The AI Financing Plumbing Is Paying Off — Until It Isn't

The five largest U.S. banks collectively reported $114 billion in capital markets revenue for the first half of 2026, a 31.5% jump over the prior year. JPMorgan posted the highest quarterly profit in American banking history. Goldman Sachs saw investment banking revenue surge 55%.

The market is reading this as proof that banks are the overlooked winner in the AI cycle. The conclusion is correct. The risk implication is where the story actually matters.

The real question isn't whether banks are benefiting from the AI capex boom. The real question is whether this is a diversified revenue tailwind or a concentrated single-point dependency — and the evidence, frankly, leans uncomfortably toward the latter.

The money flow shift: from self-funded to bank-mediated

For most of the 2020s, hyperscaler AI spending was funded internally. Amazon, Microsoft, Google, and Meta had the cash. Then the scale changed.

In 2025, AI-related capital expenditure by Big Tech climbed to roughly $400 billion — and was projected to approach $700 billion in 2026. Jamie Dimon, who actually sits at the center of the debt market where this money is structured, put it plainly on the Q2 earnings call: AI capex rose from $400 billion last year to $700 billion this year, and is on track to exceed $1 trillion next year.

That growth trajectory is the reason the financing dynamic has flipped. Capex spending, excluding dividends and share repurchases, now consumes 94% of hyperscaler operating cash flows, up from 76% in 2024. The five largest hyperscalers raised a record $108 billion in debt in 2025 alone — more than three times the average over the previous nine years. JPMorgan estimates $1.5 trillion in investment-grade bonds will be required over the next five years to fund AI data center buildout.

Put plainly: the companies that used to fund their own AI infrastructure are now turning to the bond market in structural amounts for the first time. And banks are the ones structuring, underwriting, and selling that debt.

That's where the bank revenue is coming from, and it shows up across three channels.

Investment banking. Goldman Sachs investment banking revenue jumped 55% in Q2 2026. Bank of America's rose 50%. JPMorgan's grew 30%. Goldman CEO David Solomon described the firm's deal backlog as at its highest level in five years, attributing the pipeline to AI infrastructure deal flow that is still in its "early innings."

Trading. AI-related capital markets activity is fueling equity trading volumes. JPMorgan equities trading revenue jumped 86% in Q2. Goldman's surged 72%. JPMorgan CFO Jeremy Barnum noted the activity was "downstream of the AI theme writ large on a global basis." Goldman's revenue hit $20.3 billion for the quarter — a 39% increase.

Complex structuring. Beyond straight bonds, banks are packaging data center assets into asset-backed securities. About $13.3 billion in data-center-backed ABS was issued across 27 transactions in 2025, a 55% increase over 2024. Blackstone closed a $3.46 billion commercial mortgage-backed security offering to refinance QTS debt — the largest deal of its type in 2025, versus just $3 billion in data-center-backed CMBS for all of 2024.

Banks are earning fees at every stage of the AI capital cycle. Underwriting, advisory, lending, trading. It is a structural revenue engine.

The concentration problem

Here's what changes the frame.

The Bank for International Settlements — the central bank of central banks — published a warning in its 2026 Annual Economic Report that treats this AI financing boom with the same language it reserved for the dot-com bubble, the British railway mania, and the canal speculation of the 1830s. The BIS notes that all of those episodes began with genuine technological breakthroughs. All of them ended in recession, because capital exceeded what commercial returns could justify.

The specific mechanisms matter.

Contest-theory risk. The BIS models that competitive pressure among hyperscalers is driving capex higher to the point where the net economic surplus for the sector could turn negative. Only a few players will dominate the AI market, which means the losers' investments become stranded assets. A disappointment in AI returns could trigger a sudden pullback in financing, turning the capex boom into what the BIS calls a "protracted investment bust."

Circular financing. Hyperscalers, chipmakers, and AI labs are linked through a complex web of poorly disclosed private arrangements. Chip makers take equity stakes in AI labs, which then sign multi-year purchase commitments with the same hyperscalers. $49 billion of the $53 billion in "other income" for Alphabet and Amazon in Q1 2026 came from equity stakes in private AI firms, which then use that capital to sign cloud computing deals with the same hyperscalers. If hyperscalers slow capex, the entire supply chain faces simultaneous revenue shortfalls.

Non-bank exposure. Private credit funds have quadrupled lending to the AI and IT sectors in five years, now representing about 15% of their portfolios. BIS Asia-Pacific representative Zhang Tao warned that a correction could unwind "much faster than previous banking crisis episodes" because the financing flows through hedge funds and private credit vehicles with less regulatory oversight.

On the earnings calls, bank CEOs haven't ignored this. Dimon said on the Q2 call that the market is "getting close to as good as it gets" and that "we just don't know how long it's going to last." Bank of America strategists flagged in July that AI stock concentration — the "AI Big 10" at 41% of the S&P 500 — has reached levels comparable to the precursors of previous market bubbles.

