In this issue:

• The US-China tech face-off that cannot be ignored
• Uncle Sam (Altman) might want your money soon

In Q3, exchange-traded funds (ETFs) tracking Chinese stocks dominated the Singapore Exchange’s (SGX) 10 top-performing ETFs list so thoroughly, that only two Vietnam-focused funds managed to break into the league. 

In fact, ETFs focused on Shanghai’s Star Market or the ChiNext board – the Chinese equivalents of the Nasdaq – were the top three best-performing funds, returning between 42.9 and 67.5 per cent for the quarter. 

Despite these outsized numbers in the Chinese tech space, Singapore’s investors appear to be looking elsewhere.

None of the top-performing Chinese stock ETFs were among SGX’s list of top traded ETFs by net inflows during Q3, or during the first half of the year, for that matter. 

Instead, investors busied themselves with safe-haven proxies or income plays, such as the SPDR Gold Shares ETF, the Amova SGD Investment Grade Corporate Bond Index ETF and the Lion-Phillip S-REIT ETF, among others. 

The reticence about the Chinese tech space is understandable. The Magnificent 7 stocks are sucking up a lot of oxygen in the market, and any leftover oxygen is going towards gold and cryptocurrency – ironically, to hedge out the Mag 7’s frothiness and US-dollar exposure. 

Investors might also be holding on to some Chinese stock PTSD circa 2022 when Beijing cut Alibaba and Jack Ma down to size. 

Q3’s top ETF performer, the CSOP CSI Star and ChiNext 50 Index ETF, might be up about 60 per cent year-to-date, but it tested investors’ mettle with double-digit negative returns in 2022 and 2023 before recovering the following year. 

The AI streak running through Chinese tech ETFs

The top three ETFs are partially proxies for how Chinese electric vehicle battery giant, Contemporary Amperex Technology Co Limited (CATL), is doing. The firm, up some 46 per cent year-to-date, is the largest constituent in all three ETFs, with weightages ranging from 10 to 20 per cent of overall fund allocation. CATL’s H1 revenue and profit are up 7.3 per cent and 33 per cent, respectively, despite trade tensions and a growing pressure on margins. 

More interesting, though, are CATL’s fellow ETF constituents, many of which are part of China’s artificial intelligence (AI) chip and data centre boom. 

More than half of the top 10 holdings in the CSOP CSI Star and ChiNext 50 Index ETF are benefiting from, or driving, this boom one way or another, including Eoptolink Technology, Zhongji Innolight, Semiconductor Manufacturing International Corporation and Cambricon Technologies.

Duelling models

Year-to-date, these ETFs have held their own against the AI darlings of the US market, such as Nvidia (up 46 per cent) and Microsoft (up 24 per cent). 

Where, then, is the Great US-China Tech Face-off headed? The diverging paths of both countries’ large language models (LLMs) offer some clues.

While American AI firms are largely focused on closed or proprietary LLMs, Chinese models such as DeepSeek and Alibaba’s Qwen are taking the open-weight route, in which the model’s parameters are publicly shared and modifiable. 

“Neither approach is superior to the other. However, DeepSeek’s approach is clearly more likely to generate commercially viable use cases for AI technology more quickly than the US hyperscalers, whose AI business model has shifted from asset-light to asset-heavy. This puts pressure on them to monetise their much heavier AI capital expenditure,” Kelvin Tay, managing director and chief investment officer, South Asia-Pacific, UBS Global Wealth Management, wrote in BT last month. 

Given their customisability and cost-effectiveness, open-source models have seen rapid adoption by developers and scrappy startups. “I’d say 80 per cent chance (they are) using a Chinese open-source model,” a partner at venture capital firm Andreessen Horowitz told The Economist. 

Chinese AI models are also becoming more efficient, either running faster or using fewer graphics processing units to do the same amount of work. In September, DeepSeek said that it’d spent only US$294,000 on training its R1 AI model. 

Meanwhile, many of the Magnificent 7 firms keep shovelling ever-larger piles of money into AI’s gaping maw. Alphabet, Meta and Microsoft collectively racked up US$78 billion in capital expenditures last quarter – an 89 per cent increase from a year ago. 

You might not believe in Chinese AI, but Chinese AI believes in you

Now, the US has finally allowed Nvidia to export its H20 chips to China, but Beijing doesn’t want them. The Chinese government is reportedly urging Chinese tech giants to use domestic suppliers such as Huawei Technologies and Cambricon instead. 

“They’ve made it very clear that they don’t want Nvidia to be there right now,” its CEO Jensen Huang said last Tuesday.

Even as Nvidia blitzes past a US$5 trillion market cap milestone, the US’ commercial cold war with China is costing the firm dearly. Being kept out of the Chinese market will be a US$50 billion lost opportunity for Nvidia in 2026, Huang said earlier this year. 

If China’s AI firms continue to chug along separately from (and eventually, ahead of) their American rivals, investors will have to take a closer look at last quarter’s biggest ETF winners.

OpenAI might be headed for public markets as soon as H2 next year, with a valuation of up to US$1 trillion, word has it. The AI giant behind ChatGPT is apparently looking to raise US$60 billion at the low end of the range. 

The market chatter follows news of a major restructuring that will give Microsoft a 27 per cent stake in OpenAI and pave the way for the company to become a for-profit business. 

If the public offering comes to pass, OpenAI could do with the money. Just last year, the company was estimated to have lost US$5 billion, on the back of US$3.7 billion in revenue. OpenAI’s cash burn has reportedly continued well into 2025, with an operating loss of US$7.8 billion in H1 alone.

Imagine: this time next year, you too could be part of OpenAI’s great money bonfire. 

(Disclosure: I have shares in Microsoft and Nvidia.)