※ Global Morning Briefing summarizes global news delivered by Seoul Economic Daily.
Trump Says Opposing Data Centers Means Choosing Poverty
Data centers, whose numbers have surged on demand driven by artificial intelligence, have emerged as a central issue in the U.S. midterm elections now just over 60 days away. As resentment spreads over the argument that massive electricity consumption drives up power bills and adds to living costs, skepticism about data centers is surfacing not only among Democrats but also within the governing Republican Party. Support is also substantial, however, led by labor unions arguing that data centers will create jobs, and AI companies have launched their own public campaigns.
According to major foreign media reports on the 31st of last month, states across the U.S. are rolling out regulatory measures on data centers one after another. Democratic governors have been particularly active. New Jersey Governor Mikie Sherrill signed legislation requiring data centers in the state to report water and electricity usage statistics every six months, and Chicago Mayor Brandon Johnson signed an executive order temporarily halting data center construction. Pennsylvania Governor Josh Shapiro also decided to require compliance with related conditions, and in July, New York Governor Kathy Hochul signed an executive order imposing a one-year moratorium on large data center construction. Fifteen of the 50 U.S. states are currently weighing restrictions or bans on data center construction.
Voter resentment is also spreading quickly. In a survey by the Annenberg Public Policy Center at the University of Pennsylvania, the share of respondents opposing new data centers in their communities rose to 61% in July from 49% in March. In a national Emerson College poll, opposition jumped to 63% in July this year from 42% in December last year. In more than 10 U.S. cities, recall votes are being pushed against local officials who backed data center construction. Amid this mood, even Texas Governor Greg Abbott, in a Republican stronghold, suspended all data center power grid requests and launched an audit.
AI companies have gone on the counteroffensive. The nonprofit group Build American AI plans to mobilize $50 million (about 68.68 billion won) to launch pro-data center campaigns in Kansas, Ohio and Wisconsin, and is also planning activities to support friendly candidates through a separate super PAC called Building the Future. Large unions, including the International Brotherhood of Electrical Workers with more than 900,000 members, are defending data center construction on the grounds of job creation.
President Donald Trump pushed back directly against opposition on the social media platform Truth Social, saying the only reason American communities do not want data centers is that they want to be backward and poor. He also framed the opposition as serving Chinese interests, saying China would be pleased, while rolling out price measures as well, including agreements with pharmaceutical companies to lower drug prices and a temporary tariff exemption for ground beef products.
Taiwan Trails Korea and Japan in AI Infrastructure
In Taiwan, a semiconductor manufacturing powerhouse with an AI hardware supply chain, voices of self-criticism are emerging that the island's own AI infrastructure buildout is failing to gain speed. Critics point in particular to how far Taiwan lags neighboring Korea and Japan in both the pace and the scale of investment.
According to Taiwan's United Daily News on the 1st, the 2026 Tech Sovereignty Summit Forum hosted by the Chinese Institute of Communications the previous day heard criticism that Taiwan constantly calls for the development of sovereign AI while its computing capacity and power base lag behind.
Huang Yen-nun, former minister of Taiwan's Ministry of Digital Affairs, cited this year's Global AI Index published in the U.K. and said Taiwan ranked only 16th, far behind not only the first- and second-place U.S. and China but also Singapore at third, Korea at fifth and Japan at 11th. Pointing to shortages of computing power and electricity as the causes of Taiwan's weak AI competitiveness, the former minister said Korea has already purchased 260,000 graphics processing units from Nvidia and Japan currently holds more than 40,000 GPUs, while Taiwan has only 30,000.
One reason Taiwan's manufacturing capability is not translating into computing power is that hyperscalers, or operators of very large data centers, are locking up large volumes of AI chips in advance. Nvidia is reported to have already secured more than half of the output from TSMC's advanced CoWoS packaging process through 2027.
Power shortages are also a chronic weakness of Taiwanese industry. The former minister explained that running 30,000 GPUs alone requires more than 1 gigawatt of electricity, equivalent to the output of one nuclear power plant. Taiwan pursued a nuclear phase-out after the 2011 Fukushima accident and met its target in May last year, but supply has since been unstable because of factors including the war in Iran, even after liquefied natural gas generation grew to half of total output. Taiwan depends on imports for 100% of its LNG.
Amid these obstacles, Taiwan's per-capita AI usage intensity actually fell to 2.08 this year from 2.29 last year. That contrasts with gains over the same period in Singapore, to 5.81 from 4.57, in Korea, to 3.78 from 3.73, and in Japan, to 1.91 from 1.86. Huang Chih-fang, chairman of the Taiwan External Trade Development Council, mocked the situation as being a giant in hardware and a dwarf in applications.
