AI Infrastructure Turns to Heavy Assets: Anthropic Locks In $3.4 Billion in Power Equipment as Election Politics Pummels Tech Stocks
Updated · 2026-11-09 11:51 · 5 sources cited

Competition in the AI industry is extending from chips and algorithms to power, land and regulatory approvals. Anthropic's $3.4 billion direct lock-in of gas-fired generating units marks the start of tech companies bypassing intermediaries to control power infrastructure; and the political shift after the U.S. midterm elections has made data center approvals and the AI capex cycle simultaneous market focal points. In China, the rapid growth of the core AI industry is running parallel to the retreat of land finance, and the discussion of a 'new pillar' needs to return to hard constraints of scale, employment and fiscal contribution.
1. $3.4 Billion Locks In 1.7 GW: Anthropic Buys Power Generation Equipment Directly
In late September, frontier AI company Anthropic reached an agreement with San Antonio, Texas-based equipment supplier OnePWR Solutions to spend $3.4 billion to lock in supply of 1.7 GW of heavy-duty generating units. This is the largest known single purchase of power generation equipment by an AI company to date. The deal did not acquire OnePWR itself, but covers the engineering and construction capacity needed to install these generating units; people familiar with the matter said no OnePWR employees were transferring to Anthropic.[1]
The way the deal works is telling. OnePWR ordered the generating units from Caterpillar two years earlier but needed financing to complete the transaction. Anthropic initially planned to become a OnePWR customer, but ultimately chose to take over the equipment order OnePWR had placed with Caterpillar. In other words, it bought a delivery window locked in before prices surged; the equipment can be delivered in the near term and over the next two years. As demand for gas turbines and generating units surges and manufacturing lead times stretch to years, deep-pocketed AI companies are bypassing utilities and specialized power companies to control production directly.[1]
This strategy is not an isolated case. Earlier this year, SpaceX's xAI spent $2 billion to buy mobile gas turbines; Google parent Alphabet acquired energy developer Intersect in late 2025. Anthropic's path sits between a full acquisition and simply buying equipment: it neither fully owns an energy company nor is merely a passive buyer, but shortens delivery times by taking over existing orders. Anthropic is also an important SpaceX customer; by the end of this year, its monthly payment for leased compute is expected to reach $3.1 billion; and by 2045, its compute resource commitments are already close to $900 billion.[1]
Arash Nazhad, managing director and head of clean energy and power at Stifel, commented: 'By owning part of the power infrastructure and having in-house capabilities, the company believes it can better withstand disruption risks beyond its current control.' This explains why AI companies are willing to take on heavy-asset risk: at a stage when power has become a hard constraint on compute expansion, control over power generation equipment is itself competitiveness. Anthropic is preparing for an IPO as soon as this month, which could be one of the largest listings in history; the proceeds will support more than $895 billion in compute infrastructure spending through 2045. But Blackstone has warned investors that its eventual credit rating could be 'junk' rather than the previously expected 'investment grade.'[1]
2. Blue Wave Hits AI Trade: S&P Falls 4.4% in a Day, Data Center Approvals in Focus
After the U.S. midterm elections, market concerns about regulation of the AI sector quickly intensified. The Democrats not only won the House and Senate but also won Senate seats in Kansas, South Carolina and Texas. Barclays wrote in its Nov. 6 equity market weekly: 'The historic blue wave accelerated the unwinding of the Trump trade, and regulatory concerns hit the AI sector hard.' On Wednesday, the S&P 500 rose 1.6% and the Nasdaq rose 1.4%, as the market briefly celebrated the certainty of the election result; but by Thursday, sentiment reversed sharply, with the S&P 500 falling 4.4% in one day, the largest single-day drop since April 2025, and the Nasdaq 100 falling 6.4%, the largest single-day drop since the March 2020 COVID shock, while the VIX fear index surged more than 11 points in one day to close at 28, the highest since late March.[4]
The direct trigger was the outlook for data center approvals. Barclays noted that many Democratic candidates made suspending data center approvals or raising approval thresholds a campaign issue, so investors began pricing in the risk of greater regulatory resistance for tech stocks while deepening worries that the AI capex cycle may peak. Of 47 local ballots involving data centers across the U.S., voters chose to restrict data center operations in 44, with measures ranging from zoning limits and temporary moratoriums to outright bans on new construction. Texas and Georgia together account for about one-third of U.S. AI data center spending; a Barclays credit strategy report noted that by 2027, about 80% of U.S. utilities' reported high-probability data center project pipelines will be in states with Democratic governors, versus less than 30% in 2026.[4]
