The market is re-pricing the obstacles faced by the infrastructure construction of artificial intelligence ( AI ) in the United States.
On November 20th local time, U.S. President Donald Trump stated at a White House press conference following his participation in the APEC summit that American companies should not build data centers overseas, otherwise they would face "huge taxes" or "severe penalties." However, Trump also acknowledged that this idea has not been fully discussed with his team, and there are currently no official executive orders or specific tax rate details available.
After the results of the U.S. mid-term elections were announced in early November, due to strong opposition to data centers at the local level, the AI sector of the U.S. stock market experienced a sharp correction. In the month up to November 22, the Nasdaq index fell by 13%, with NVIDIA's stock price shrinking by nearly 41%, and AMD falling by more than 16%.
On one hand, the White House is attempting to keep computing infrastructure within its own borders through taxes and punitive measures, while on the other hand, there are growing concerns regarding environmental protection, power supply, and approval processes after the election. What kind of impact will this have on data center construction in the United States and Asia? Casanova ( Carlos Casanova ), a senior economist for UBS ( UBP ) in Asia, told CBN reporters that although geopolitical fragmentation has pushed up global inflation and interest rates, "we are in a technological race, and no party will voluntarily stop investing in data centers at this time; doing so would be very unwise."
He further stated that the bank still maintains a positive attitude towards the AI sector and is currently focusing on entering markets with high profitability prospects, especially those strong companies that can continue to benefit from substantial capital expenditures, such as AI and leading semiconductor enterprises.
The White House's "localization" requirements and political resistance after the election
Policy signals surrounding data centers in the United States are showing clear signs of tension.

At a White House press conference on November 20th, while warning companies not to relocate their data centers overseas, Trump also attempted to address the concerns of local public opinion. He called on local communities not to shut down data centers but to make use of them, demanded that developers provide direct economic benefits to the communities, and reiterated that by building their own power plants, data centers could actually reduce local electricity costs and increase tax revenues. If California or Texas were not welcoming to such projects, they could be directed to energy-rich states like West Virginia.
This continues Trump's consistent policy tone. In February of this year, he proposed a "protection commitment for electricity bill payers" in his State of the Union address ( Ratepayer Protection Pledge ). Since then, he has also repeatedly demanded that tech giants meet their own electricity needs, bear all the costs of power grid upgrades, and in exchange for local acceptance, build their own power plants and pay "dividends" to communities.
However, in the U.S. mid-term elections that concluded on November 3rd, the opposition at the local level turned into tangible political constraints. The Democratic Party not only seized control of both houses of Congress but also won governorships in key states such as Texas, Ohio, Georgia, Nevada, and Iowa. At the same time, due to concerns about high electricity consumption driving up residential electricity prices, as well as issues related to water resources and noise, according to statistics from Ballotpedia which tracked the results of these mid-term elections, voters supported restrictions on data center development in 40 out of 41 local referendums. Exit polls also showed that 56% of voters believed that AI and data centers were important or key factors affecting their voting in Congress.
Research institutions have conducted a quantitative re-evaluation of this political variable. Allianz Trade ( Allianz Trade ) stated in a report replied to Yicai that approximately one-third (33%) of the planned new data center capacity across the United States is in the “risk zone of suspension.” Just Texas (25.9 gigawatts), Georgia (6.0 gigawatts), and Ohio (4.0 gigawatts) account for 33% of the planned new capacity in North America before 2030. The newly elected governors of these three states all supported an immediate or conditional suspension of approvals during their campaigns. Texas has suspended the approval of new licenses since September of this year in anticipation of an audit.
US domestic data centers are slowing down rather than shutting down
Several economists from investment research institutions interviewed by Yicai News said that under the competition between federal and local policies, the pace of data center construction in the United States may slow down, and the cost per watt may increase, but the investment cycle is far from over.
From the perspective of demand-side certainty and capital expenditure, Coulton ( Brian Coulton ), Chief Economist at the global rating agency Fitch Ratings ( Fitch Ratings ), told CBN reporters that this round of investment cycle still has sufficient momentum for the next year or so. As of 2026, investment in the US IT hardware and software sectors has grown by about 15% year-on-year, and it is expected that the growth rate will slow slightly to around 12% to 13% in 2027, while it will still maintain double-digit growth by 2028.
"Although there have been some fluctuations in the valuation of AI-related stocks recently, we believe that there will not be a substantial correction in 2027. We are still in a relatively early stage of the capital expenditure cycle, and it will continue to be an important support for economic growth for at least the next 18 months to two years," said Coulton.
BMI Chief Economist Xie Habu ( Cedric Chehab ) also told Yicai News that the current market concerns mainly focus on three aspects: first, whether open-source large models will weaken the business models of cutting-edge AI laboratories, thereby reducing their dependence on massive computing power; second, the mismatch between the current debt investments made by large technology companies and hyperscale cloud providers and the future commercial returns; third, the impact of regulatory policy changes on the construction cycle.
