The end of AI is not just about electricity, but also the electricity bills! As data centers become the focal point of the midterm elections in the United States, investments in AI infrastructure are facing a "ballot pressure test."

As the most critical pillar of the artificial intelligence industry chain, spending on data center infrastructure is expected to reach new heights by 2026 and is projected to set multiple historical spending records at least until 2030. In the face of this unprecedented infrastructure construction boom, opposition to data centers has sharply escalated across the United States this year, evolving from a niche local issue into a key discussion topic ahead of the November midterm elections in Congress. Almost overnight, candidates across various political positions have adjusted their campaign messaging to respond to grassroots opposition from the public against this AI computing infrastructure wave, supported by hundreds of billions of dollars in technology industry spending. Both major parties in the U.S. are striving to adapt promptly to shifting public opinion and determine their respective strategies in response to the AI data center construction frenzy. However, President Donald Trump continues to advocate for rapid construction progress. "Other countries are delighted about this anti-data center movement," he tweeted on August 31. His stance has put many Republicans in a dilemmasome wish to support increasingly agitated constituents but do not want to incur the wrath of the current U.S. president, a paramount figure in the Republican Party. Rising electricity costs are increasing the burden on American consumers, provoking discontent among lower- to middle-income voters. The competition for Pennsylvania's 7th congressional district, a crucial swing district for both parties, may reflect the significance of the AI computing capacity initiative in 2026. The construction constraints facing data centers have already become concrete. Texas Governor Abbott on August 3 demanded a comprehensive review of data centers that are advancing through the grid interconnection process managed by Texas' grid operator, prohibiting any projects from proceeding until the review is completed. The state government disclosed at that time that the interconnection applications involved an additional electricity demand exceeding 474 gigawatts, with about 90% coming from data centers; these figures reflect application scales, not operational loads or confirmed orders. The review encompasses electricity usage, water consumption, financial incentives, and community impacts, indicating how bipartisan life cost disputes tightly linked to the midterm elections can cascade into investments in AI infrastructure through interconnection conditions and project timelines. The research report released by Michael Hartnett, a senior strategist at Bank of America nicknamed "Wall Street's Most Accurate Strategist," incorporates U.S. policy conditions and funding costs into the valuation analysis of AI computing power: if Republicans fail to win the Senate and consecutively lose both the Senate and House, project approval delays will postpone computing power go-live and the realization of actual AI revenues and profits; financing rate increases driven by a Democrat-controlled Congress will elevate the costs of data center construction and reduce the present value of future cash flows. Following this valuation framework, global semiconductor and data center power supply chain equipment firms, as well as AI computing leasing companies dominated by "new cloud" forces, will be negatively impacted by deteriorating expectations, ultimately validated through actual orders, deployment rates, capital expenditures, and free cash flows. This selling pressure may first manifest as valuation contractions for AI-related stocks; should approval and financing constraints lead cloud vendors to further delay construction and equipment procurement, the impact will cascade along the supply chain to supplier orders, revenues, and operator cash flows, causing policy concerns to evolve further into earnings pressures. Hartnett identifies a Democrat-controlled Congress as a significant pressure scenario: heightened market worries around tax increases, stricter regulations, and constraints on AI construction will pressure corporate earnings expectations and valuations simultaneously; accordingly, he posits that U.S. stocks could drop by over 10%, the dollar could weaken, and bond yields could decline. In contrast, he views a Republican-controlled Congress as a scenario that expands risk appetite, referring to a Republican-controlled Senate and Democrat-controlled House as a "Goldilocks deadlock" mildly favorable to risk assets. The trillion-dollar AI computing expansion raises new challenges: voters, power supply, and financing costs. Wall Street financial giant Morgan Stanley's insight regarding the so-called AGI epoch initiated by OpenAI's Astra large model highlights that enhanced capabilities of AI models render more workloads economically viable, thus intensifying supply constraints in power, substrate, and storage manufacturing. Its scenario analysis indicates that the power capacity corresponding to super-scale cloud vendors' computing deployments will expand from approximately 35 gigawatts in 2025 to about 145 gigawatts in 2028, achieving a nearly 4.1-fold increase. The launch of Astra has fueled market expectations for artificial general intelligence (AGI), resulting in a more robust trajectory of computing demand, particularly as the new growth model of "pay-for-results" is anticipated to generate stronger overall computing needs. More direct evidence of immediate computing demand has emerged from AI research and development processes themselvesspecifically, the "recursive self-improvement (RSI)" research trajectory, where AI begins to "create AI." As the advanced cutting-edge large models led by Astra generate increasingly vigorous demand