Comments by Texas Governor-elect Gina Hinojosa were widely cited by the market. She pledged to pause new data center approvals until the legislature establishes stricter power, water and community protection measures, and said: 'I don't hate data centers. What I hate is the way they take advantage of Texas, and that will stop when I am governor.' Barclays' judgment is that AI construction will not be stopped, but will become 'more expensive and slower'—scarce power and permits are reshaping the AI competitive landscape, and industry competition may shift from a 'prisoner's dilemma' to a 'war of attrition,' favoring companies with deeper financial resources.[4]
Institutions clearly disagree on whether the market has fallen enough to buy the dip. Barclays believes some of the selloff may exceed what actual policy risk can explain, because even if Democrats control both chambers, passing aggressive federal data center legislation still faces a high bar of presidential veto; but Jefferies is more cautious on semiconductors, believing valuations and the industry cycle are not yet sufficient to support a clear buy-the-dip call. The SOX index fell 15% in the first four days of election week and is down 38% from its mid-June peak, but is still up 27% year to date; global semiconductor sales grew 144.3% in August 2026, the highest on record. Financing costs are also rising: in recent investment-grade bond issues by Meta and Oracle, new-issue concessions were as high as 30 basis points, versus about 10-15 basis points for Apple. Amazon announced a community investment commitment on Oct. 2, 2026, and OpenAI launched a 'Grid Protection Plan' on Nov. 2, showing that the industry is seeking support from local communities and governments.[4]
3. Can AI Replace Real Estate as a Pillar? The Triple Mismatch of Scale, Employment and Fiscal Contribution
China's third-quarter economic data show clear structural divergence. GDP in the first three quarters was about RMB 107 trillion, up 4.6% year on year; high-tech manufacturing grew 14.3%, digital product manufacturing 13.6%, equipment manufacturing 10.2%, and value added in information and software services 10.4%, sectors mostly related to the global AI race. At the same time, real estate development investment, infrastructure investment and retail sales remained weak. The core AI industry has exceeded RMB 1.2 trillion, with more than 6,200 companies; meanwhile, national revenue from sales of state-owned land use rights fell from a peak of RMB 8.71 trillion in 2021 to RMB 4.15 trillion in 2025, and further to RMB 977.8 billion in the first half of 2026, continuing a sharp year-on-year decline.[2]
As a result, the saying that 'real estate is finished, AI will be the new pillar' has become popular. But by macroeconomic definition, a pillar industry needs sufficient scale, sufficient job creation, sufficient fiscal and tax contribution, and close ties to the balance sheets of banks, local governments and households; a leading industry emphasizes fast growth, high technology density and strong spillovers. China's GDP in 2025 was about RMB 140.2 trillion, and the core AI industry totaled RMB 1.2 trillion, about 0.86% of GDP; even including the broad core digital economy, it accounts for only about 10% of GDP. In the same period, real estate value added was about RMB 8.3 trillion, 5.9% of GDP; real estate and construction together accounted for 12.1%, and the broad real estate chain's combined contribution is often estimated at about 25% of GDP. Real estate alone is already about seven times the core AI industry, and real estate plus construction is about 13 times.[2]
The employment gap is equally significant. Real estate is labor-intensive with a very long chain; direct and indirect employment exceeds 100 million people, about 15% of urban employment nationwide; it is characterized by low barriers to entry, regional dispersion and coverage across all age groups. The core AI industry has about 8 million direct jobs; adding cloud, devices, industrial software, data services and AI manufacturing, it can accommodate about 35 million jobs. These jobs require high-skill collaboration and are concentrated in developed regions such as Beijing, Guangdong, Shanghai, Zhejiang and Jiangsu; AI creates 'new-skill employment,' not 'low-barrier employment.' It restructures the employment structure but can hardly absorb real estate's employment base.[2]
The fiscal contribution gap is even more critical. At real estate's 2021 peak, revenue from sales of state-owned land use rights alone reached RMB 8.71 trillion; land finance plus real estate-related taxes accounted for one-third of local revenue, more than 50% in some provinces and cities, and 90% in some districts and counties; real estate credit (on- and off-balance-sheet) exceeded RMB 100 trillion, about 30% of total banking assets. The AI industry's fiscal contribution is mainly corporate income tax, VAT and personal income tax; currently the combined profits of domestic AI companies are only in the RMB 100 billion range, and related tax contributions are in the RMB 10 billion range, which cannot fill the gap of several trillion yuan in annual land sale revenue.[2]
