Japanese tech stocks took a tumble on Thursday as AI infrastructure spending worries on Wall Street crossed the ocean into the Asian markets, with AI-related stocks declining.

Softbank Group Corp was among the top losers in the benchmark Nikkei 225, falling as much as 7.25%, with the index leading losses in Asia, down 1.23%. The group pared some losses and was last trading 3% lower.

This decline comes as the tech-heavy Nasdaq Composite fell 1.81% overnight, dragged by losses in Oracle, Broadcom, Nvidia and other AI plays.

The losses in Oracle came after the Financial Times reported on Wednesday that Blue Owl Capital’s plans to finance the cloud infrastructure company’s $10 billion Michigan data center had stalled. The company last week had refuted a report that said it had delayed some projects for AI major OpenAI to 2028.

Tech-focused SoftBank has seen sharp volatility in its stock over the past month as fears over AI-related spending have gripped the market.

At the start of the year, the group had revealed plans to invest $500 billion in AI infrastructure in the U.S. along with OpenAI, Oracle and other partners, and in September it announced five new U.S. AI data center sites under Stargate, OpenAI’s overarching AI infrastructure platform.

Other Japanese tech stocks also fell. Semiconductor equipment supplier Advantest, dropped as much as 5%. Counterparts Lasertec, Renesas Electronics and Tokyo Electron declined between 3% and 4%.

Jesper Koll, expert director at Tokyo-based financial services firm Monex Group, said much of what goes into data centers, power centers, and AI hardware enablers is “Made in Japan, and can only be made in Japan.” That makes Japanese tech, especially AI-related stocks more vulnerable to any worries around U.S. tech spending.

On Wednesday, Japan’s trade numbers showed that exports of electrical machinery jumped 7.4%, and semiconductor-related exports surged 13% year on year. Koll said the U.S.-led boom in tech spending was translating into growing exports of specialized machinery and equipment.

Losses were less pronounced in South Korean chip heavyweight Samsung Electronics at 0.93%, while SK Hynix reversed course to gain 0.73%. Taiwan’s TSMC, the world’s largest contract chip manufacturer, was marginally down.

The apprehension investors have surrounding Oracle has spilled over from manifesting in its stock price — which has fallen nearly 50% from its all-time high on Sept. 10 — to affecting its projects.

Asset management firm Blue Owl Capital reportedly pulled out from Oracle’s $10 billion data center project over unfavorable debt terms, according to the Financial Times, as concerns about the tech giant’s high level of debt mount.

The latest development adds fuel to worries that Oracle could delay the completion of data centers for OpenAI, which were first flagged by Bloomberg on Friday, though the cloud company has denied the report.

Shares of Oracle fell 5.4% Wednesday, putting its month-to-date losses more than 11%. They weighed down related names, such as Broadcom Nvidia and Advanced Micro Devices.

As a result, major U.S. indexes fell. The S&P 500 retreated 1.16% and the Dow Jones Industrial Average dropped 0.47%, while the Nasdaq Composite lost 1.81% in its worst day in nearly a month.

Despite the recent pullback in artificial intelligence stocks, the Bank of America thinks “the AI trade may still have room to run into 2026” — with the important caveat that shares going up does not mean a bubble isn’t forming.

“In our view, such progression validates our thesis that a larger AI bubble continues to build,” analysts at Bank of America wrote.

The trouble, as always, is pinpointing the exact moment before the bubble pops — if that’s even possible.

— CNBC’s Jaures Yip contributed to this report.

Three holds and a cut? Europe’s central banks are about to make their final calls of 2025

Investors are gearing up for the last interest-rate decisions of 2025, with four of Europe’s central banks announcing their monetary policies and macroeconomic outlooks on Thursday.

The European Central Bank, Bank of England, Riksbank and Norges Bank are all meeting, but only one of them is expected to change its rate.

— Holly Ellyatt and Annette Weisbach

It felt like déjà vu when shares of chipmaker MetaX Integrated Circuits soared 700% in its Shanghai market debut on Wednesday. Moore Threads surged over 400% on its first day of trading just two weeks earlier.

