AI Chip Valuations Now Rival the Mississippi Bubble

A Bloomberg survey of 32 investment managers across three continents finds 80% bullish on equities, yet elevated Treasury yields are the most commonly flagged threat to the very AI stocks driving those gains. The contradiction sharpens with each upward revision to hyperscaler capital expenditure and each tick higher in the 10-year yield.

On the eve of NVIDIA’s most consequential earnings release in years, two competing narratives have converged. Alphabet, Microsoft, Amazon, and Meta are committing a combined $650-$725 billion in AI infrastructure capex for 2026, a figure that dwarfs any prior coordinated corporate capital programme. Simultaneously, institutional short sellers, historical analogues, and a 10-year Treasury yield sitting at 4.66-4.68% are generating a credible counter-case that the equity market has already priced in outcomes that have yet to be earned. What follows is a framework for holding both narratives simultaneously and knowing which signals to watch.

The signals short sellers and strategists are reading right now

The most concrete bearish position disclosed this month belongs to Leopold Aschenbrenner’s Situational Awareness fund. A Q1 13F filing revealed put options against ten entities spanning the full semiconductor supply chain: ASML, AMD, Broadcom, Corning, Intel, Micron, NVIDIA, Oracle, Taiwan Semiconductor, and the VanEck Semiconductor ETF. The breadth is the signal. This is not a single-name short thesis; it is a bet against the entire value chain.

Bank of America’s Michael Hartnett widened the frame to the macro level, flagging four concurrent warning signals:

• Extreme price momentum in a narrow set of AI-linked leaders
• Elevated market capitalisation concentration in those same names
• Compressed volatility despite stretched valuations
• Equity-to-bond yield pressure as stocks compete with a 4.66-4.68% risk-free rate

Hartnett’s four-signal cluster: extreme price momentum, elevated concentration, compressed volatility, and equity pressure on bond yields, all firing simultaneously.

The Sohn Investment Conference on 14 May 2026 brought multiple practitioners to the same bearish theme on the same day. That convergence, rather than any single participant’s position, is what matters analytically. Stocks identified as exhibiting parabolic price behaviour since late March include Micron, AMD, SK Hynix, Marvell, and Intel. When named practitioners with detailed sector expertise converge on a directional view at the same moment, the pattern warrants monitoring regardless of whether the positions prove profitable immediately.

Paul Tudor Jones’s runway estimate of one to two remaining years for the AI bull market, offered on 6 May 2026 and drawing a direct parallel to 1999, sits alongside the Aschenbrenner and Hartnett positions as a third institutional data point framing the timeline rather than disputing the technology’s validity.

How the Mississippi Bubble comparison actually works, and where it breaks down

The historical pattern is well documented. Transformative technologies generate genuine long-run economic value while simultaneously producing speculative excess in publicly traded proxies that arrives years before actual returns materialise. Railroads reshaped global commerce; most investors who bought railway equities at the peak of 1840s mania lost money. Radio created an entirely new industry; many early listed-radio stocks failed within a decade. The internet changed the structure of the global economy; the dot-com Nasdaq still took 15 years to recover its March 2000 peak in nominal terms.

CNBC-cited commentary has claimed that AI chip valuations, by at least one concentration metric, now rival the French Mississippi Bubble and exceed dot-com Nasdaq concentrations. Concentration as a metric measures how much of a market’s total capitalisation is held by a small number of names. When concentration is extreme, the index becomes a proxy for those few stocks rather than a diversified equity exposure.

Joyce Meng of Fact Capital, speaking at the Sohn Conference, framed the AI analogue through the railroad and internet lens: many early participants ultimately failed despite the technology’s success. The analytically precise risk is not that AI fails as a technology. It is that current equity prices leave no room for the technology to succeed while shareholders still lose money.

Where the AI case differs from a clean mania

One genuine structural difference separates AI semiconductor stocks from the speculative vehicles of prior bubbles. Hyperscaler capex commitments of $650-$725 billion for 2026 represent contracted, multi-year demand, not speculative future revenue. Technology adoption risk appears low given the scale of spending already committed. Valuation risk, however, remains high given the sensitivity of these multiples to interest rate movements. The demand is real; the question is whether the prices already reflect it.

What $650 billion in hyperscaler commitments actually proves, and what it does not

Combined 2026 hyperscaler capex guidance stands at $650-$725 billion, up from approximately $380-$410 billion in 2025, representing one of the largest coordinated capital programmes in corporate history.

The scale of the commitment deserves honest treatment. This is not hypothetical demand. Four companies have publicly revised their spending upward, each independently.

Alphabet’s CFO Ruth Porat stated that capex would be “notably higher in 2025 than in 2024,” driven by investments in technical infrastructure “particularly for AI.” The 2026 revisions have since pushed that trajectory steeper still.

