In theory, what would happen to the US economy, writ large, if we were to wake up and find out hyperscaler AI ROI or AI-linked revenue projections no longer support the bullish CapEx narrative? Despite having now received Anthropic’s S-1, the jury remains out.
Admittedly, this is merely one of the many themes gripping the market today as it attempts to divine the sustainability of thematic AI momentum. Other causes for consternation include the risk that Anthropic’s IPO is poorly bought or traded. Period. We could get a year’s worth of bad or inconclusive conference calls out of Anthropic. We could see an escalating series of NeoCloud corporate failures, de-facto admitting to ourselves that the entire AI monetization thesis will run through maybe 5-7 hegemonic companies. Apollo research recently divulged flat revenue productivity per worker amongst the S&P 493, asserting that productivity has yet to disseminate into the market beyond technology companies. Erego, broader AI use-cases could prove to be a narrower opportunity set than originally conceived. CoreWeave spreads could widen, cutting off its access to the capital markets. CoreWeave acts as the pricing stip for all non-rated, private-company data center leases. Such an outcome would strip the national datacenter development market of the ability to finance a datacenter built on the back of private or non-rated corporate company credit. This assumes circular financing and chip-company credit backstops dry up, which is already happening. As a matter of fact, each additional dollar of CapEx spent in attainment of super productive AI, each marginal data center lease obligation signed, serves to further lever us to these risk factors. The path to everything working out as originally planned is becoming increasingly obscured. Our point in this piece is to attempt to think through whether a 75th percentile outcome is actually a wholesale bad thing. Because, despite the risks, such an outcome may not prove to be disastrous.
In summary, what happens if datapoints turn, hyperscalers pull back on CapEx and overall animal spirits retreat from cautiously optimistic to outright concerned? What is the risk to the US economy?
A few things to consider. First, if you’ve read our previously published missives with any measure of consistency, you’ll know we think about the economy as a bottom’s up, consumption driven machine run by individuals and households. Second, US households are currently experiencing a wealth effect like never before due, primarily, to the proliferation of US equity market ownership across the American socioeconomic stratification. Third, real wages are flat to negative at present. Finally, nominal savings in checking in accounts are healthy, healthier than ever before – excluding idiosyncratic data coming out of Covid due to stimulus checks. However, all-told, the current real earnings and savings environment is insufficient to support present household consumption trends alone. Americans need dividend and equity capital markets liquidity to support 4%+ consumption readings in quarterly GDP. If the value of household investments drop, it follows that the majority will revert to well-established cash savings trends of 7%-10% of income, up from about 4% today (FRED). The big question being define drop? We can’t exactly, but our suspicion is it won’t take a full bear market. We are confident a pullback in US equity market performance, particularly one that metastasizes into pessimistic forward expectations, would produce a meaningful slowdown in total consumer spending.
Thus, what we’ve come to realize is that US consumer spending is tied to public equity prices like never before. Consider this: US household holdings of equities and mutual funds reached a record 278% of disposable personal income in the second quarter of 2026, up from just 29% in 1982, according to BCA Research. Similarly, Gallup polls put equity market ownership at approximately 65% of Americans. We’ve referenced and seen polls which reflect household exposure to US equities in excess of 70% all US households. While BofA checking account data shows a savings surplus in consumer accounts about 25% - 30% above inflation-adjusted pre-covid levels, we’re talking about a relatively thin nominal cash cushion compared to the wealth impact from S&P 500 ownership, which has provided a 12% - 13% 5-year CAGR for those invested.
