Abstract:
With the latest changes in the U.S. political arena causing market concerns about the prospects of the artificial intelligence industry, some investors have begun to worry that data center construction may encounter large-scale restrictions, which will affect the future development of the AI industry. However, the latest public information shows that even if some state governments are more cautious about data center expansion, most construction plans for new computing power in the United States in the next few years are still expected to continue.
Recently, the U.S. capital market has become significantly concerned about the prospects of AI infrastructure construction. Some analysts believe that after the Democratic Party gains political advantages in some key areas, broader data center construction restrictions may appear in the United States in the future. However, judging from the current actual construction progress, a large number of projects have entered the advanced stage, and the possibility of being affected by the policy in the short term is relatively limited.
According to reports, OpenAI is discussing slowing down the pace of its computing power expansion. One of the reasons is that about 40% of its planned new capacity in 2027 is now located in Democratic-controlled areas, including Texas, Ohio, Georgia and New Mexico. At the same time, OpenAI also stated that it will no longer be required to sign a confidentiality agreement when cooperating with the public sector, and promises to bear the cost of power and grid infrastructure, while publicly disclosing water and electricity usage.
Microsoft has also begun to strengthen its review of some "new cloud service provider" partners, including companies such as Nebius, IREN and Nscale, to assess potential risks and ensure the smooth progress of its AI infrastructure construction plan.
Texas occupies an extremely important position in the U.S. AI infrastructure landscape. It is expected that by 2030, about a quarter of the planned data center capacity in the United States will be located in the state, and the scale of high-load power applications from local power grid operator ERCOT has reached 474GW.
It is against this background that newly elected Texas Governor Gina Hinojosa has become the focus of the industry. The Democrat has previously publicly advocated the elimination of subsidies for data centers, requiring the use of unionized labor, banning confidentiality agreements, and requiring companies to prove that related projects will not drive up residents' electricity and water bills. Therefore, many industry insiders regard it as a representative of the cautious side of data center expansion.
However, the latest developments show that the situation may not be as severe as the market had previously feared.
Meta President Dina Powell McCormick recently stated that a newly elected Democratic governor who made data center issues an important issue during his campaign has proactively contacted Meta and hopes to hold a meeting after he officially takes office. According to McCormick, the reason why the other party did this is very straightforward, that is, "they don't want these jobs to flow to other areas."
This statement was seen by the outside world as a signal to seek balance between local governments and technology companies. Although future projects may face more scrutiny, local governments are also unwilling to lose the job opportunities, investment funds and related industrial chain development brought by data center construction.
Meta remains one of the most active AI infrastructure investors in the world. The market had previously estimated that the company's capital expenditures in 2027 could reach up to US$145 billion, involving multiple ultra-large projects, including the Hyperion project in Louisiana, the Prometheus project in Ohio, and the third large-scale park in Texas that are being promoted in partnership with BlackRock. In addition, Meta is also studying offshore data center solutions.
McCormick acknowledged that the approval process for new projects is expected to receive more scrutiny in the future, but expansion of data center campuses that are already built or in operation will be relatively easy to support because local communities have already reaped real benefits from job growth and investment inflows.
At the same time, the latest analysis from consulting firm Aterio shows that as of November 5, a total of approximately 86.7GW of data center capacity in the United States is under construction. About 50GW of them have entered the main structure construction stage or later, which means that the building shell has been basically completed.
What is more noteworthy is that the capacity of approximately 22.5GW of data centers has now reached or exceeded 75% of construction completion, and the remaining work is mainly concentrated on equipment installation and power access.
Analysis believes that even if new data center construction restrictions are introduced in some areas in the future, the 22.5GW project that is nearing completion will most likely not be significantly affected. Operators can also bypass some grid access restrictions through "behind-the-meter power supply" solutions such as natural gas turbines or Bloom Energy fuel cells.
In addition, Hinojosa has made it clear that all data center projects that have been approved will be allowed to complete construction. This means that most of the new capacity expected to be operational in 2027 can still come online as planned.
As for new projects scheduled to be put into operation between 2028 and 2029, the state government hopes that companies will accept conditions such as canceling subsidies, using union labor, proving that they will not increase residents’ energy and water burdens, and allowing local communities to have a greater say in project decisions.
Based on the current situation, even if some regions in the United States tighten data center construction approval policies in the future, the 22.5GW of new computing power that has entered the advanced stage will continue to be put into operation. The new capacity of this scale is considered to be enough to meet the incremental computing power needs of the US AI industry until 2028. Therefore, compared with the recent intense market concerns about the prospects of the AI industry, the actual situation may not be as pessimistic as outsiders imagine.
Comments