The household balance sheet angle also cannot be ignored. Housing accounts for nearly 70% of Chinese household assets, and more than 80% for some families, versus 28% in the U.S., 38% in Japan and 25% in Germany. In recent years, home prices have generally fallen by more than 30%, which for households is not just asset shrinkage but the disappearance of borrowing capacity against collateral and a decline in willingness to consume. AI wealth is mainly held in the form of listed company equity, concentrated among a few shareholders, employee options, platform companies and state-owned funds, and does not enter ordinary household balance sheets. Therefore, AI's rapid development can hardly boost the household wealth effect, nor can it drive household consumption. Overall, AI will become China's most important 'leading industry' and 'efficiency engine,' but it is hard for it to replace real estate's pillar function as a trinity of 'investment, fiscal revenue and employment.'[2]
4. The New Bottleneck of the Platform Economy: Algorithmic Distribution and Where Content Producers' Profits Go
The evolution of AI and platform algorithms is changing the survival conditions of content producers. In the short-drama industry, production technology is becoming more mature, and AI has further lowered the cost of scripts, voiceovers and video production, but production companies are not necessarily finding it easier to make money. After a work is completed, it still needs to compete for platform traffic; some companies even need to keep buying traffic and rely on subsequent revenue sharing and rebates to make a profit; once platform cooperation rules change, original investment plans may also be affected. These individual experiences cannot represent the entire short-drama industry, but they raise a question worth studying: why does technology keep lowering the cost of starting a business, yet entrepreneurs do not necessarily accumulate more profit?[6]
Harvard Business School professor Carliss Baldwin proposed in the second volume of *Design Rules*, published in 2024, that technology not only changes production efficiency but also changes the division of labor among firms and their organizational forms. Digital platforms, through standardized interfaces and modular design, allow work that once had to be done by a single company to be handed to many independent firms, which is why the internet could incubate a large number of startups. But Baldwin also distinguishes technical bottlenecks from strategic bottlenecks: firms create value by solving technical bottlenecks, but often need to control hard-to-replace key links to keep capturing value. In the short-drama industry, AI has removed old technical bottlenecks such as shooting, editing and voiceover, but an increase in content supply does not mean consumer attention increases proportionally; the distribution system that can effectively screen content, find audiences and complete monetization has instead become a scarcer resource.[6]
Hongguo Short Drama offers a case worth observing. According to reports, Hongguo not only cooperates with a large number of external copyright holders and production companies but also has its own content production system; external companies can either provide independently developed works or take on production tasks organized by the platform. In 2026, Hongguo's adjustment to the guaranteed-minimum mechanism for some commissioned projects caused operating difficulties for some production companies and industry controversy. This shows that a platform does not need to own all production companies to deeply organize production activities across the industry through script development, content procurement, recommendation and distribution, and revenue rules. The platform not only organizes the supply of works but also deeply participates in matching audience demand; its influence on producers is no longer limited to whether a work can be released, but can extend to how many times each work is shown, what kind of audience it reaches, and how revenue is distributed.[6]
5. The Fragility of an Interconnected World: Cybersecurity Becomes a Hidden Cost of Heavy-Asset Expansion
As AI infrastructure expands rapidly, cybersecurity risks are also rising. An FT editorial pointed out that a series of cybersecurity incidents have made governments, companies and individuals realize again the fragility of a highly interconnected world, and these incidents offer some common lessons. As compute, power and data become increasingly concentrated in a few large facilities and platforms, the more interconnected the system, the greater the potential impact of a single point of failure. For AI companies pouring huge sums into power generation equipment, data centers and compute clusters, cybersecurity is no longer just a software-level issue but a hidden cost that must be counted in heavy-asset investment.[3]
From Anthropic directly buying power generation equipment, to U.S. data center approvals becoming an election issue, to China's debate over AI and real estate's pillar status, several threads point to one trend: competition in the AI industry is shifting from 'light-asset expansion' to 'heavy-asset deepening.' Power, land, approvals, fiscal revenue and platform rules have all become key variables determining investment returns. Going forward, watch Anthropic's IPO progress and changes in its credit rating, the actual implementation of local data center restrictions in the U.S., the degree to which China's AI industry matches the land finance gap, and the continuing impact of platform distribution rule changes on content producers.
Sources
- [1] Anthropic斥资34亿美元接手发电设备 · 2026-11-09 00:00 · www.ftchinese.com
- [2] AI取代房地产成为中国支柱产业? · 2026-11-09 00:00 · www.ftchinese.com
- [3] FT社评:黑客攻击揭示联网世界缺陷 · 2026-11-09 00:00 · www.ftchinese.com
- [4] “蓝色浪潮”重创科技信仰,AI交易何时企稳?未来如何演变?华尔街密集研判 · 2026-11-08 14:00 · wallstreetcn.com
- [6] “字节时间”之下:平台如何改变企业命运? · 2026-11-09 00:00 · www.ftchinese.com