They’re the latest Chinese AI chip companies the country’s investors are ploughing money into, as it races to develop its own semiconductors and challenge Nvidia’s dominance in the face of U.S. export curbs.

Both are developing graphics processing units (GPUs), the type of chip manufactured by Nvidia and used for advanced AI.

Investor enthusiasm around Chinese AI-chip IPOs is partly shaped by longer-term expectations that China will build a self-sufficient semiconductor ecosystem as tensions with the U.S. continue, Macquarie’s equity analyst Eugene Hsiao told CNBC.

Washington has barred sales of Nvidia’s most advanced semiconductors to the country. While U.S. President Donald Trump relaxed export curbs for some Nvidia chips, regulators in the country were planning to limit access to the company’s processors, the Financial Times reported earlier this month, as it looks to wean itself off overseas tech in the AI race. 

None of China’s AI chipmakers — which include a cohort of tech giants like Huawei, Alibaba and Baidu — have been able to develop processors comparable to Nvidia’s most advanced so far.

But while significant barriers remain in overcoming export control restrictions in some areas of its chip supply chain, like equipment, it’s made significant strides in others, such as memory. 

Here’s how the market of China’s AI chip Nvidia rivals is shaping up.

Privately-owned tech giant Huawei develops the Ascend series of chips, with its next-generation model, the 950, to be launched in 2026. Nvidia told CNBC that “competition has undeniably arrived” when the new systems were announced.

While its previous Ascend models have not been considered competitive with Nvidia’s on a chip-by-chip basis, Huawei has been able to build high-performance “clusters” to rival the US chipmaker’s most advanced systems by linking more of its processors.

“This strategy relies on high-speed, potentially optical interconnects to move data quickly across large clusters – a setup that doesn’t require top-end chips and therefore suits China’s current strengths,” Brady Wang, associate director at Counterpoint Research, told CNBC in November.

China’s biggest search platform, Baidu, has increasingly funneled more resources into AI and is a majority shareholder in chip designer Kunlunxin. In November, the company unveiled a five-year roadmap for its Kunlun AI chips, unveiling new processors in 2026 and 2027.

Baidu, which is traded on the Nasdaq, uses a combination of self-developed chips and Nvidia products in its data centers to run its in-house AI models. The company has looked to position itself as a “full-stack” provider, producing chips, servers, data centers, and AI models and applications.

“Kunlunxin has emerged as a leading domestic AI chip developer, focusing on high-performance AI chips for large language model (LLM) training and inference, cloud computing, and telecom and enterprise workloads,” analysts at Deutsche Bank said in a note in November.

JPMorgan said in a November note that it viewed the Kunlun AI chip as one of the “best-positioned” as Chinese hyperscalers increasingly source from local solution providers.

E-commerce giant Alibaba — which is sometimes compared with Amazon as it is one of the biggest cloud providers locally — began developing AI chips in the late 2010s. It was developing a new AI chip in August, CNBC reported, specifically designed for inference rather than training.

Alibaba’s share rose in September after reports that the company had secured a major customer for its AI chips.  

“Improved performance of its self-developed chip” was one of the factors that supported revenue growth in the cloud division at Alibaba, Morningstar analyst Chelsey Tam said in September.

Cambricon, which is developing chips for AI training and inference, posted record profits in the first half of 2025 as revenue surged. The chipmaker, founded in 2016, said revenue rose more than 4,000% year-on-year to 2.88 billion Chinese yuan ($402.7 million) and net profit hit a record 1.04 billion yuan.

“We view Cambricon as the most plausible winner in China’s AI accelerator market, which is still at its early stage when compared to the US market on chip accessibility issue,” Jamie Mills O’Brien, investment director at investment group Aberdeen, told CNBC by email.

“We see multiple roadblocks being digested in next 1-2 years, including fab maturity, client acceptance, and ecosystem formation, which is likely to set Cambricon a ‘good enough’ alternative to Nvidia’s downgraded chips in China.”

Biren Technology, founded in 2019, is designing high-performance GPUs and was approved for an IPO by the Chinese regulator on Tuesday, according to reports from Reuters and the South China Morning Post.