Taiwan's government shares that concern. Lin Yi-ching, minister of Taiwan's Ministry of Digital Affairs, said at the event that work is under way to build a corpus for training Taiwan's sovereign AI, and that the government plans to secure more than 10,000 GPUs in Taiwan within a year by attracting private investment.
Anthropic Signs $35 Billion Cloud Deal With Nvidia-Backed Lambda
Anthropic, developer of the AI model Claude, has secured large-scale data center computing resources from a cloud company backed by Nvidia.
The Wall Street Journal reported on the 31st of last month, citing sources, that Anthropic signed a cloud computing contract worth $35 billion (about 47.9 trillion won) with Lambda, an AI cloud company backed by Nvidia. Through the deal, Anthropic will gain long-term access to AI computing capacity at a data center in New Braunfels County, Texas, that Lambda will operate, the newspaper said.
Infrastructure developer Hut 8 is building the data center, and Nvidia has contracted with Hut 8 to secure leasing rights to the facility's capacity. Lambda will install Nvidia GPUs there and supply Anthropic with the cloud computing needed to train and run AI models, with the facility reaching up to 700 megawatts in scale. In other words, it is a multi-layered structure in which Hut 8 builds the data center, Nvidia provides the leasing rights and the GPUs, Lambda operates the GPU cloud, and Anthropic buys the resulting computing resources.
The arrangement is seen as aimed at smoothing the financing needed for large data center construction. Rather than having Anthropic, whose credit rating has not been disclosed, sign a long-term infrastructure contract worth tens of billions of dollars on its own, Nvidia's holding of the leasing rights lowers the credit risk borne by Hut 8 and financial institutions.
Anthropic, which is preparing to list by year-end, relies heavily on Nvidia for AI model development. In early August, it signed a $45 billion (about 74 trillion won) contract with Nscale, another cloud company backed by Nvidia, to lease Nvidia computing capacity in West Virginia.
Nvidia is supporting OpenAI, the developer of ChatGPT, in a similar fashion. On the 17th of last month, OpenAI signed a contract with SB Energy and Nvidia to secure 8 gigawatts of infrastructure at the Potts Pike data center campus in Pike County, Ohio.
China's CXMT Starts Small-Scale HBM3E Output, Trails Big Three by 3 to 5 Years
ChangXin Memory Technologies, China's largest memory chipmaker, has already begun producing HBM3E, an advanced high-bandwidth memory product, on a small scale, the technology publication The Information reported on the 31st of last month, citing multiple sources.
CXMT plans to ramp up output in earnest from next year, according to the report, narrowing its technology gap to a single generation behind products already in mass production at the memory Big Three of Samsung Electronics (005930), SK hynix (000660) and Micron.
According to the sources, T-Head, a chip design unit under Alibaba, and Cambricon Technologies are testing CXMT's HBM3E in their in-house processors, and related products could reach the market early next year if the plans proceed as scheduled. These companies are expected to be able to sell their own chips containing CXMT's HBM3E to AI data center operators in China.
The Information said orders from these fabless companies will provide a foothold for CXMT to accumulate HBM mass production know-how and customer feedback, and that this could mark a turning point for China's semiconductor industry as it coincides with a surge in HBM demand driven by the AI boom. CXMT plans to invest the $8.6 billion (about 11.8 trillion won) raised through its listing on the Shanghai exchange last month in expanding production.
Obstacles remain, however. The U.S. has restricted exports of advanced HBM to China since 2024, and Chinese companies must obtain prior approval from the U.S. government even when buying lower-specification memory. CXMT faces limits on purchases of advanced foreign equipment as well as low yields, and its technology gap with leading companies is put at three to five years. SK hynix, Samsung Electronics and Micron have already begun supplying samples of HBM4E, a full generation ahead.
OpenAI's ChatGPT Ad Revenue Tops $1 Billion Ahead of Listing
Annualized revenue from advertising on OpenAI's ChatGPT chatbot has passed $1 billion (about 1.4 trillion won).
OpenAI said on the 31st of last month that ChatGPT advertising revenue, currently generated in about 40 countries, had reached $1 billion. That came roughly 200 days after the company began showing ads to users in the U.S. on Feb. 9.
OpenAI shows ads to free ChatGPT users and subscribers to its lower-priced Go plan, and started the advertising service in Korea in June this year. ChatGPT's advertising revenue accounts for 2.5% of OpenAI's total annualized revenue of $40 billion, disclosed recently.
OpenAI said tens of thousands of advertisers are currently using its advertising platform, and that advertising forms one pillar of a diversified business model alongside consumer subscriptions, enterprise services and its usage-based application programming interface.
OpenAI's disclosure of advertising results is seen as a move to highlight its performance to investors as it pursues an initial public offering. With rival Anthropic recently overtaking it in valuation and its listing pushed back from this year to early next year, the company is read as emphasizing advertising revenue to showcase a diversified business base.
