In the span of just one quarter, the market has asked investors to believe two opposing narratives. Ninety days ago, the story was one of energy shock, inflation fears, and SaaS disruption (among others). The VIX, Wall Street’s so-called fear gauge, read 25 – nearly 30% above its historical average.1 The popular prediction market Polymarket pegged the probability of a recession by the end of 2026 at ~36%.2 By the close of the quarter, the VIX read 16, or ~16%, below the average. Enthusiasm around the demand for compute (e.g., memory chips and semiconductors) along with a forecasted AI-driven earnings super cycle had propelled the S&P to an all-time high. The probability of recession had fallen to ~12%. The pendulum had swung.
This is not, in itself, a criticism of either narrative. Energy shocks are real, as is the demand for compute. But a pendulum swinging from one extreme to the other rarely indicates which extreme was correct. More often, it’s the result of investor sentiment moving considerably faster than the underlying business fundamentals, often in an attempt to forecast the future before market prices have a chance to catch up. Over the past 90 days, the demand for AI compute has not dramatically changed. What changed, in large part, was the story investors told themselves – particularly surrounding the duration of that demand.
We raised a version of this question in our March letter, although our focus then was oil rather than silicon. We asked, in effect: “Is $100 per barrel the new normal, or are we experiencing a temporary spike destined to fade back toward $75?” The answer mattered enormously to the value of an oil company. A durable, decade-long repricing of oil implies a vastly different intrinsic value than a one-year spike followed by reversion.
The same question is worth asking today about the demand for compute. Is the current pace of AI-related capital expenditure – the kind propelling memory and semiconductor names to some of their best quarters on record – a genuine, decades-long shift in the normal level of demand for chips and data center capacity? Or does it include some measure of the same dynamic that has whipsawed semiconductor cycles before: customers over-ordering out of fear of scarcity, pulling forward years of demand into a handful of quarters?
We do not pretend to know the answer with precision. Our crystal ball, as we’ve said before, gets blurry well before the horizon these questions require. That admission, though, doesn’t leave us helpless. We don’t need to forecast the future with absolute precision to ask whether AI-related names are selling at a significant discount or premium to intrinsic value under a reasonable set of assumptions.
Consider the following test: take a set of reasonable assumptions and work backward to the revenue an industry would need to generate to justify what’s currently being spent on it. If that amount of revenue looks plausible, then the spending looks justified. If it doesn’t, then the spending may be running ahead of the opportunity. We at Harris Associates have conducted this test. (Spoiler alert: the annual required revenue needed to satisfy even a conservative set of assumptions is astounding.)
Doing so required answering a sequential set of questions that each in their own right had a range of plausible answers. How much capital expenditure will be deployed in the coming years? How much of that capital expenditure is used to expand compute (i.e., not to serve existing customers in existing businesses)? How much is in excess of depreciation? What will the return on invested capital be? What will the tax rate be? What will the EBIT margin be?
As you can imagine, it is quite difficult (if not impossible) to provide an accurate point estimate to any one of these questions with a high degree of confidence – much less each of them, concurrently. And we aren’t the only ones struggling.
Goldman Sachs predicts AI capital expenditure will be ~$765B this year, but cautiously notes, “The scale of required investment for the AI build-out is itself more uncertain than commonly assumed. Estimates rest on a number of assumptions that, if changed, can significantly increase or decrease the amount of capital required.”3 JPMorgan estimates 2026 capital expenditure of ~$650B from the hyperscalers alone.4 Bank of America’s estimate tops $800B.5
Considering that this wide range of estimates applies to a calendar year we are already halfway through, it comes as no surprise that estimates disperse even further the farther out predictions extend. All for only the first of several required questions…
The important takeaway is that the accuracy of any conclusion quickly erodes as you begin to stack wide-ranging assumption on top of wide-ranging assumption, in an effort to work backward towards your final answer.
This is what we refer to as a wide range of outcomes. Can we enter a plausible set of assumptions that make even the most speculative AI investment look like a good investment? Absolutely. Just as easily, we could create a set of assumptions (again, that are plausible) that suggest the very same investment is vastly overvalued. The difference between us and the market is in the confidence around the range of assumptions. At present, the market seems quite confident that an aggressive set of estimates is accurate.
We want to be careful about what conclusion to draw from this. It would be easy to overreach in either direction. A wide range of outcomes is not, by itself, evidence that AI infrastructure spending is unjustified. Some assumptions on the optimistic end of the range are entirely plausible. Businesses that generate durable pricing power, genuine scarcity, or contracted long-term revenue streams could comfortably clear even the higher end of what’s required. Our point is narrower and, in some ways, more useful: a wide range of plausible outcomes should be met with a price that reflects that width. A stock priced for the narrow, optimistic slice of outcomes leaves very little room for error. That is a different failure than simply being wrong about the direction of a business. It is a failure to rigorously assess the uncertainty of one’s own analysis.
