Everything About AI
NextFin News - The artificial-intelligence boom has produced the clearest schism on Wall Street in a generation: the executives financing the build-out are celebrating it, while the money managers who must price it are warning that it will break. At the Hong Kong Monetary Authority's Global Financial Leaders' Investment Summit on Tuesday, some of the world's most powerful bank chiefs and asset managers laid out two irreconcilable views of the same trade, and the gap between them is now wide enough to matter for every investor holding AI-linked assets.
Pacific Investment Management Co. Chief Executive Emmanuel Roman warned that AI-related financing could grow to about $10 trillion over the next five years and eventually become excessive. His warning carried a specific prediction about what happens when leverage outruns cash flow:
"At some point in time, there's going to be a downturn, and then things are going to break."
Roman cited rising U.S. interest rates, war in the Middle East, or a euro-zone crisis as potential triggers. Bain Capital chair John Connaughton has joined a growing chorus of investment managers flagging bubble risks after the sector's breakneck expansion.
On the other side, bank bosses and private-market chiefs continue to describe demand for capital to fund AI infrastructure as insatiable. The dispute is no longer academic. It sits directly on top of a market that has already begun to wobble: 47 AI and chip stocks have shed roughly $5 trillion in market value since early October, even as the S&P 500 trades near record territory after a run that took it to a year-to-date high of 7,844.52.
The Two Camps at the Hong Kong Summit
The split running through Hong Kong was not a subtle difference of degree. It was a disagreement about the fundamental nature of the AI investment boom: is this a durable industrial revolution, or a credit-fueled mania that will end in a familiar bust?
The bearish camp is led by the asset managers who will be left holding the losses if the boom turns. Roman's $10 trillion figure is the most concrete number on the table. It represents financing — debt, equity, leases, and structured credit — flowing into AI infrastructure over the next five years, not revenue and not profits. His point is structural: when financing grows faster than the cash flows that must service it, something breaks. The trigger does not matter. It could be rates, a war, or a currency crisis; the vulnerability is built into the leverage itself.
Connaughton, whose Bain Capital sits at the center of private-equity dealmaking, has echoed the warning. The significance of a private-market veteran joining a bond-market veteran like Roman is that the skepticism has crossed asset classes. This is not just a fixed-income concern about duration and credit spreads; it is a broader concern about capital allocation at scale.
The bullish camp is led by the institutions financing the physical build-out. Stephen Schwarzman, chairman and CEO of Blackstone, acknowledged the risk of excess before dismissing it for his own book. "When you have an investment boom like we're seeing today, there's no surprise that there will be some misallocations of capital," Schwarzman said, before adding:
"But when I think about what we do in AI infrastructure, I have a tremendous amount of conviction and confidence."
His defense rests on counterparty quality: almost everything Blackstone finances in AI infrastructure is tied to hyperscalers with average market capitalizations of $2.5 trillion, signing contracts lasting 15 to 20 years.
That is the crux of the disagreement. The bulls see contracts with the largest companies in the world. The bears see those same contracts as the transmission belt for a much larger mistake.
The Money Behind the Boom
To judge who is right, the first question is how much money is actually being committed. The scale is now large enough that small errors in assumption compound into very large errors in outcome.
The four largest hyperscalers — Meta Platforms, Microsoft, Amazon, and Alphabet — collectively plan to allocate about $725 billion to capital expenditures in 2026, up roughly 77% from the $410 billion spent in 2025. Amazon has guided to approximately $200 billion, Alphabet to between $175 billion and $185 billion, Meta to a range of $115 billion to $135 billion, and Microsoft's fiscal-year run rate points to roughly $145 billion. Goldman Sachs expects the four to spend a combined $5.3 trillion on capex from fiscal 2025 through fiscal 2030.
A broader estimate puts six hyperscalers' capex at $1.3 trillion in 2027, up from a projected $870 billion this year and $470 billion in 2025. The same analysis found that only Microsoft is forecast to generate positive free cash flow next year. That is a striking detail: the companies with the strongest balance sheets in corporate America are collectively moving toward negative free cash flow to fund a technology whose revenue payoff remains unproven at this scale.
Moody's has flagged the credit consequences directly. Rating analysts warned that the capex plans could "threaten credit quality" at Alphabet, Microsoft, Amazon, Meta, Oracle, and CoreWeave as they make a costly transition from asset-light software models to asset-heavy infrastructure models requiring "unprecedented levels of investment and capital raising." Moody's estimates direct debt at the six companies at roughly $460 billion, and noted that additional debt financing "could lead to increasing leverage and additional risk being borne on their balance sheets."
The financing is not confined to investment-grade balance sheets. Nvidia, which controls an estimated 85% or more of the market for AI graphics processors, has partnered with six Wall Street firms to raise more than $500 billion in outside capital to fund data-center build-outs. The structure lets Nvidia collect revenue on GPUs without carrying the assets on its own balance sheet — a form of leverage migration that looks efficient in good times and opaque in bad ones.
