Made in Ch-AI-na

Jensen Huang recently described China as being “nanoseconds behind” the US in its AI capabilities. Which invites the question: what does it mean to be globally competitive in AI?

In this piece we explore the developments across the full AI value chain in China, from its inputs (power, compute, data), through to its commercial application and the investment implications.

Electricity Generation and Transmission

AI is highly energy intensive, presenting a challenge for developed markets that have historically underinvested in power infrastructure (generation and transmission). By comparison, China has proactively invested into both renewable energy generation, energy storage, and ultra-high voltage transmission (UHVT) lines over the last decade.

In 2024, China added 373GW of domestic renewable capacity (75% of which was from solar), compared to only 58GW of renewable capacity added in the US. In 1H25 alone, China added more additional solar capacity than the entire installed based in the US. China’s manufacturing dominance across these value chains is critical in this rollout. China manufactures over 80% of the world’s solar panels and batteries, providing a cost advantage that allows them to add “generation + storage” capacity at a highly competitive levelized cost of energy (LCOE).

China has also invested heavily in ultra-high voltage transmission (UHVT) infrastructure to transport electricity from areas of production (mostly Northwest China) to areas of heavy consumption (mostly East and South). Over the last 5 years China has built 8200 miles of UHVT compared to 375 miles in the US. Over the next 6 years China plans to spend a further $800bn on transmission.

Semi-conductor Manufacturing

China’s chip manufacturing capability currently lags behind global leaders. This includes equipment, logic chips and memory.

SMIC, China’s leading foundry, has achieved production of 7nm chips, but it remains several generations behind industry leader TSMC, which first introduced 7nm chips in 2018. TSMC is expected to release its 2nm node production this year. Chips below 7nm typically require EUV lithography equipment. Following US restrictions, China is unable to purchase those from ASML, and does not yet have the capability to produce them domestically.

In memory, Chinese firms such as YMTC and CXMT can produce standard NAND and DRAM chips. But they are not capable of producing leading edge high-bandwidth memory (HBM) for AI applications, where SK Hynix and Micron currently dominate globally.

Despite these gaps, China is moving quickly to catch up. US export controls have made technological self-sufficiency a strategic imperative. This is prompting Beijing to deploy its usual playbook: mobilising the enormous domestic demand base, mandating local sourcing, and providing substantial state subsidises. Recent policy ‘guidance’ requires data centres to source 50% of their chips domestically, while also introducing controls on chips purchased from Nvidia to force a reduction in dependency. This also forces LLMs developers to optimise models for local capabilities/deficiencies (i.e. Deepseek).

Data and Large Language Models

A key differentiator in developing successful AI capabilities is the breadth and quality of data on which models can be trained. In 2020, the Chinese State Council designated “data” as a factor of production (alongside Land, Labour, Capital and Technology). This move reflects Beijing’s view of data as a strategic, centrally coordinated resource to fuel economic growth and industrial upgrading.

While this raises significant concerns about personal privacy and state surveillance, it also potentially creates an advantage for China in training large-scale AI models. As an analogy, one might view China as having a more centrally coordinated “data lake,” whereas the West has more fragmented corporate “walled gardens.”

China’s best-know LLM is Deepseek. However, there are a range of leading models including Qwen (by Alibaba), Baichuan, Ernie (by Baidu) and Hunyuan (by Tencent). These all appear to perform well in international benchmarks. Interestingly, these are all open source models and are increasingly designed to run on domestically produced chips. That creates a fiercely competitive landscape for model development in China, as breakthroughs can be quickly replicated by peers, accelerating the rate of collective improvement and therefore the system-wide pace of AI roll-out.

Commercial Roll Out

China is adopting a “good enough” philosophy in AI roll-out, and is prioritising the speed of adoption, rather than perfection in performance and governance. A recent AI paper from the Chinese State Council gave a target of having 70% of people in China using AI services by 2030. The same paper called for a “trial-and-error and mistake-tolerant governing system”.

Although domestic chips are not comparable to the performance of leading Nvidia GPUs, they are sufficient for commercial uses such as chatbots and recommendation engines. We have yet to read evidence that hardware is a constraint to Chinese companies in rolling out AI applications in their businesses. Rather, we see evidence that Chinese companies are rolling out AI at pace, with tangible benefits being achieved.

Investment Implications – Tencent

We think that Tencent is a prime beneficiary of the rapid roll-out of AI services in China.

The company’s WeChat super-app has over a billion active users. The app accumulates data on these users through its various functionalities, including social media, payments, mini-programs, entertainment, and search. This offers an unparalleled data source from which to train its in-house LLM, Hunyuan.

WeChat is also a powerful ecosystem in which to monetise AI capabilities. The super-app allows Tencent to control the entire online consumer journey. That starts at product “discovery”, where Tencent can influence produce awareness in a targeted way through its advertising across the platform’s real estate. Tencent can also influence the bottom of the consumer funnel, known as “transaction conversion”. Consumers can click directly through to e-commerce mini-apps within WeChat, and complete the transaction seamlessly using WePay.

We feel that Tencent’s huge proprietary dataset and a captive user base, combined with an end-to-end consumer ecosystem, creates a powerful environment in which to deploy AI. We have already had data points to support this. Improvements in ad targeting have improved click-through rates up to 3x. That gives Tencent more pricing power with advertisers, and has allowed for an acceleration in ad revenue growth to +20%yoy. Improving personalisation of recommendations is also driving user growth and “time-spent” in its entertainment businesses, such as Tencent Video and Tencent Music.

We also view it as positive that the capital investments being made by Tencent into data centres are primarily for its in-house use. That limits capital intensity, and allows the company to invest only where there is a clear line-of-sight on ROIs and future capacity requirements.