Politicians Fight the AI Boom. Pension Funds Bank on It

America’s backlash against data centers has a retirement problem. Elected officials want to protect communities from higher electricity bills, pressure on water supplies and jobs lost to automation. Pension funds under their oversight own stakes in the companies driving that expansion. Someone campaigning against a new computing campus may also be counting on the industry’s profits to help pay their pension.

The scale is substantial. According to The Wall Street Journal, 13 large US public pension funds accumulated $138 billion in paper gains on AI-linked stocks from late 2022 through June. Fifteen companies, including cloud giants and chipmakers, accounted for 27% of their US stock portfolios by June, up from 12% when ChatGPT launched. Those holdings generated roughly 57% of the portfolios’ domestic stock gains. The calculation left out foreign shares, private equity and private debt. A handful of AI winners supplied more than half the gains from the funds’ US stocks.

The Journal traces several versions of this political split. New York State paused certain state permit applications for large data-center projects, without suspending local permits, while its pension fund’s AI holdings kept growing. Pennsylvania’s treasurer opposes further data-center expansion while helping oversee a pension system with billions in AI assets. California union leaders want pension managers to account for the damage AI could do to workers and communities. CalPERS, with roughly $650 billion in assets, has committed $500 million to a DigitalBridge fund whose digital-infrastructure mandate includes data centers. New York State’s comptroller has pressed technology companies over residents’ concerns about electricity, land and water while overseeing a fund that profits from those businesses. Projects that encounter resistance at public hearings can still look excellent in a pension fund’s investment report.

Stock-market gains and construction finance follow different routes. Buying an existing share sends money to its seller; new capacity gets funded through securities issuance, loans, investment funds and direct equity commitments. The $138 billion is also an unrealized gain. But pension portfolios already depend on the industry, and managers did not need to make an explicit decision to bet on superintelligence to get there. Owning the broad market was enough as technology giants outpaced the rest.

Equable Institute tried to measure that largely unplanned exposure. Among 23 major US public pension funds with available disclosures, a basket of roughly 50 AI-linked listed companies represented 8.6% of total assets. Equable’s national estimate puts the exposure at no less than 8–10%, or about $513–642 billion. Externally managed investments and private markets are only partly captured. That suggests the estimate may understate total exposure. Its denominator also matters: Equable is measuring a share of all pension assets, whereas the Journal’s 27% refers only to US stock holdings.

This creates a problem for the usual diversification playbook. In a Bloomberg investigation, pension and sovereign funds try to work out how often they have made the same bet in different investment wrappers. The $94 billion Los Angeles County Employees Retirement Association, or LACERA, estimated its AI exposure at anywhere from 8% to 19%, depending on the definition. Chief investment officer Jonathan Grabel sees that assessment as a starting point for investment decisions. Monte Tarbox, investment chief of New York City’s pension systems, turned down a private-equity fund proposal because AI exposure was already too concentrated, both in the proposed strategy and across the city’s portfolios. The city and state funds are separate investors, each with its own exposure to assess.

Managers disagree about how much is too much. Kari Vatanen, head of asset allocation at Finland’s Elo, argues that being in the right sector matters more than getting the asset-class mix right, according to Oninvest’s account of Bloomberg’s reporting. There is a clear investment case for that view: stepping away from a technological shift too early can mean years of lagging returns. Britain’s Pensions Regulator wants trustees to examine the downside just as closely. As Reuters reported on October 22, it urged funds to assess direct and hidden AI exposure and stress-test their portfolios against a severe downturn. Different headings in an investment report do not guarantee independent sources of return.

“Having the right sector in your portfolio is more important than having the right asset class.” — Kari Vatanen, Elo, quoted by Oninvest

Some of the largest international funds are also supplying capital directly to the buildout. In October 2024, Canada’s CPP Investments and Singapore’s sovereign fund GIC announced a joint venture with Equinix. Each fund committed $2 billion in equity for a 37.5% stake; Equinix’s share was 25%. Planned borrowing would bring the venture’s total investable capital to more than $15 billion. Executives at CPP Investments and GIC explicitly tied the project to growth in AI and cloud computing. Here, Canadian retirement savings and Singaporean state reserves become capital for facilities where cloud providers will install their equipment.

Sovereign investors have national objectives to consider as well: access to computing capacity, energy and technology supply chains. According to Global SWF, sovereign funds invested about $66 billion in AI and digitalization in 2025, a combined category extending well beyond model developers. The buyer group for Aligned Data Centers included Abu Dhabi’s MGX, the AI Infrastructure Partnership and BlackRock’s Global Infrastructure Partners. Kuwait Investment Authority and Temasek were named as financial anchor investors in AIP. Aligned’s announced enterprise value was roughly $40 billion, including debt. That figure does not reveal each participant’s contribution, but the buyers’ identities show how international the financing of America’s computing boom has become.

Norway illustrates another route into the race: the stock market. Its sovereign fund’s 2025 annual report shows technology accounting for 28.5% of its equity portfolio, with Nvidia, Alphabet and Broadcom among the biggest contributors to returns. The fund owned shares in 7,201 companies, yet US technology leaders had an outsized influence on its performance. Chief executive Nicolai Tangen has also urged businesses to use AI to raise prosperity broadly, warning that using it simply to cut jobs risks a political backlash. Norway faces a familiar tension: earning money from automation while keeping public support for its consequences.

Sovereign managers are well aware of the concentration. The 2026 Invesco Global Sovereign Asset Management Study, covering 90 sovereign funds and 54 central banks, found that 52% of sovereign respondents to its AI-risk question identified market concentration as the leading concern. Valuation or bubble risk was cited by 38%, and rapid obsolescence by 36%. Infrastructure offers an appealing way to earn revenue from the demands of several AI developers. Yet a data-center landlord, a lender to a cloud provider and a shareholder in a chipmaker may all depend on the same small circle of customers being able to pay for computing.

We have examined who will ultimately pay for AI’s infrastructure boom. Pension and sovereign funds supply an important part of the answer. A single fund can own shares in the equipment supplier, a stake in the building and an interest in the credit fund financing the tenant. Each investment has its own reporting, but a demand shortfall can damage cash flows throughout the chain. Widespread adoption would not settle the question of returns, either. Cheaper computing benefits users while potentially eroding the value of expensive infrastructure financed by early investors. The replacement cycle matters especially when, as in our breakdown of a $30 billion data center, chips and servers account for more than half the budget and need replacing much sooner than the building.

In a defined-benefit pension system, a sustained investment shortfall becomes a fight over who makes up the difference. Years of weak returns put pressure on contributions and public budgets. A retiree’s financial interest in AI’s success is real. So is a worker’s interest in keeping a job, affordable electricity and a reliable water supply. Public investment funds have put those interests on the same balance sheet. Politicians promise protection from AI’s costs today; pension managers are counting on its expansion to help fund retirement tomorrow. What it costs to keep both promises is only beginning to emerge.