Private credit has become the other hidden pillar. The industry has grown into roughly a $2 trillion market, with about $1.5 trillion expected to come from private-credit funds. Outstanding loans to AI companies from these funds have surpassed $200 billion, mostly to mid-sized borrowers carrying debt at five to seven times earnings, with no public credit ratings for most of the loans. The Financial Stability Board, the global regulatory watchdog, estimated that banks have around $220 billion of drawn and undrawn credit lines to private-credit funds — and noted that commercial estimates suggest the true figure could be more than twice as large. The industry, one analyst put it, has never been tested during a real economic downturn at its current size.
There are already cracks. CoreWeave's unsecured bonds due in 2032 traded below 80 cents on the dollar, yielding more than 15%, a level widely considered the threshold for financial distress. Oracle's $165 billion Project Jupiter data center is mired in permitting problems; the company invoked force majeure after a key energy permit was denied, and the debt financing the project trades below 90 cents on the dollar. A 15% yield is not a rounding error; it is the market pricing a real probability of distress.
Why the Bulls Think They Are Safe
The bull case is not built on denial. It is built on a specific claim about risk distribution. Schwarzman's argument — contracts with $2.5 trillion market-cap counterparties, locked in for 15 to 20 years — is the cleanest version of it. If the tenants are the largest companies in the world, and the contracts are long-dated, then the infrastructure owner is insulated from a demand shock. The hyperscaler absorbs the risk, not the financier.
There is a second pillar to the bullish view: this is not the first automation wave, and employment fears have consistently overstated the damage. Standard Chartered CEO Bill Winters offered a data point from his own bank when asked about AI's evolution at the summit. "There's a fear factor," he said. He noted that the lender employs roughly 100,000 people, about the same as when he joined 11 years ago, then described what actually happened to those roles:
"We automated away 25,000 jobs or so along the way, and we replaced those roles with other roles."
The third pillar is the banking system's own experience. Goldman Sachs and JPMorgan Chase have reported strong earnings while executives described demand for capital to fund AI infrastructure as insatiable. For bank CEOs, the AI boom is not a speculative bet; it is a fee-generating reality. Trading, advisory, and financing activity around AI deals has been a direct contributor to profits.
Finally, the optimists argue that today's mood is not euphoria. Some strategists have described the market as priced for perfection rather than bubbly, noting that the largest tech companies remain strong on their balance sheets and that investor sentiment, while optimistic, has not reached the manic levels that typically mark a top.
The Cyclical Versus Structural Call
Here is the judgment the market must make, and the two camps are answering it differently. Is the AI boom a cyclical investment wave that will mean-revert, or a structural regime shift that will not?
The evidence for a structural shift is real and should not be dismissed. General-purpose technologies — electricity, the internal combustion engine, the internet — do reorganize the economy, and the capital intensity of the transition can last a decade. AI has the hallmarks: a new production function, broad applicability across industries, and a capacity to substitute for cognitive labor, not just manual labor. Winters' employment data from Standard Chartered is consistent with this read: automation destroyed 25,000 roles and the headcount did not fall, because new roles absorbed the labor. That is the historical pattern of a structural shift.
But a structural technology can still produce a cyclical financial bubble. This is the distinction the market is blurring. The railroads were a structural revolution in the 19th century; the railway mania of the 1840s still ended in a crash that wiped out most of the listed railway companies. The internet was a structural revolution; the Nasdaq still fell 78% from its 2000 peak. The dot-com bust did not disprove the internet; it repriced the claims on the internet's cash flows.
Three pieces of evidence point to a cyclical financial overlay on top of the structural technology shift. First, the financing is growing faster than the revenue: $10 trillion of financing over five years against a revenue base that analysts still struggle to forecast with confidence. Second, the credit structure is deteriorating as the boom extends — from investment-grade hyperscaler balance sheets down to unrated private-credit borrowers at five to seven times earnings, and into bonds yielding 15%. That is the classic late-cycle pattern of capital chasing the same trade into lower-quality corners. Third, the market has already shown it can reverse violently: a $5 trillion loss in AI and chip stock market value within weeks, even while the broader index held near records, is not the signature of a steady structural repricing. It is the signature of a crowded trade.
The correct call, therefore, is both: AI is a structural technological shift, but the financing of AI infrastructure has entered a cyclical boom that will mean-revert. The technology will not go away. Many of the claims on the technology's cash flows will be repriced, possibly severely.
The Second-Order Question the Market Is Not Asking
The first-order debate — bubble or not — is already priced into the volatility. The second-order question is more important: what happens to the banking system when the AI credit cycle turns?
