The U.S. equity market is witnessing a once-in-a-decade euphoric run since it briefly dipped in March 2026. As the major indices are stretched further to the upside day after day, so is the frequency and the loudness of a specific alarm bell from media sources — Berkshire Hathaway’s growing cash pile, currently sitting on $397 billion in cash and short-term Treasuries, the largest in the company’s history. The logic runs like this: the most celebrated investment institution in history can’t find anything worth buying, therefore the market must be overvalued, therefore a correction is coming. It is hard to falsify this narrative because the market is indeed very extended and a pullback is unlikely to surprise any pundits. The question is: does Berkshire’s cash pile have anything to do with the current market, and hence its credibility in foretelling the headwinds.
Everyone is watching the wrong number.
Berkshire’s cash pile is not a forecast. It is a side effect. Berkshire’s operating businesses — insurance, railroads, energy, manufacturing — generate roughly $40 to $45 billion per year in operating earnings. Unless the company actively deploys that cash into acquisitions or stock purchases, it accumulates by default. Cash hitting new records is not a signal; it is the natural state of a compounding machine that has run out of targets large enough to absorb its output. Berkshire needs to deploy at scale to matter. Its three largest acquisitions — Burlington Northern at $44 billion, Precision Castparts at $37 billion, Allegheny at $12 billion — confirm the magnitude required. Berkshire does hold smaller positions, some under $500 million, but a $500 million equity bet moves $397 billion of cash by 0.1 percent. The cash pile is not a choice to hold cash. It is the residual of a machine that generates faster than it can deploy. When the targets large enough to absorb that output are expensive — and large-cap US quality has been expensive for most of the last decade — cash builds. That is arithmetic, not prophecy.
The track record does not hold up.
The narrative survives because we remember the hits and forget the misses. Cash was elevated before the 2008 financial crisis and before the 2020 COVID crash. Those two episodes have been cited so frequently that they have become financial folklore. What gets mentioned less: cash grew from $77 billion to $116 billion across 2013 to 2017 while the S&P 500 rallied more than 80%. It has been at or near record highs continuously since late 2023, during which the S&P 500 returned approximately 70% in three years — one of the strongest stretches in modern market history. If you had gone defensive every time Berkshire’s cash balance made headlines, you would have missed more rallies than you avoided drawdowns.
Contrast this with signals that do carry lead time. Bond prices, for instance, have historically peaked one to two years before equity market tops — a structural lead driven by the interest rate cycle. That signal has a causal mechanism (tightening liquidity) and a consistent lead time. Berkshire’s cash level has neither. It fires at random intervals, for reasons that may or may not relate to the market, and offers no mechanism connecting the cash build to a subsequent equity decline.
There is a subtler issue. Cash as a percentage of Berkshire’s total assets tells a less dramatic story than the absolute number. Berkshire itself has grown enormously — its market capitalisation now exceeds $1 trillion. The cash position relative to the enterprise is elevated but not historically extreme. Reporting the absolute figure without adjusting for scale is the financial equivalent of noting that GDP is at an all-time high and concluding that the economy has never been this good.
The near $400 billion question.
There is a more uncomfortable reading of the cash pile that the financial press rarely explores: what if it is not evidence of prescience, but the receipt for the largest missed opportunity in modern investing history?
Since early 2023, the dominant investment theme has been AI infrastructure. It was visible, articulable, and investable from the moment ChatGPT made the demand signal unmistakable. NVIDIA was trading at 25 times forward earnings in January 2023 — not expensive by any historical tech standard. The hyperscalers — Microsoft, Alphabet, Amazon, Meta — were among the highest-quality balance sheets on earth, deploying hundreds of billions into a technology their own earnings confirmed was driving real revenue growth. This was not a speculative meme. It was the defining capital deployment cycle of the decade, and the evidence was in every quarterly report Berkshire’s own analysts could read.
Berkshire largely sat it out. Zero shares of NVIDIA — the centrepiece of the entire cycle. Microsoft was never owned — a longstanding principled avoidance citing conflict of interest. No Meta, no AMD. Amazon was held but cut by 77% in late 2025 to a negligible $525 million position. To Berkshire’s credit, a $4.3 billion stake in Alphabet was initiated in Q3 2025 — but by then Alphabet’s share price had already more than doubled from its early 2023 level, and the position represents less than 2% of the portfolio. It was a toe in the water, not a conviction bet. Meanwhile, the Apple position — once over 40% of the equity portfolio and Berkshire’s largest source of indirect AI exposure — has been sold down by 76% over ten quarters, from 915 million shares to roughly 217 million, just as Apple Intelligence launched. What remains is railroads, insurance, Coca-Cola, and Chevron.
In the meantime, NVIDIA returned over 800%. Meta returned over 400%. Alphabet more than doubled before Berkshire bought its first share. A simple equal-weighted basket of the four largest AI beneficiaries, bought in early 2023 with just a third of Berkshire’s current cash pile, would have generated more than $350 billion in gains — roughly equal to the entire cash reserve that is supposedly keeping the company safe.