However, Xie Habu believes that the current fundamental conditions of enterprises are not yet sufficient to cause the market to worry about a substantial contraction in investment: the earnings growth rate of the components of the S&P 500 index remained at a strong level of 26% to 27% in the third quarter, and none of the technology giants in their latest quarterly financial reports signaled any intention to cut capital expenditures. BMI expects that global hyperscalers in cloud computing will spend slightly less than $1 trillion on capital this year, with the potential to increase further to around $1.4 trillion next year, representing a year-on-year increase of nearly 50%. He said, "Unless profitability is severely impacted or the regulatory environment undergoes drastic changes, companies will continue to make sustained investments in power generation, transmission networks, data centers, chips, and the entire technology stack of cutting-edge models."
From the perspective of the specific delivery timeline on the supply side, according to Allianz Trade Report's estimates, out of the anticipated 33.8 gigawatts of capacity to be delivered, about three-quarters are already under construction, and the impact of the election results is limited; the real exposed risks are concentrated in projects that are scheduled to begin construction in 2028 and thereafter. Approximately 30 gigawatts of the planned capacity for 2028 across the United States are located in states where candidates advocated for a suspension of approvals during the election campaign. The licensing suspensions in these states have directly affected about 20% of the ongoing pipeline projects nationwide.
Allianz Research believes that a political shift will not end this construction cycle; rather, it is likely to push some of the production capacity planned for 2028 to later years and to states that are more conducive to the development environment. Current policy obstacles have mainly increased construction costs and prolonged the cycle, without yet significantly reducing the scale of construction. For example, developers are facing new expenses such as the requirement to provide their own power supply, payments for grid upgrades, prepaid mortgages, and the withdrawal of tax incentives. Large-scale cloud providers have the capability to absorb these costs, while smaller developers and projects that have not secured power supplies are the most vulnerable. However, the report also warns that if suspension orders in places like Texas and New York are further strengthened, the original "cost constraints" could evolve into "scale constraints" around 2028, thereby limiting the ultimate scale and speed of AI's computing power expansion.
Impact on Asia: Spillover Opportunities and "Dual Constraints"
How will the policy changes faced by data centers in the United States be transmitted to the Asian market?
Xie Habu analyzes that if stricter regulations in the United States hinder local construction, and yet there remains a demand for computing power infrastructure by businesses, some capital may indeed flow to other regions. Asia already has a wealth of ongoing and planned projects, which are well-equipped to accommodate this demand. Public data shows that Southeast Asia is the region with the fastest growth in data centers globally. Malaysia, particularly Johor, is at the forefront: Malaysia currently has nearly 1 gigawatt of operational data center capacity, with 2 gigawatts under construction and about 3 gigawatts planned. Moreover, the industry is upgrading along the value chain, shifting from simple cloud storage to more profitable computing power and inference centers. Additionally, until the geopolitical situation in the Middle East becomes clearer, Southeast Asia will continue to be an important direction for multinational tech capital allocation.
However, the Asian market is not without its limitations. While it is capable of accommodating potential spillover demand, Southeast Asia faces constraints in terms of physical infrastructure and cross-border compliance. Xie Habu reminds us that electricity has become the primary limiting factor for the expansion of global data centers. Taking Malaysia as an example, data centers currently consume about 70% of the country's newly added power generation capacity, whereas just a few years ago, this proportion was only 20%. Other Southeast Asian economies such as Indonesia and Vietnam have also formulated ambitious plans, but they face even more severe pressures regarding power grid upgrades.
Looking across the entire Asia-Pacific region, other bottlenecks include long waiting times for grid connection, lagging transmission line construction, shortages of land and water resources, as well as delays in approval due to overlapping functions among multiple departments. All these factors have slowed down the conversion of planned production capacity into actual commencement of projects, said Xie Habu.
However, from a more macroscopic perspective of the industrial chain, Asia is still deeply tied to the global capital expenditure cycle. Huang Liyang, a senior economist at Allianz Trade for the Asia-Pacific region, provided a set of data to Yicai News: The manufacturing industry contributes about 25% of the economic growth in the entire Asia-Pacific region, and Malaysia's export of manufactured products accounts for over 60% of the regional growth. If we precisely track the exports of intermediate goods and final products used globally for infrastructure construction, Asia controls more than 65% of the global share. Among the top ten global economies in terms of manufactured product exports, seven are located in Asia.
Huang Liyang stated that Allianz Trade's baseline judgment is that the absolute amount of global AI capital expenditure will not peak until 2028, by which time it is expected to exceed one trillion US dollars. However, the peak growth rate of investment expansion may arrive earlier than that. Influenced by factors such as the mid-term elections in the United States, corporate investment pace next year may slow down compared to this year, which could lead to a return of Asian-related export growth rates from their exceptional levels this year to more normal levels. "The entire Asia-Pacific region is still expected to maintain an economic growth rate of over 4% in 2027 and 2028. The AI cycle is one of 'slowing down' rather than 'recession'; it will continue to support Asian exports and provide a foundation for regional economic growth," said Huang Liyang.
Author: Gao Ya











