Macro Pulse for November 2: Is AI in a Bubble? Where AI Has Room to Run, and Where It May Hit a Wall

• Though the Fed delivered the expected 25bp interest rate cut at its October 29^th meeting and announced the end of its liquidity-draining quantitative tightening (QT) program for December 1, our read of the meeting was decidedly hawkish. Chair Powell emphasized that a further cut in December “is not a foregone conclusion,” in line with our base case; market expectations baking in this cut have had to adjust. We expect 50-75bps of further easing by year-end 2026, barring meaningful upside risks to inflation.
• The U.S.-China trade deal lowered U.S. tariffs on Chinese imports (from 57% to 47%) and lifted Chinese bans on rare earths exports and purchases of U.S. soybeans. We await the U.S. Supreme Court ruling (oral arguments begin November 5) on constitutional tariff authority before drawing conclusions on the nature of tariff policy going into 2026.
• As AI-related Q3 corporate earnings roll in, the market is rewarding strong capex and companies that find a fast path to monetization of cloud services. We discuss these dynamics in detail in today’s note. 

Last year, I was on the course. This year, I’m reflecting on the top question on investors’ minds: Has AI gone out too fast? Is it going to hit a wall at mile 18?

Is AI in a bubble?

This question is daunting because it’s about much more than just equity price action. A “bubble” can apply to so many aspects of how the AI investment thesis is developing, along the centers of gravity illustrated in chart 1:

1. Is AI simply too expensive? Debate about market concentration, price action, valuation, and earnings quality hinges on an investor’s view of the following questions.
2. Is the AI sphere building more than it can handle? AI capacity has seen a rapid buildout, fueled by extensive capex. How this capex is financed, including via corporate cash, as well as both public and private debt and equity, depends on the cost of capital, related to but distinct from capital markets conditions.
3. Are there enough resources to sustain the great hope of AI? From investor demand to electricity and rare earths, AI faces resource constraints. Efforts to address these resource constraints are creating competition.

There’s no doubt that AI valuations are high, and that the Magnificent 7’s market concentration (alongside other AI-related equities) is, too. But we still believe the AI thesis has room to run. We are far from reaching overcapacity in AI infrastructure, and we very much stand by our 2024 view that AI is a megatrend influencing the next 10 years of investing.

However: AI investment has happened at breakneck speed, and may hit some bumps in the next 24 months or so. Two peripheral dynamics are moving more central to the AI thesis – the likely “canaries in the coal mine” if sentiment becomes stretched, capital markets conditions shift, or capex intentions falter. First is the rise of debt financing in the AI sphere; second, competition is fueling speculation in companies along the AI supply chain.

Is AI simply too expensive?

The top defense of gravity-defying price action in AI is the strength of corporate earnings: even up against spectacular valuation re-rating (chart 2), earnings growth has consistently been, is currently, and is expected to continue to be the primary driver of price performance among Magnificent 7 hyperscalers (chart 3). 

The pushback we’ve had in countless conversations with both retail and institutional investors in the U.S., Europe, and Asia does not question these earnings – more on this in a moment – but whether investors should pay quite so much for these earnings.

Comparisons to the dot-com era are imperfect, but since it is the most direct comparison we have: AI looks cheap relative to dot-com (chart 4), but may face similar hiccups around rapid capacity creation (chart 5 – and see section 3 for our views). 

Our take: Our (market-weight) neutral conviction in U.S. large cap equities implies we’re only marginal buyers at today’s valuations. Market momentum around AI, which we see supported by robust economic activity, policy tailwinds, strong sentiment, and ample liquidity, could reverse at any time. But as long as earnings expectations and capex remain healthy, driven by the two factors we discuss below – we’d buy that dip. 

Is the AI sphere building more than it can handle?

Let’s return to the topic of earnings. As we discussed in our 2024 Megatrends report, the key players have not fully figured out how to monetize generative AI models. Should companies pay per unit of compute power they use? Will consumers be willing to pay for a more intelligent version of a search engine? How will energy costs, which fluctuate based on power supply and region, factor into this pricing?

Though cloud-related revenues are growing, how to monetize foundational AI models is still a moving target. In the meantime, hyperscalers are, on aggregate, recycling cash flow from the profitable arms of their businesses into AI.