for AI computing power, Morgan Stanley forecasts total capital expenditure for the four major North American supercloud and AI application firms' data centers will rise from $917 billion in 2026 to $1.47 trillion in 2027, and $1.64 trillion in 2028. Meanwhile, the projected deployment capacity is expected to increase from 35 gigawatts in 2025 to 145 gigawatts in 2028. The sustainability of AI investment hinges on whether capital returns and operational cash flows can support expansion. Morgan Stanley's scenario analysis suggests that model companies providing API services using their own infrastructure achieve a return on invested capital (ROIC) of about 46%, which surpasses the 31% for cloud vendors renting GPUs, and the 25% for those reliant on third-party infrastructure to provide API services. This underlines that firms managing models, computing power, and commercialization channels can achieve higher returns, although these are quantitative model estimates. Concurrently, Morgan Stanley projects that the combined operational cash flows of the four major giants will increase from $739 billion in 2026 to $1.23 trillion in 2028, while new debt financing needs will decrease from $238 billion to $90 billion. Resistance against data centers is expanding the constraints on AI investment from chip and power supply to construction permits and community acceptance. Annenberg reported in August that a survey conducted between June and July indicated that the proportion of adults opposing new local data centers had risen from 49% to 61%, although public opinions on the overall impact of AI did not show significant changes. Meanwhile, Data Center Watch confirmed that at least 75 projects valued at approximately $130 billion were stalled or postponed in the first quarter. For the AI industry chain, its direct impact is the increased uncertainty around the timeline for new computing capacity to become operational, and delays in equipment delivery, cloud service expansion, and market-focused AI-related revenue confirmations might follow, leading the AI computing industry chain to face a correction due to valuation compression. The core economic contention revolves around who ultimately bears the costs of new power generation, transmission, and water supply. After technology companies signed the "Electric Bill Payer Protection Commitment," there remain discrepancies in cost-sharing. Media reports in early September indicated that Microsoft was appealing a new regulation in Virginia that requires data center developers to bear the upfront costs of transmission infrastructure. This case illustrates that the concept of "companies bearing their own electricity costs" involves questions around timing of payments, scope of infrastructure, and risk distribution during implementation. If more costs need to be borne upfront by developers, the initial capital investment and financing needs will increase; if costs are passed on to residents, it could lead to further political resistance. How these commitments are written into pricing agreements and construction conditions will directly affect the investment returns of data centers. Hartnett's warning signals regarding a "bond-dominated bubble," combined with local construction resistance, point towards the cash recovery cycle for AI projects. Prolonged elevated yields in the long-term U.S. Treasury bond market will undoubtedly significantly increase the financing costs associated with the AI computing infrastructure being built by tech giants and depress the present value of future cash flows; delays in approvals will also push back the timeline for projects to start generating revenue. Together, even if the long-term outlook for computing demand remains overwhelmingly robust, annual project returns could decline. How does the public view data centers? Public dissatisfaction is rising rapidly. A survey by the Annenberg Public Policy Center at the University of Pennsylvania conducted in August found that approximately 61% of American adults oppose the construction of new data centers in their areas, an increase of 12 percentage points from a survey conducted between February and March. In an increasingly politically polarized country, this has become a rare bipartisan issue: according to a summary of a Reuters/Ipsos poll conducted in June, 75% of Democrats and 63% of Republicans expressed opposition to data centers being built in their local communities. The Reuters/Ipsos survey also revealed that 47% of registered voters in the U.S. listed living costs as the most significant factor influencing their midterm election votes, while 71% of American adults disapproved of Trump's handling of cost-of-living issues; in parallel surveys, only 33% of respondents approved of his overall governance amidst soaring energy costs and resurgent inflation. These early poll results conducted by the media cannot be directly converted into congressional seats or electoral victory probabilities but largely highlight the energy, electricity prices, and consumer burden and indicate that the resistance to data center construction has become a necessary policy backdrop. In a general climate of public discontent regarding living costs and economic conditions, voters are increasingly concerned about the impact of data centers on their electricity bills. The largest power grid in the U.S. covers 13 states in the East and Midwest. A report from the independent market monitoring agency Monitoring Analytics revealed that due to data center demand, the overall electricity prices for this grid surged by a record 76% in the first quarter of this year. An analysis by Bloomberg News found that from 2020 to 2025, wholesale electricity costs in areas with high concentrations of data center activities increased by as much as 267%. While U.S. tech giants have committed to paying separately for the electricity required for the development and construction of AI data centers, ensuring they will accelerate the construction of large