Bank of America's $250 Billion Puts the AI Bubble Question on a Bank Balance Sheet

Bank of America has announced a $250 billion financing commitment to new data center construction and other infrastructure projects, joining Morgan Stanley and Goldman Sachs, which have made comparable announcements ^[1][2]. That completes a pattern among the big three: whether the current AI frenzy is a bubble is now, in part, a question about bank credit books, and the banks also face the challenge of squaring this buildout with their past promises of environmental friendliness ^[3].

Start with what the number is made of. According to Karen Fang, Bank of America's global head of infrastructure and sustainable finance, the $250 billion spans on-balance-sheet lending, project finance bonds, and project-level capital including equity and debt in both public and private markets ^[4]. Those are very different risk positions. On-balance-sheet lending is retained; bonds and third-party project capital are distributed. American Banker's account lists the categories without giving proportions, so the share of the $250 billion the bank actually intends to hold is unknown ^[5].

The riskiest slice is named explicitly. For some data center projects, the bank is helping developers raise greenfield capital, meaning money used to secure permits, equipment, labor contracts and offtake contracts, the agreements between producers and buyers ^[6]. That is capital committed before a project has a permit or a customer. Other work in the program is more conventional: roads, ports and water infrastructure ^[7].

On accounting, one transaction that closed this April counts toward the $250 billion ^[8]. So the figure is a tally that includes completed work, not purely forward capacity. Fang calls it "one of the largest private sector announcements to date to mobilize within a very condensed short time frame," while also noting that none of the big banks' announcements happened overnight and that everyone has been working on this for some time, "witness the AI-driven infrastructure [capital expenditure] boom" ^[9][10]. Her stated framing is urgency: "Our energy and power infrastructure is dated," and the bank feels "a tremendous sense of urgency to get all the investors, banks, developers, corporations, governments and consumers together to build this infrastructure now" ^[11][12].

The sustainability collision is structural rather than rhetorical. Fang runs infrastructure investment and sustainable finance in the same seat, which she says demonstrates the bank's dedication to a cleaner, lower-carbon economy ^[13]. She also says the country needs renewables and batteries plus gas power generation and potentially nuclear and advanced geothermal ^[14]. Separately, the bank has a 10-year, $1.5 trillion sustainable finance goal, to which Fang says several renewable and storage projects contribute ^[15]. The infrastructure commitment is equal to about one sixth of that decade-long target ^[16], and unlike the target it carries no stated clean-generation restriction.

The community economics are worth reading closely. At the Saline Township, Michigan data center, developer Related Digital and Blackstone are rebuilding a neighborhood recreational center and have paid for new fire trucks, and they promised local Detroit Edison customers that power bills would not increase for the next 10 to 15 years because Oracle and OpenAI are paying for the power, the substation upgrades and the batteries ^[17][18]. That ratepayer protection is only as durable as those two counterparties. Fang says the site uses closed-loop water cooling that will consume less water than an office building or a comparable farm ^[19], and that "if the communities don't love it, the project isn't viable for the long term" ^[20].

Steve Rubinow, former chief information officer at the New York Stock Exchange and now an associate teaching professor at the Illinois Institute of Technology, told American Banker that efficient designs and water recycling serve cost control as much as climate, but pointed to reports that Lake Mead and Lake Powell are at their lowest recorded levels, and that in some parts of the country there is not much water to allocate ^[21][22].