This is precisely the discipline we described in our March letter, applied at a larger scale. There, we distinguished between a durable repricing of oil and a temporary spike destined to fade. Here, the same distinction applies to an entire industry’s capital cycle, layered across many compounding assumptions instead of one commodity price. In both cases, our job is to understand the direction and width of the range well enough to recognize when an asset has not only been mispriced, but by enough to compensate for any error in our forecasts.
The furthest distance a pendulum swings to one side is known as the amplitude. The greater the amplitude, the wider the swing back – the same energy that carried it out must carry it back. Markets are not physics, and sentiment is not conserved as energy is. But they do behave similarly and for similar reasons: confidence in estimates that have a wide range of plausible answers, once set in motion, tends to carry prices past the point that fundamentals alone would justify – both in fear, and, just as reliably, in enthusiasm. Fear compresses prices below value, euphoria inflates them above it. Both are the pendulum doing what pendulums do. Neither is a signal to be brave or fearful so much as it is a signal to check, carefully, which side of fair value the market has carried us to and to act accordingly, in either direction.
We do not know when, or if, this particular pendulum will swing back. We do know that the size of a swing is itself a measure of risk, independent of its direction. A market pricing in only the optimistic arc of that swing is a market that has, whether knowingly or not, assumed the pendulum has stopped moving. Our job has never been to predict when the pendulum stops. It has been, and remains, to calculate what a business is actually worth across a realistic range of outcomes and to stay disciplined enough not to let the pendulum’s current position – wherever it happens to be swinging – lure us into paying for only its most favorable arc.
As always, we thank you for entrusting us with your investment assets and your continued support. Lastly, the best compliment we can receive is a referral from a satisfied client. We appreciate your referrals and handle them with the utmost care.
1 Source: FactSet
2 https://polymarket.com/event/us-recession-by-end-of-2026
3 https://www.goldmansachs.com/pdfs/insights/articles/tracking-trillions-the-assumptions-shaping-scale-of-the-ai-build-out/GS_Apr_GSGI_Tracking_Trillions.pdf
4 https://fortune.com/2026/06/25/what-bubble-jpmorgan-5-5-trillion-ai-capex-explosion-profitable-for-now/
5 https://www.cnbc.com/2026/04/30/ai-boom-big-tech-capital-expenditures-now-seen-topping-1-trillion-in-2027-.html
Past performance is no guarantee of future results. Current performance may be lower or higher than the performance data quoted. The investment return and principal value vary so that an investor’s shares when redeemed may be worth more or less than the original cost.
The specific securities identified and described in this report do not represent all the securities purchased, sold, or recommended to advisory clients. There is no assurance that any securities discussed herein will remain in an account’s portfolio at the time one receives this report or that securities sold have not been repurchased. It should not be assumed that any of the securities, transactions, or holdings discussed herein were or will prove to be profitable.
This material is not intended to be a recommendation or investment advice, does not constitute a solicitation to buy, sell or hold a security or an investment strategy, and is not provided in a fiduciary capacity. The information provided does not take into account the specific objectives or circumstances of any particular investor or suggest any specific course of action. Investment decisions should be made based on an investor’s objectives and circumstances and in consultation with his or her advisors.
The information, data, analyses, and opinions presented herein (including current investment themes, the portfolio managers’ research and investment process, and portfolio characteristics) are for informational purposes only and represent the investments and views of the portfolio managers and Harris Associates L.P. as of the date written and are subject to change without notice. This content is not a recommendation of or an offer to buy or sell a security and is not warranted to be correct, complete or accurate. Certain comments herein are based on current expectations and are considered “forward-looking statements”. These forward looking statements reflect assumptions and analyses made by the portfolio managers and Harris Associates L.P. based on their experience and perception of historical trends, current conditions, expected future developments, and other factors they believe are relevant. Actual future results are subject to a number of investment and other risks and may prove to be different from expectations. Readers are cautioned not to place undue reliance on the forward-looking statements.
The S&P 500 Index is a float-adjusted, capitalization-weighted index of 500 U.S. large-capitalization stocks representing all major industries. It is a widely recognized index of broad, U.S. equity market performance. Returns reflect the reinvestment of dividends. This index is unmanaged and investors cannot invest directly in this index.
All information provided is as of 06/30/2026 unless otherwise specified.