The opportunity cost is permanent.
The nature of opportunity cost is that it does not reverse. A dollar that sat in Treasuries at 4.5% from 2023 to 2026 earned roughly 14 cents in cumulative interest. The same dollar in NVIDIA earned over eight. The delta between those two outcomes — the $350 billion in foregone returns — is not a paper loss that recovers in the next cycle. It is a compounding gap that widens every year. The gap between those two trajectories is not closed by waiting for a correction. It is closed only by a market meltdown so severe that it erases years of compounding — and even then only brings you back to where you would have already been.
The stated justification for holding cash is dry powder for when opportunities arise. But the opportunity did arise. It simply did not arrive in the form Berkshire’s pattern-matching was calibrated to recognise. It was not a distressed bank in a panic. It was not a high-quality industrial trading below book value. It was the companies selling the pickaxes during the largest gold rush in corporate history — a structure Berkshire understood perfectly when it bought Coca-Cola for its brand moat and Apple for its ecosystem moat. The AI infrastructure moat — NVIDIA’s CUDA dominance, hyperscaler cloud lock-in, enterprise switching costs — is structurally identical. The playbook was there. It was not run.
For any other investor — a fund manager, an allocator, a family office — sitting on over 35% cash for two years and not investing in AI while the market’s dominant theme generates returns of 100 to 800 percent would be a career-ending performance gap. Opportunity cost at this scale is not a footnote. It is the story.
This is not 1999.
The comparison to 1999 is the strongest card in Berkshire’s defence, and it is being played vigorously — most recently at the May 2026 annual meeting, and in commentary that frames the AI boom as a repeat of the dotcom bubble waiting to burst. But the parallel is weaker than it appears. In 1999, the technology companies driving the market had no earnings — the boom was built on projections. In 2026, NVIDIA alone generates over $40 billion in quarterly profit — more than the combined annual earnings of Microsoft, Intel, and Cisco at the dotcom peak. The hyperscalers have committed $700 billion in capital expenditure backed by cloud revenue growing 28 to 63 percent. The question is not whether the technology is real — it is whether the spending is sustainable. That is a capex cycle question (which we examined in The AI Capex Machine), not a valuation question, and Berkshire’s cash pile answers neither. Sitting out of dotcom was vindicated because the thesis was wrong. Sitting out of AI infrastructure will only be vindicated if this thesis is also wrong — and so far, every earnings report says otherwise.
Circle of competence or circle of comfort?
“We don’t invest in what we don’t understand” is admirable humility when applied to biotech or cryptocurrency. Applied to cloud computing and artificial intelligence in 2023 to 2026 — technologies now used by virtually every business Berkshire owns — it starts to look less like discipline and more like inertia. GEICO is deploying AI for claims processing. BNSF is using machine learning for freight logistics. Berkshire Hathaway Energy is integrating AI into grid management. The technology is not foreign to the portfolio. It is running inside it. What Berkshire chose not to invest in is the infrastructure layer that makes all of it possible.
The honest question is at what point does “circle of competence” become “circle of comfort”? Competence can be expanded. Comfort resists expansion by definition. The distinction matters because one is a strategy and the other is a limitation. When the limitation costs $350 billion in foregone returns that do not compound back, it is no longer a philosophical preference. It is a material impairment to shareholder value.
A stopped clock is right twice a day.
Strip away the market-timing narrative, strip away the reverence, and the cash pile has one honest message: Berkshire did not invest in AI. Whether that proves to be late-stage patience or institutional rigidity depends entirely on what happens next — and neither the cash balance nor anyone watching it can tell you which. The Buffett Indicator — total market capitalisation divided by GDP — has been structurally broken for over a decade by globalisation, the shift to asset-light business models, and the dominance of high-multiple technology companies. It has rarely been championed by Berkshire since the original 2001 endorsement. The Berkshire cash pile is its spiritual successor: a metric that captures a real observation while offering zero actionable value. Both belong in the category of indicators that are permanently right about the level and offer no reliable lead on timing. That they happened to coincide with the 2000 and 2020 crashes does not make them timing tools — a stopped clock is right twice a day, and a permanently elevated indicator will inevitably overlap with every correction. The question is whether it told you anything you could act on before the fact. It did not.
The hardest sentence in market analysis is also the most honest one: we don’t know what the market will do next, and neither does Berkshire’s cash balance.
Note on data
Berkshire Hathaway financial data is sourced from SEC 13F filings (Q4 2025, filed February 2026) and the Q1 2026 10-Q (filed May 2026). Cash and Treasury holdings of $397.4 billion reflect Q1 2026 as reported. Apple share counts reflect the Q4 2025 13F with Q1 2026 estimates from portfolio tracking services; the Q1 2026 13F is due by May 15, 2026. Alphabet position data from Q4 2025 13F disclosure. AI stock returns are calculated from January 2023 to April 2026 using adjusted closing prices. Market performance references use the S&P 500 Total Return Index. This article does not constitute investment advice.