This use of corporate cash is a cornerstone of why the AI thesis has looked so healthy: hyperscalers are cashflow-generative enough to conduct share buybacks, issue dividends, and invest in AI capex (to the tune of 60% of operating cash flow) all at once (chart 6). In turn, hyperscalers and investors alike hope – but haven’t proven – AI capex will drive additional revenue growth in the future. 

Enter: debt financing. Per Project Finance, data center debt financings doubled from roughly $30B in 2024 to $60B in 2025. Total estimated data center financing needs (including cash, debt, equity) measure in the hundreds of billions on an annual basis, and in the trillions between now and 2030. Though the share of debt financing in AI is a drop in the bucket thus far, it’s become something of a red flag, perhaps implying that companies no longer think investing in AI is the best use of their cash, or that they have overextended their cash-based investment commitments.  

Our take: U.S. hyperscalers clearly have more than enough cash to sustain their current pace of investment, meaning the shift toward debt financing is a capital structure choice, not forced by cash constraints. On its own, debt financing of AI capex could be a complete non-issue, as long as capex doesn’t take over operating cash flow.

Our concern, however, is that debt has the power to turn an interest rate-insensitive market force into the opposite. Think about it two ways: 1) Since we expect ample interest rate volatility, duration is not where we want to take risk. Leverage pushes AI capex upward from a base of no-to-low duration. 2) At a systemic level, rising leverage increases the risk that any sort of AI-related blowup becomes systemic. Whether triggered by AI qualms or private credit jitters, a financial “accident” is not our base case; overall leverage levels are still healthy, and slowing liquidity drains (the Fed’s QT program ends December 1^st) ease concerns about market triggers. But there is a premium on quality within every leveraged asset class, and AI may be in too early innings for quality to be readily assessed.

Are there enough resources to sustain the great hope of AI?

According to Paul Krugman, large AI players are spending $370B on data centers in 2025, and $1.5T in utility power investment over next 5 years. But this growth in supply of AI infrastructure is being met with just as much demand (chart 7); data center vacancy rates are low, and the construction boom in data centers is just now starting to normalize.

Our take: AI is still at a stage of development in which overcapacity isn’t the near-term concern; under-capacity is. The most common under-capacity concern – and we agree – is that electricity and other resource supply will not be able to keep pace with demand, driving prices for everyone higher (discussed in investment implications below). Even more important, in our view, is a different lens: undercapacity is fueling competition, which is fueling speculation. While we see select examples of this trend across the AI supply chain, competition is fiercest – driving valuations that are resoundingly speculative – among power generation companies, particularly in alternative and nuclear power. While compliance concerns won’t allow us to put specific companies on the spot, we’ll call out a few examples of the froth: a zero-revenue nuclear power company with no licenses or contracts, valued at $20B+; and a data center REIT that expects no revenue until 2027, which received a $15B valuation at IPO.

As we discussed in our 2023 megatrends report, the energy supply chain faces constraints. GenAI has only accelerated this perspective. And though we stand by our view that nuclear energy will need to be a part of the solution, we raise our eyebrows at the extent of valuation run-up in companies that aren’t yet providing power (or power commitments) (or earnings).

What should investors make of all this?

Ensure allocation is aligned to your level of belief in the long-term AI thesis. We are too early in AI development for the “winners and losers” to be decided. But we know the largest players have the greatest chance of being among those winners, because they are able to develop AI models and tools in-house, compete for scarce resources (from investor interest to electric capacity), and acquire capabilities they lack themselves (inorganic growth).

We also believe that higher power costs and data center and chip investment at scale are some of many tailwinds behind inflation in the medium term. In a moderate inflation environment, we like a commodity satellite, perhaps 10% of a portfolio as a hedge for upside inflation surprises.

For the keen eye that noticed the potential “Gold 2.0” bubble in this chart, stay tuned for our views on gold in the coming weeks.

Where to deploy capital today should align to your view on possible near-term hiccups:

Keep on keeping on in U.S. large cap growth. We expect earnings and free cash flow to continue supporting AI capex. This thesis will change when and if operating cash flow falters, capex intentions falter, or debt financing begins to exceed that of cash.

We still believe in broader access points in AI-related infrastructure: electric utilities, materials, energy, data center-related real estate. Where these providers are expanding capacity to meet real demand, we’re all in. Where providers do not yet have capacity or revenue, moments of stellar returns will be met with stellar volatility.