This year, the new stock market in Hong Kong has made a strong comeback, with a total annual fundraising of about 280 billion HKD, and is set to return to the top spot in new stocks globally. However, at the same time, the basic underwriting rate of Chinese issuers continuing to rely less on international investment banks this year, the basic underwriting rate of investment banks has hit a 25-year low, an average of only 1.46%. The industry believes that under the continued fierce competition, the future rate levels will hardly show significant improvement. The basic fee rate is more than 20% less than the peak According to LSEG statistics, as of December 15, this year, the average underwriting rate for local IPOs and second listings in 2025 (only basic fees, not discretionary incentives)
Amazon rose nearly 3% in the night session. According to The Information, OpenAI is in negotiations with Amazon to raise at least $10 billion from Amazon and use Amazon's artificial intelligence chips
There are about a hundred new cloud merchants operating like CoreWeave around the world, borrowing funds from banks, bond markets, private equity funds, or issuing stocks and equity bonds through the stock market, then buying GPUs from Fedda to install their hardware in the cabinets and leasing their hardware to AI companies, including OpenAI and Meta, which seem worry-free. However, there are some flaws in their business model. One is that the added value of the business is not significant, and the cost is very high. As seen from a report released by McKinsey last month, the evolution and next step of new cloud services, 36% of their revenue is gross sales costs, because they do not own the power and hydropower systems of their factories and data centers, they need to rent from the owners
Temu, a subsidiary of Pinduoduo, has reportedly been subjected to a surprise inspection by the European Union, fearing that it may receive subsidies from the Chinese government. Last week, the EU's competition regulator raided Temu's European headquarters in Dublin, Ireland, to investigate whether it had received unfair subsidies, according to foreign media reports. Suspected of receiving unfair subsidies from China The Commission said it had conducted an unannounced on-site inspection of a company actively engaged in e-commerce within the EU under the Foreign Subsidies Regulation, but did not disclose the name of the company involved. The Foreign Subsidies Regulation aims to combat unfair competition caused by non-EU companies receiving government subsidies, and companies that violate the
According to sources cited by Reuters, Nvidia has developed location-validation technology that can show which country its chips are operating in, helping to prevent their artificial intelligence (AI) chips from being smuggled into countries that are prohibited from exporting. The news indicates that this new feature of Nvidia utilizes the confidential computing power of its image processor (GPU). The company has demonstrated this feature privately in recent months, but has not publicly released it, allowing customers to choose to install it when updating the software. Utilize GPU monitoring. Not publicly released The report points out that the software's purpose is to enable customers to track the overall computing performance of the chips, which is a common practice for businesses when p
Latest on DeepSeek's Alleged Use of NVIDIA Blackwell GPUs As of December 11, 2025, the big story buzzing in AI and tech circles is a report alleging that Chinese AI startup DeepSeek has been accessing thousands of NVIDIA's Blackwell GPUs—despite U.S. export bans—through an elaborate smuggling operation. This comes amid escalating U.S.-China tech tensions, where advanced chips like Blackwell are restricted to prevent military or strategic advantages for China. Here's a breakdown based on recent reports and discussions, drawing from U.S. media, NVIDIA's statements, and online chatter. The Allegations - Core Claim: According to a December 10 report by The Information, DeepSeek obtained around 2,000 Blackwell chips to train its upcoming large language model (LLM). The scheme reportedly involve