The bulls' defense rests on the assumption that risk stays with the hyperscalers. But the financing chain has already migrated. Banks are exposed through direct lending, through the private-credit funds they advise and finance, and through the revolving credit lines that roughly half of all private-credit borrowers also draw on. The Financial Stability Board's estimate of around $220 billion in drawn and undrawn exposure is a floor, not a ceiling, because it does not fully capture the indirect exposure through funds and facilities.
If AI infrastructure demand disappoints, the first losses will not appear at the hyperscalers. They will appear in the unrated private-credit book and in the single-name infrastructure debt — the CoreWeaves and the project-finance vehicles — where a 15% yield already signals stress. Only after that will the damage travel up the chain to the banks that stand behind the funds. This is the same transmission pattern as the 2008 crisis, where the first visible losses were in subprime mortgages long before they were on bank balance sheets.
There is also a cross-asset transmission that the market has not fully digested. A disorderly AI repricing would hit the exact segment of the market that has carried the S&P 500. Strategists estimate that roughly half of the index's $57 trillion market capitalization has significant or moderate exposure to AI. A 30% decline in AI-exposed equities would therefore translate into a roughly 15% decline in the index even if the other half held flat. The concentration that made the rally so strong is the same concentration that makes the downside so efficient.
This is why Roman's statement matters more than the specific trigger he named. He did not say AI would fail. He said a downturn would reveal what breaks. In a highly leveraged, highly concentrated, opacity-heavy credit structure, the answer is usually the part nobody was watching.
The Strongest Counter-Thesis
The most serious argument against the bubble thesis does not come from a cheerleader. It comes from the structure of the counterparties themselves. The hyperscalers are not 1999-era dot-coms burning venture capital. They are mature, cash-generating companies with pricing power, and they are signing 15-to-20-year contracts before building. If demand for compute grows even at a moderate pace, the infrastructure is fully utilized and the financiers are paid. The misallocation, in this view, is concentrated in a few speculative names like CoreWeave, not in the core of the trade.
This counter-thesis has real force, and it is why a blanket "sell everything AI" is not the conclusion. But it has two vulnerabilities. First, it assumes the hyperscalers' own revenue from AI grows fast enough to justify $725 billion in one year of capex, rising to $1.3 trillion. If AI revenue grows but not at that pace, the hyperscalers absorb the loss — and their stock prices, which embed the growth assumption, fall with it. Second, it assumes the private-credit and project-finance tail stays small. It is already more than $200 billion and growing into a $1.5 trillion private-credit allocation. A tail that grows sevenfold is no longer a tail.
The falsifying signal for the bubble thesis is specific and observable. If the four largest hyperscalers report, over the next two quarters, that AI-related revenue is growing at or above the pace of their AI capex — and if CoreWeave's bonds recover above 90 cents on the dollar while private-credit default rates on AI loans stay below 2% — then the financing is being absorbed by real demand, the credit structure is sound, and the bubble thesis is wrong. Conversely, if hyperscaler AI revenue growth decelerates below capex growth for two consecutive quarters while sub-investment-grade AI credit spreads widen by more than 300 basis points, the cyclical boom has turned and the downside scenario is in motion.
What to Watch and What It Means
The split between Wall Street's CEOs and its money managers is unlikely to resolve quickly, and investors should expect volatility to remain elevated until the revenue data catches up with the capital commitments. The watch list is concrete: quarterly AI revenue disclosure from the hyperscalers, the direction of CoreWeave and similar single-name AI credit, private-credit default rates — Moody's latest reading put the 12-month rate at 4%, rising to 5.8% when payment-in-kind amendments and maturity extensions are counted as distressed — and any sign that banks are tightening underwriting standards for AI-related lending.
In the short term, sentiment and liquidity will dominate. The market has already shown it can sell off $5 trillion in AI-linked value within weeks while the broader index holds near records. That divergence can persist, and it can reverse just as fast on a single strong earnings print.
In the medium term, fundamentals will decide. The hyperscalers must convert capex into cash flow. If they do, the skeptics look like they missed a revolution. If they do not, the credit losses will travel up the chain exactly as Roman described.
In the long term, the structural call stands regardless of the cyclical outcome. AI will reorganize work, and companies that adopt it will gain productivity. But owning the technology and owning the securities that claim its cash flows are different propositions. The first can be true while the second loses money.
The base case is a bifurcated outcome: the hyperscalers survive and consolidate, the speculative periphery defaults or resets at lower valuations, and the financing chain takes losses that are manageable for the system but painful for specific holders. The upside case is that AI revenue accelerates faster than capex, spreads tighten, and today's skeptics are remembered as the investors who called every boom a bubble. The downside case is that revenue disappoints, private credit cracks, and bank exposure proves larger than the disclosed estimates.
Wall Street's CEOs are being paid to build the boom. Its money managers are being paid to survive it. When those two incentives diverge this sharply, the prudent assumption is that the boom is real — and so is the bill that comes after.
Explore more exclusive insights at nextfin.ai.