Beyond a drag on consumer spending, what additional effects would our economy experience following a change in AI fortunes? Our view is the answer is plenty, but not all at once and, in all likelihood, not quickly. First, it is unlikely the aggregate capital spent on the “AI buildout” over the coming two years will change materially in almost any scenario. Total long-term CapEx spend estimates may suffer under a revised AI efficacy regime, but short-to-medium term spend actually pushed through the US economy will likely not. We can say this with some degree of confidence because much of the capital to be spent on infrastructure over the next two to three years is already committed under agreements with penal termination clauses. These credit-enhanced, usually airtight agreements are what backstop the construction of large real-assets, including datacenters and energy generation facilities. Public equity markets can reprice overnight after catching the flu from negative AI sentiment, the reality is they usually recover. Long-term capital commitments into real assets, on the other hand, need to be made years in advance and are generally adhered to unless the capital backstopping those agreements goes bankrupt. Most datacenters actually being constructed today include some sort of financeable, bankable credit. According to JLL, 95% of the 66GW of data centers under construction are pre-leased. The majority of those executed leases (percentage unknown, but rationally it’s probably most of them) are backed by investment grade credit of some sort. So, this gives us confidence that the CapEx component of GDP growth will remain firm for years to come.
Second, it is unlikely the market pulls the handbrake on AI productivity estimates overnight. Markets will gyrate around sentiment and news until they rationalize a sustainable capital spend level based upon some blend of observed and projected return on AI investment. Our view is this is likely to be a long rationalization process. If Anthropic really can earn $100B+ a year as a frontier model developer, if Goole and Meta can continue to leverage their existing businesses to efficiently monetize capital spend, if OpenAI can actually follow the Anthropic listing with a productive listing of its own…if, if, if…our experience has been that powerful stakeholders, such as the aforementioned entities, can weave a compelling narrative counter to any emerging strain of negative sentiment. In this theoretical case, the market starts to turn bearish on AI, stakeholders push back with new products and creative ideas, and the total impact is blunted over time. Truth is, this exact process is playing out already – which is a great thing. The AI thesis has been pummeled by competition risks (see DeepSeek selloff), litigation risks (see HuggingFace event), CapEx risks (see CoreWeave CDS widening selloff), existing business cannibalization risks (see Alphabet, Meta, Microsoft earnings), etc. Generally, companies have maintained guidance and pushed new products to market which substantiate their competitive position in this broader narrative. Thus, when we evaluate the forward risks listed above, we can’t look at them in isolation. It’s likely these massive stakeholder companies will have a say in how the market digests negative news or data. Actually, it’s more than likely, you can guarantee it. Our point is, we don’t think AI “guppy-like death” happens quickly or soon.
Our final thought on the matter is to remind our readers that an entire industry earning 50%+ profit margins is unsustainable and thankfully so. Chip makers earning 75%+ margins, LLM’s creating 85%+ margins, GE Vernova gross margin has nearly doubled since 2023, Corning performance, Caterpillar, etc. Neither the US economy nor the S&P 500 require 30% YoY aggregate earnings growth to retain disproportionate access to global capital. We have no idea how much revenue needs to be earned by AI companies to not tank the US economy. We all would love to have that bogey. But even if AI related revenue falls short of its lofty expectations, we’re comforted by the fact that nearly all members of the physical AI supply chain can crunch margins and remain strongly profitable. Cost to build drops, future hurdle AI revenue to create ROI drops and the market resets.
Will the above outcome blow the top off the S&P? Probably. But, again, this all must be taken in context. AI projected revenues drop, CapEx estimates pull back, supply chain members reduce margins to maintain marketshare, companies spin narratives around future revenue growth and, ideally, the market survives the rebalancing effort. Stronger confidence around future assumptions could push forward multiples up to compensate for lower growth. We admittedly overuse this example, but Tesla is a great barometer for how US public equity investors treat “hope” themes. This isn’t a perfect comparison, but Elon’s ability to spin narratives and bring new theoretical products to market in order to produce an unjustified value is a Harvard case study for both AI company CEO’s and investors, alike. There are likely to be short-term, one-to-three month periods of pain as this whole process plays out. It happens all the time in Tesla stock. But importantly, we’ve actually experienced and survived versions of these macro events in the public markets over the last 18 months, notably associated with the aforementioned AI related selloff events. And the market survived, allowing consumers to spend and GDP to grow.
In short, we’re starting to lean more into the thesis that any material thematic AI reversion will be met with a sharp equity selloff, followed by a recovery and ultimately a gradual slowdown in associated spending. Basically, that’s a slow puncture event, not a blown tire.