China’s chip gap

As with any investment, your capital is at risk.

In the race for AI leadership, China appears to be neck-and-neck with the US in software. Its open-source models routinely outperform American peers on popular benchmarks, often at less than 10 per cent of US pricing.

However, the hardware gap remains wide. Even Huawei’s most advanced chip still trails Nvidia’s H200, an older generation that President Trump recently authorised for sale to China. Nvidia’s latest chip, Blackwell, is roughly four times as powerful as the older H200, and both AMD and Google’s newest chips are more than twice as powerful. On a per-chip performance basis, American chipmakers remain well ahead.

Can China close this gap? How quickly, and with what implications for the rest of the world?

Progress and bottlenecks

In chip design, China is shifting from an Nvidia-centric stack to a hybrid ecosystem anchored by Huawei. It’s trying to narrow the per-chip performance gap by stitching thousands of chips into large clusters to match system-level performance. The trade-off is lower efficiency and reliability.

Access to the H200 will provide short-term relief, but Beijing is still likely to reserve a meaningful share of demand for national champions, especially in state-owned datacentres.

In chip manufacturing, China’s leading foundry, SMIC, has made notable progress after years of stagnation. It has achieved workable 7 nanometre (nm) yields, but remains far behind TSMC, which is entering 2nm. SMIC is likely to ramp up 7nm production aggressively in the near future.

While one can do plenty with 7nm – not least autonomous driving chips and graphics processing units (GPUs) – more advanced nodes are unlikely to be scalable without ASML’s extreme ultraviolet (EUV) lithography machines. SMIC’s current deep ultraviolet (DUV) approach already implies materially higher costs and operational complexity versus TSMC’s EUV-based process at comparable nodes.

EUV denial also affects memory, though the impact is smaller than for logic chips. CXMT, China’s primary DRAM (short-term working memory) producer, is estimated to be one to two generations behind Samsung Electronics and SK Hynix. Its early high-bandwidth memory effort (essential for frontier-scale AI training) also faces lithography machine licensing and advanced packaging constraints.  

In summary, China’s semiconductor roadmap resembles its own version of ‘Moore’s Law’: it is likely five-plus years behind Taiwan and Korea on leading-edge nodes and still far behind Nvidia on single-chip performance. But with existing DUV capacity, architectural workarounds, cheap energy, and policy effectively forcing volume towards local champions, its solutions could be ‘good enough’ to support a domestic AI stack.

Where does this leave us?

Chinese foundries remain unattractive investments on current economics, as state priority has allowed less focus on profitability. That said, they are likely to keep taking domestic share in nodes that do not require EUV, and could become viable, cost-efficient competitors to the China businesses of our holdings TSMC, Samsung Electronics, and SK Hynix. However, all three are shifting towards more leading-edge production. Their revenue exposure to nodes where China can realistically compete is in the single digits to low teens: immaterial in our view.

Other local suppliers will also try to replace their foreign peers wherever possible along the domestic value chain. Local players are estimated to capture about 60 per cent of China’s planned AI-chip spending, a sharp increase from just a few years ago, when US firms accounted for over 80 per cent of sales. This will not be a straight line: it is simply too large a market for international competitors to abandon. Selection therefore matters.

Some of our emerging market strategies recently added Montage, a memory-interface chip designer for datacentres, and are re-testing the thesis for our holding in Silergy, an analogue chip designer that has detracted in previous quarters.

What’s next?

Most Chinese semiconductor companies we spoke to see no short- to medium-term path to EUV. Over the longer term, however, China could eventually repeat its electric vehicle playbook in semiconductor hardware and disrupt incumbents. The probability looks low today given geopolitics, but we should stay open-minded and track progress closely.

More broadly, China is building a parallel AI ecosystem to the US. From chips to models and applications, its low-cost yet capable solutions are gaining domestic traction and attracting global interest. Yet equity markets still appear to underprice this shift.

As mentioned by Sophie Earnshaw, an investment manager in our China team, in a recent podcast – the total market cap of China’s three largest AI and cloud players – ByteDance, Tencent and Alibaba – is barely comparable to Meta’s. That is likely to change.

For now, our preferred exposure is through platform companies and selected hardware suppliers, but we are also keen to research newer disruptors, including independent AI labs, niche chip designers, and semiconductor equipment manufacturers. Our Shanghai team has flagged a promising IPO pipeline from the sector for 2026.

Risk factors 

The views expressed should not be considered as advice or a recommendation to buy, sell or hold a particular investment. They reflect opinion and should not be taken as statements of fact nor should any reliance be placed on them when making investment decisions.

This communication was produced and approved in February 2026 and has not been updated subsequently. It represents views held at the time of writing and may not reflect current thinking.

Potential for Profit and Loss 

All investment strategies have the potential for profit and loss, your or your clients’ capital may be at risk. Past performance is not a guide to future returns.

This communication contains information on investments which does not constitute independent research. Accordingly, it is not subject to the protections afforded to independent research, but is classified as advertising under Art 68 of the Financial Services Act (‘FinSA’) and Baillie Gifford and its staff may have dealt in the investments concerned.

All information is sourced from Baillie Gifford & Co and is current unless otherwise stated. 

The images used in this communication are for illustrative purposes only.

189253 10061288