data centers that will initially not connect to the grid, it remains unclear how these commitments will be implemented in practice as demand for computing power continues to grow in the short term. Some critics are also concerned about the water usage of data centers, although industry leaders believe these worries are exaggerated. Others are dissatisfied with the loud humming produced by these facilities or the lucrative tax incentives they enjoy. Moreover, the public's broader unease and resistance towards artificial intelligence often manifest as opposition to data centers. Many believe AI poses a threat to jobs and privacy. How do key political figures respond? Progressive figures such as Senator Bernie Sanders from Vermont and Representative Alexandria Ocasio-Cortez from New York were among the early prominent politicians to call for a pause in data center constructions this year. Subsequently, many lawmakers across the political spectrum, including Republicans from the "Make America Great Again" camp and centrist Democrats, followed suit, urging a series of policy measures to ensure local residents would not bear the burden of rising energy costs or environmental disruptions. Even once staunch supporters of data centers have shifted their stance on this issue, such as Democratic Pennsylvania Governor Josh Shapiro and Republican Texas Governor Greg Abbott. Last year, Abbott described Texas as "the center of AI development" during the announcement of a $40 billion investment by Alphabet-owned Google. By mid-August 2026, he had implemented measures to pause new data center constructions. In Pennsylvania, Shapiro initially helped expedite data center development but later signed an executive order imposing one of the strictest regulatory frameworks for such constructions in the U.S. Across states like Wyoming, Ohio, Florida, and Pennsylvania, aggressive campaign ads have highlighted opposition to data centers. As of August, campaign teams had spent a total of $31 million on ads referencing data centers. In Florida, Republican gubernatorial candidate Byron Donalds ran ads promising to ensure that data centers would not raise energy bills. In Ohio, Democratic Senate candidate Sherrod Brown has attacked the prior support shown by Republican incumbent Senator Jon Husted for these facilities. As the midterm elections approach, which party will benefit more? The Republican Party is clearly on the defensive. The Democrats have more swiftly strengthened their opposition to data centers, while the political right continues to struggle with Trump's support for the industry. Some Republicans are fully supporting a halt to data center constructions, while others are trying to devise more nuanced strategies. For example, in Florida, Donalds proposed a plan resonating with Trump's "Electric Bill Payer Protection Commitment." This is a voluntary agreement signed by tech companies aimed at protecting American consumers from price hikes caused by data center electricity usage. However, there are rising concerns within the Republican Party that they are losing the narrative battle surrounding data centers. In a memo sent to AI firms in August, the National Republican Senatorial Committee warned that Brown is positioning opposition to data centers as a central part of his campaign to attack Husted. The committee cautioned in the memo: "In this campaign, nothing is dragging down Husted's fortunes more than data centers." What electoral processes might data centers influence in 2026? The Ohio Senate election stands out as the most prominent case dominated by the data center propaganda war, but several other elections nationwide are experiencing similar phenomena. In Wisconsin, Republican gubernatorial candidate Tom Tiffany has invested in aggressive advertising tying his Democratic opponent to data centers. In Michigan, Republican Senate candidate Mike Rogers announced his support for a one-year statewide pause on new data center constructions. In this critical swing state election, his stance on this issue aligns with that of his Democratic opponent Abdul El-Sayed. It is difficult to gauge how many voters will prioritize a candidate's views on data centers when casting their votes, but the propaganda against data centers has already become an important means for candidates to showcase their populist and anti-establishment positions. How are core AI companies and data center operators affected by the wave of resistance? A report published by the investment firm Kimmeridge Energy Management on August 26 indicated that up to half of the planned large AI data centers in the U.S. face risks of delays or cancellations, with political opposition being a significant factor. According to Data Center Watch, in just the first three months of 2026, at least 75 data center projects have been stalled or postponed due to local public opposition, with a total value reaching approximately $130 billion. Data center projects are facing serious delays and cancellations for multiple reasons, including some power grids being unable to accommodate the immense power they require and challenges in sourcing critical electrical components. However, political opposition is increasingly becoming a significant factor slowing down construction. Even if projects proceed, companies like Google, Amazon, Oracle, and Microsoft are investing substantial resources to gain favor with local communities to avoid backlash. To date, this remains a formidable task. AI companies and their supporters have pledged to invest hundreds of millions of dollars in campaigning during this year's midterm elections. The initiative organization "Build the AI America," supported by the pro-AI super political action committee "Leading the Future," has committed to investing millions to actively promote data centers through advertising in swing states like Ohio, Kansas, and Wisconsin and proactively disclose plans for grid construction that insulate data centers from local electricity systems.