What this means is not that the AI cycle is ending. It means that the banks riding this wave have their revenue increasingly tied to a single, self-reinforcing feedback loop. If hyperscalers pull back on capex — whether because returns disappoint, regulation shifts, or capital markets seize — underwriting, advisory, lending, and trading revenues could weaken simultaneously.

The valuation picture

The good news is happening. The question is whether the prices already assume it keeps happening.

Bank	Forward P/E	TTM P/B	Dividend Yield	YTD Return
JPM	16.4x	2.54x	1.57%	+11.0%
GS	18.8x	2.47x	1.54%	+18.3%
MS	22.5x	2.89x	1.85%	+21.9%
BAC	16.0x	1.47x	1.77%	+14.9%
C	14.0x	1.06x	1.78%	+15.7%
WFC	13.4x	1.45x	2.06%	-6.4%

Goldman Sachs and Morgan Stanley are the purest plays on investment banking and trading revenue, and they're priced accordingly — at 18.8x and 22.5x forward earnings, respectively. That's not cheap for financials. Citigroup and Wells Fargo trade at significant discounts, at 14.0x and 13.4x forward, with P/B ratios around 1.0x-1.4x.

JPMorgan is the most interesting data point here. Revenue growth of 13.5% year-over-year, with Q2 2026 revenue hitting $57.3 billion and EPS of $7.70 — both well above the consensus estimates of $51.1 billion and $5.59. The stock trades at a forward P/E of 16.4x, which is a premium to the broader bank sector but below Goldman and Morgan Stanley. It has the largest franchise, the deepest balance sheet, and the most diversified exposure across lending, investment banking, trading, and asset management.

Goldman, by contrast, is the highest-beta play. Revenue growth of 17.8% YoY, with Q2 revenue of $20.3 billion and EPS of $20.98, both well above consensus of $16.2 billion and $14.51. That kind of beat is impressive. But at 18.8x forward P/E with 2.47x book, Goldman's pricing assumes this momentum persists.

Where I put the capital

I believe the AI financing tailwind for banks is real and structural, not a blip. The shift from self-funded to debt-funded hyperscaler capex represents a genuine change in how capital flows through the financial system. The fee income from underwriting, trading, and advisory is already showing up in the earnings.

However, the concentration risk is not theoretical. It's the same risk that turns a bull story into a bear story when the underlying cycle reverses. The BIS warning about circular financing, non-bank vulnerabilities, and contest-theory overinvestment isn't noise. It's the kind of signal that tends to be right six months too early and wrong at no point in time.

My allocation view:

• JPMorgan is the core position. Diversified franchise, record earnings, forward P/E of 16.4x is not stretched, and it has the strongest balance sheet to absorb a capex slowdown. If I'm long banks in this cycle, JPM is the one I want the most.
• Goldman Sachs is the high-conviction, high-beta play. The investment banking and trading tailwinds are strongest here, but the forward valuation at 18.8x means less margin of safety. The position size should reflect that.
• Citigroup at 14.0x forward P/E and 1.06x book is the value entry point, though its investment banking franchise is smaller than JPM's or Goldman's. The turn from Brian Moynihan's leadership is still mid-cycle.
• Morgan Stanley at 22.5x forward P/E is the most expensive. The wealth management book provides some diversification, but the valuation assumes AI-driven deal flow stays elevated.

The break condition is clear: if hyperscaler capex guidance starts decelerating — if the sequential growth rate of AI spending drops materially below the 50-70% range we've seen — the bank tailwind evaporates fastest in investment banking and equities trading. Goldman and Morgan Stanley would be the first to feel it. JPM would cushion the blow through its consumer and commercial banking operations.

The debate isn't whether banks are benefiting from the AI cycle. They are, unequivocally. The debate is whether the concentration of that benefit — across revenue lines, across the same counterparty group, across the same macro assumption — is still a reason to build position or trim it.

In my opinion, the answer depends on your time horizon. For a three- to five-year window, I believe the AI financing cycle has enough runway to justify bank exposure. But much of the return profile is likely back-half weighted, and the risk is front-loaded. Size positions accordingly.

The closing point

This cycle is producing record bank profits right now. That's a fact, not a prediction. The $114 billion in first-half capital markets revenue, the 55% investment banking surge at Goldman, the 86% jump in JPMorgan's equities trading — none of that is disputed.

But the BIS warning, the circular financing web, the 94% capex-to-cash-flow ratio approaching the limit of internal funding — these signals are also real. They don't mean the party is over. They mean the leverage is rising.

The smartest move isn't to bet against the AI cycle or to assume it runs forever. It's to recognize which bank franchise gives you the best exposure to the upside while limiting the damage when the cycle turns. JPMorgan, at a forward P/E of 16.4x, is that franchise. The pure-play investment banks are higher reward and higher risk. The value names offer margin of safety but less tailwind participation.

Demand remains robust. The risk is rising. Size accordingly.