Today’s CTR
China technology has spent the past 24 hours oscillating between triumph and vertigo. ChangXin Memory Technologies’ spectacular market debut offered Beijing a potent symbol of semiconductor self-reliance, while the broader chip sell-off reminded investors that artificial intelligence enthusiasm still rests on highly cyclical hardware economics. Tencent is assembling an increasingly formidable artificial intelligence portfolio, Chinese model developers are exploring ways to charge more for their intellectual property, and BYD is attempting the automotive equivalent of winning an away match in Tokyo. The prevailing mood is confident, well-funded and strategically ambitious — but no longer immune to awkward questions about valuations, profitability and execution.
CXMT’s debut turns China’s chip ambitions into a market spectacle
ChangXin Memory Technologies (CXMT), China’s largest producer of dynamic random-access memory (DRAM), rose 466 per cent on its first day of trading in Shanghai. The company raised about US$8.5 billion in the mainland’s largest initial public offering (IPO) since 2010 and briefly became China’s most valuable listed company.
The debut gives Beijing’s semiconductor campaign something it has often lacked: a highly visible capital-markets success. CXMT remains behind Samsung, SK Hynix and Micron in technology and global market share, but its listing provides substantial funding for production expansion and high-bandwidth memory development.
The valuation also contains a warning label. China’s IPO pricing system, a limited free float and intense demand for strategic technology stocks helped amplify the first-day rise. A share price can leap several generations faster than a fabrication process.
The closing thought: CXMT has secured capital and political prestige; it must now prove that both can be converted into competitive memory chips.
Hefei emerges as the unlikely venture capitalist behind the chip boom
Hefei’s municipal investment funds are sitting on a paper gain of roughly 5,000 per cent from their early support for CXMT. Government-linked entities reportedly retain about 30 per cent of the chipmaker, a stake now valued at close to Rmb1 trillion.
The windfall strengthens the case for Hefei’s distinctive industrial policy. Rather than relying solely on subsidies, the city has repeatedly taken equity stakes in capital-intensive technology companies, including display maker BOE and electric-vehicle group Nio.
The reach extends beyond one fortunate investment. Other Chinese local governments are likely to study Hefei’s model as they compete to attract semiconductor, robotics and artificial intelligence companies. The danger is that copying the financing structure is easier than reproducing the city’s technical judgement and willingness to tolerate years of losses.
The closing thought: Hefei has made state capitalism look rather like venture capital — except the limited partners are taxpayers.
Tencent quietly builds an artificial intelligence conglomerate
Tencent’s expanding artificial intelligence (AI) portfolio now stretches across model developers, semiconductor designers and computing infrastructure. Its interests include prominent companies such as Moonshot AI and DeepSeek, alongside a broader collection of investments intended to give the group exposure to multiple layers of the AI supply chain.
The strategy resembles Tencent’s earlier approach to China’s consumer internet: place numerous bets, supply traffic or infrastructure where useful, and allow the strongest companies to emerge. That is less theatrical than developing every breakthrough internally, but potentially more resilient.
Tencent also has an advantage that many model start-ups lack: distribution. WeChat, cloud services, gaming and enterprise software provide ready-made channels through which AI products can be tested and commercialised. The difficult task will be coordinating those assets without smothering entrepreneurial teams beneath corporate process.
The closing thought: while rivals argue about the best model, Tencent is trying to own a portion of the entire tournament.
Chinese model developers consider putting a price on open weights
Chinese artificial intelligence companies may increasingly adopt “paid-weight” licensing, according to Goldman Sachs. Under such arrangements, model parameters could remain available to outside cloud providers and developers, but commercial users would pay for broader deployment rights.
The shift would mark an important change in China’s AI economics. Permissive open-weight releases have helped companies such as Moonshot AI and Zhipu AI attract users, developers and international attention. They have been less effective at turning that attention into recurring revenue.
A paid model could improve revenue capture without requiring Chinese developers to imitate the fully closed systems favoured by several American competitors. It could also fragment the ecosystem, however, as developers confront different licences, hosting restrictions and usage fees.
The closing thought: open weights helped China’s models travel quickly; the next question is whether they can travel profitably.
BYD drives directly into Japan’s most defended car segment
BYD launched the Racco, an electric vehicle designed specifically for Japan’s miniature “kei” car category. The entry-level model is priced below ¥2 million after subsidies and offers an estimated range of about 210 kilometres, with a longer-range version also planned.
Kei cars account for roughly 40 per cent of Japanese new-vehicle sales, making the category central to the domestic manufacturers’ hold on their home market. BYD is therefore not merely exporting another Chinese model; it has developed a product around Japanese dimensions, regulations and consumer habits.
The commercial hurdle remains formidable. Japanese buyers overwhelmingly favour domestic brands, while BYD sold only a few thousand cars in the country last year. Yet even modest success would show that Chinese manufacturers can compete through localised product design rather than price alone.
The closing thought: challenging Toyota globally is ambitious; challenging Japan’s tiny-car establishment on its own narrow roads may be bolder still.
The artificial intelligence chip trade meets its cyclical alter ego
A broad sell-off in semiconductor shares deepened across Asia, with South Korea’s Kospi falling more than 10 per cent and major memory producers suffering sharp declines. Investors are questioning whether the extraordinary spending on artificial intelligence infrastructure can continue generating adequate returns.
China sits on both sides of the argument. Advances by companies such as Moonshot AI demonstrate that strong models may be developed with fewer resources than previously assumed, challenging the assumption that computing expenditure must rise indefinitely. At the same time, CXMT and other Chinese manufacturers are preparing substantial new capacity.
That combination could accelerate innovation while adding to longer-term supply pressure. Memory chips remain a notoriously cyclical business, regardless of how many artificial intelligence labels are attached to the servers consuming them.
The closing thought: the AI revolution may be unprecedented, but semiconductor investors have seen the capacity cycle before.
