Urgent.News

What's breaking now, across hundreds of outlets.

AI

故宮院長怒斥光電AI圖誤導! 蔡詩萍臉書悄悄撤圖 網酸:沒錯幹嘛刪?

台北市文化局長蔡詩萍日前在臉書發文宣稱擋下國立故宮博物院屋頂設置太陽光電案,並附上故宮正館屋頂鋪滿光電板的AI生成示意圖,遭故宮院長蕭宗煌痛批顛倒黑白、惡意曲解公文並要求撤圖道歉...…

The New York Times reports that China's AI ambitions are facing financial challenges, as it struggles to keep up with the United States in the development of artificial intelligence. Both countries see AI as a key area for growth, with US-based Anthropic recently securing over $800 billion in funding through its IPO. However, China's venture capital industry lags behind the US, with several major technology firms like Alibaba, Baidu, and Tencent investing in some of the leading Chinese AI companies.

Yet, many of these firms are now struggling with negative cash flow. Investor interest in AI has also waned, with China's China Creation Board 50 index falling 43% since the end of July, surpassing the decline in US tech stocks. Rhodium Group predicts that China's largest tech companies' capital expenditure will double this year to $1.39 trillion, while AI lab revenue accounts for less than 8% of that.

Research from Brookings Institution suggests China's AI labs are running on a treadmill, where speed increases but difficulty grows. A Brookings analyst, John Thornton, commented on the challenges China faces in competing with the US in AI. Additionally, US restrictions on exporting advanced semiconductors to China have forced Chinese AI labs to find expensive alternatives, such as renting data centers equipped with Nvidia chips in Southeast Asia.

However, the financial constraints may be an even bigger hurdle, as China's revenue-to-capital expenditure ratio is worse than that of the US. Epoch AI data shows that China's six largest AI companies generate about 10% of the revenue of OpenAI and Anthropic combined. China's embrace of open-source AI has exacerbated this imbalance, as it allows developers to freely use and modify models, accelerating their global spread but weakening revenue potential.

The shortage of advanced Nvidia chips, which are considered the gold standard for training powerful AI models, is another obstacle. In 2022-2026, US companies are projected to have computational power equivalent to about 20 million Nvidia H100 chips, while China will have around 26 million, many of which may have been acquired illegally.

On the second-hand market, Nvidia's DGX B300 supercomputing servers are priced at nearly $3 million, about eight times their US price. Demand for these servers has doubled since early this year, reflecting the high costs Chinese companies face to acquire cutting-edge computing power. The situation could worsen, as Nvidia CEO Huang Renxun recently resigned amid a sexual misconduct investigation, having previously championed business in China.

Nvidia has announced it will cease exporting chips to China and has established an export control committee.

Rewritten by urgent.news; may contain errors — check the original.

Sources: talk.ltn.com.tw · rti.org.tw · hani.co.kr · ec.ltn.com.tw · news.ltn.com.tw

This story

This is one outlet's version. Read the fullest account.

Every outlet that reported this, and when →

Read the original at news.ltn.com.tw →

Report a problem with this story

More in AI

More from Sunday 22 November →