, the more powerful the chip).

Given the strategic importance of AI, the US has moved from company-specific interventions to banning the export to China of any semiconductor technology surpassing 14 nm. Meanwhile, Dutch company ASML², the world’s only manufacturer of the leading-edge lithographic technology needed to make high-density chips, has been banned by its own government from exporting its most advanced machines to China.

Deprived of the necessary technology, Chinese companies are struggling to innovate and maintain commercial momentum, with Nvidia’s latest offering far outpacing Huawei’s² (see Figure 2). For example, Huawei is struggling to ship 5G-enabled phones and is losing market share as a result.

FIG 2. China’s top AI semiconductors trail the performance of Nvidia’s best chips⁴

It is notable that the US is limiting China’s access to technologies that are three to five years behind the latest advancements. This goes some way to balancing US national security objectives and the commercial interests of US firms, for whom China remains an important market. For example, in the 2022 fiscal year, Qualcomm² earned 62.4% of its total revenue from Chinese sales, while Tesla² (21.8%), Apple² (17.7%) and Microsoft² (12.4%) also depended on the Chinese market for a significant part of their revenue.⁵

In response to the ban, Nvidia² began marketing a set of chips specifically designed for the Chinese market, of which the H20 is the most powerful. While Huawei² claims to offer comparable performance for its in-house SAI chipset (the Ascend 910B), supply chain and system-level performance issues mean most Chinese buyers prefer the H20 for now.

Overall, US protectionism appears to be meeting the government’s aims: despite investing USD 150 billion in subsidies over the past decade, China currently has a domestic semiconductor industry that is noticeably behind the curve.

However, small-scale Chinese cloud providers are reportedly renting access to Nvidia's H100 and A100 chips, indicating that chips have been smuggled in to the country. For other firms, innovation is the answer: one Chinese startup uses a ‘mixture of experts’ approach to developing LLMs, delegating specific tasks to the best-suited networks instead of relying on the processing muscle of cutting-edge chips.

Place your chips: semiconductor investing opportunities

Several key factors arising from the impact of the US-China rivalry have implications for investing in the semiconductor sector. They are:

Capacity reshoring. Governments are offering chip makers incentives to increase production in their jurisdictions – although skilled labour shortages and high costs (even after subsidies) remain a barrier. Countries have created schemes offering subsidies and tax breaks to encourage semiconductor firms to move their operations:

US. The CHIPS and Science Act and the Facilitating American-Built Semiconductors Act provide over USD 52 billion in tax credits to boost research, development and production and to support the construction and improvement of manufacturing facilities. It has incentivised TSMC and Samsung² to build foundries in Arizona and Texas, respectively (see Figure 3).

EU. The European Chips Act commits USD 46 billion to strengthen R&D, develop the regional workforce and increase production capacity from 10% to 20% of the global market.

China. The Integrated Circuit Industry Investment Fund (known as the ‘Big Fund') delivers more than USD 50 billion in tax breaks over two phases.

Taiwan. The Invest Taiwan Initiative offers tax credits to semiconductor manufacturers, along with help in securing land, water and electricity.

South Korea. The K-Semiconductor Belt project offers tax credits with the aim of attracting USD 450 billion in private investment by 2030.

Such programmes are ambitious. However, beyond these incentives, the skills and consistent investment and development required to compete in this space are steep barriers to entry. For now, the supply chain for advanced chips is deeply integrated – from Taiwanese foundries and Dutch lithography to Korean memory expertise. The current opportunity set is clear and investors need to stay attuned to the progress of onshoring attempts to gauge how it is shifting.

FIG 3. Hot CHIPS: US funding for increased onshore semiconductor manufacturing (USD billion)⁶

Geopolitical risk. Some companies in the Chinese supply chain have been impacted due to an unavailability of key components or technology due to export controls. In contrast, some firms, like Taiwan-based Chroma², are gaining market share as Chinese customers seek to reduce their dependence on US and Japanese equipment⁷.

China localisation. As the increasingly bipolar world influences the supply chain, some Chinese semiconductor manufacturers will benefit from becoming local champions. For example, printed circuit board supplier Shengyi¹ has been gaining market share, while leading Chinese semiconductor equipment supplier Advanced Micro-Equipment² doubled its revenue from 2021 to 2023. Meanwhile, industry players including major tech firms Baidu, Alibaba and Tencent² are making significant efforts to develop gen AI chipsets inhouse. These efforts have taken place amid the USD 120 billion in semiconductor spending under the Made in China 2025 programme.⁴

Given the presence of key semiconductor and memory suppliers in Asia – from TSMC to SK Hynix and Samsung² – gen AI is a focus for our high-conviction equity strategy concentrating on the region. The role of semiconductors in the growth of renewable energy and the potential for gen AI to support more inclusive healthcare treatments also makes the technology a theme for fundamental equity strategies in our holistiQ¹ sustainable investment franchise. 

related content.

rethinking through the noise.

learn more

Corporate

Rethinking adversity to seize opportunity

September 27, 2024

Adapting to geopolitical risk: diversification helps To understand how investment strategies could be impacted by sustained geopolitical tensions, we assessed the historical returns of three portfolio types during periods of high, medium and low risk as gauged by the Geopolitical Risk Index from 1960-2024. The simulated portfolios are a capital allocation strategy with a 50/50 stock and bond split and two risk-premia approaches: 50/50 stock and bond, and another with stocks, bonds and commodities. The results are shown in figure 4. During episodes of heightened geopolitical risk – such as the Bay of Pigs Invasion, Vietnam War, fall of the Berlin Wall, 9/11 and the Russia-Ukraine conflict – average returns and risk/reward ratios declined. In these periods, as well as times of low geopolitical risk, the portfolio including commodities performed the strongest. This infers that diversification can help mitigate the market impact of major geopolitical events, while ensuring exposure to whatever return drivers lead recoveries. FIG 4. Average returns (left) and performance-to-risk ratios (right) for different portfolio types during various geopolitical risk environments, 1960-20248 Diversifying by risk premia, as opposed to capital allocation, is fundamental to our All Roads multi-asset franchise. Since its 2012 inception, no single asset class has consistently driven portfolio returns: sometimes duration leads; at others, equities and alternative risk premia do. Yet by maintaining market exposures to diverse risk premia and adjusting these levels in response to the environment, we aim to harness the next effective diversifier as it comes to the fore. Combined with our ability to reduce market exposure in favour of cash at times of stress, this helps us weather volatility in pursuit of a smooth return profile.