KBW Reads $1.75 Trillion Commitments from BofA and Morgan Stanley as Bet on AI Capex Beyond 2027
Analyst note by Christopher McGratty published this Monday interprets infrastructure initiatives from both banks as a sign of longevity for the data center investment cycle.
KBW published a note this Monday interpreting the recent infrastructure commitments announced by Bank of America and Morgan Stanley as evidence that the AI capex cycle is not peaking this year but is just the beginning of a decade-long origination for America's major banks. "We view the longer-term commitments as clear signs of confidence from the largest lenders that this capex cycle has room to run beyond 2026," wrote analyst Christopher McGratty.
The total calculation is substantial. Bank of America announced on August 12 the Critical Infrastructure Finance Initiative, aiming to mobilize $250 billion over 18 months until the fourth quarter of July 2027. Morgan Stanley had launched two days earlier with the U.S. Innovation Infrastructure Initiative, sized at $1.5 trillion to be facilitated over ten years. KBW also includes the mobilization of over $500 billion that Nvidia arranged with Goldman Sachs for computing financing.
Where the Money is Actually Going
A close reading of the announcements matters. BofA's commitment is not on-balance sheet: it accounts for each bookrunner mandate in data center bonds, every advisory on plant purchases, and every co-investment in grid optimization. It’s a fee pipeline, not a risk one. Morgan Stanley's is more ambitious and more diversified: it encompasses AI, quantum, semiconductors, cyber, aerospace, defense, pharmaceuticals, and critical minerals, providing the bank room to reclassify operations it would already be undertaking.
The central economic point of KBW's note is not the absolute number; it's the duration. American banks have been reporting double-digit growth in commercial and industrial credit since the first quarter, something not seen since 2022. KBW argues that the average maturity of this credit has increased, and this elongation only makes sense if lenders believe that the demand for electricity for AI servers will continue past 2027.
The Comparison Global Markets are Making
Across the Atlantic, Deutsche Bank and UBS have yet to announce initiatives with proprietary names on an equivalent scale. The average ticket for data center origination in Europe remains concentrated in the hubs of Frankfurt, Amsterdam, and Dublin, with a strong participation from private credit funds rather than bank balance sheets. Frankfurt closed the first half of the year with 640 megawatts of new contracted capacity, according to CBRE, but financing is predominantly structured.
The Japanese angle is more concrete. MUFG announced in June a $100 billion program for medium-term technological modernization investment, and Mizuho expanded its credit line for domestic data center operators in Chiba and Kansai. The scale difference with the U.S. is significant, but the mechanism is the same: banks with large balance sheets are partially replacing hyperscalers in origination, and gaining fees. Nomura estimated in a July report that 12% of the investment banking fees from the Japanese market in 2026 is expected to come from energy and infrastructure for AI.
What Weakens the Thesis
KBW's note is not uniformly optimistic. McGratty acknowledges that nearly 40% of companies measured by Bain & Company in its Automation and AI Pathfinder 2026 study achieved AI savings below 10%, while the declared target was between 11% and 20%. If enterprise return in productivity does not accelerate, token consumption slows down, and the argument for loan duration becomes less solid.
The second counter-argument is what analysts at JPMorgan Asset Management have been calling "funding circularity": Nvidia invests in a hyperscaler that buys Nvidia chips, the bank finances the construction of the data center hosting the chip, and the same hyperscaler hires the bank to manage the next bond. The math works as long as AI revenue growth continues; if it stalls, the cascading effect is the opposite.
The distinction that investor discussions insist on ignoring is between operationally profitable hyperscalers financing capex with their own cash flow and AI labs burning venture capital to train models. BofA and Morgan Stanley are originating for the first group; exposure to the second, more fragile group, remains small on the balance sheets of major American banks. This significantly separates this cycle from the private credit cycle of 2021.