NVIDIA’s Q1 FY2027 consensus reflects this demand: approximately $78-$79 billion in total revenue, roughly $73 billion from the data centre segment alone, and adjusted earnings per share of $1.75-$1.77. Those figures imply approximately 77-80% year-over-year revenue growth.

The analytical limitation is precise. Demand-side spending validates near-term revenue. It does not validate equity multiples priced on earnings growth extending many years into the future. A company can report record revenue every quarter while its stock declines if the market decides the multiple is too rich for the rate environment. The capex data settles the demand question. It does not settle the valuation question.

Inference cost economics represent a second-order risk sitting beneath the capex headline figures: if the cost of running generative AI applications at scale remains structurally unprofitable for hyperscalers, the commercial justification for sustaining $650-$725 billion in annual hardware spending weakens regardless of near-term demand visibility.

Why rising yields make the valuation math asymmetric

The 10-year US Treasury yield closed at 4.66-4.68% on 19 May 2026, according to FRED data. That level sits well above the near-zero rate environment that enabled the initial expansion of growth stock multiples in 2020-2021.

The discount rate mechanism is straightforward. When analysts value a high-growth company, they project cash flows far into the future and discount them back to present value. The higher the discount rate, the less those distant cash flows are worth today. For a company like NVIDIA, where a significant portion of the theoretical intrinsic value depends on earnings projected five, seven, or ten years ahead, a higher yield compresses the present value of those future earnings more severely than it compresses the value of a company whose earnings are front-loaded.

Goldman Sachs strategist Dominic Wilson has described the market environment as “increasingly complex,” with overly aggressive expectations potentially creating a valuation overhang. JPMorgan strategists have warned that elevated real yields reduce the equity risk premium, making richly valued growth stocks more vulnerable to re-rating. Morgan Stanley strategists have flagged similar sensitivity among AI-linked leaders if yields remain elevated or rise further.

The synchronised sovereign bond selloff extending across the US, UK, and Japan in May 2026 reinforces why the 10-year yield headwind is not a US-specific phenomenon; when the term premium rises across multiple sovereign markets simultaneously, the discount rate pressure on long-duration growth stocks becomes harder to offset through earnings upgrades alone.

The Bloomberg survey captures the tension precisely: 80% of 32 investment managers across the US, Asia, and Europe maintain a bullish outlook, yet elevated bond yields are flagged as the primary rising risk. Near-term earnings upgrades have so far offset the theoretical multiple compression from rates. That offset, however, depends on a specific condition continuing.

The asymmetric downside scenario

The asymmetry becomes visible under two conditions occurring simultaneously rather than in sequence:

1. Earnings growth disappoints or forward guidance softens, removing the primary justification for current multiples.
2. The rate headwind, which was previously offset by earnings upgrades, compounds the disappointment by compressing the multiple at the same moment revenue expectations decline.

Under this scenario, drawdown velocity exceeds what either factor would produce alone. This is a risk scenario, not a prediction, and its probability depends on the three variables tracked throughout this analysis. The distinction matters: the risk is not that rates are high, but that high rates remove the margin of safety if the earnings narrative falters.

What NVIDIA’s earnings will and will not resolve

NVIDIA reports Q1 FY2027 earnings on 20 May 2026. The consensus bar is high: approximately $78-$79 billion in total revenue, roughly $73 billion from the data centre, and adjusted EPS of $1.75-$1.77.

Consensus implies approximately 77-80% year-over-year revenue growth for NVIDIA’s Q1 FY2027, a figure that captures the scale of expectations already embedded in the stock price.

What earnings can confirm:

• Whether near-term demand is tracking to consensus
• Whether forward guidance is consistent with the hyperscaler capex narrative
• Whether Blackwell architecture uptake is meeting internal targets

What earnings cannot resolve:

• Whether current valuation multiples are sustainable over a three-to-five-year horizon
• Whether hyperscaler returns on AI capex will justify continued spending at this rate
• Whether concentration risk in the sector has been adequately priced

The guidance commentary will carry more analytical weight than the reported quarter itself. Management’s language around demand visibility, order backlog duration, and Blackwell ramp trajectory will be the market’s forward signal. The backward-looking revenue figure, even if it beats consensus, is already partially embedded in the stock price.

One timing detail is analytically notable. The Aschenbrenner puts disclosure was filed on 19 May 2026, the day before NVIDIA’s earnings. Institutional bearish positioning formally entered the public record at the last possible moment before the most anticipated data point of the quarter.

The framework investors need before the sector re-rates

Three variables will determine whether AI chip valuations sustain or compress from here:

1. Earnings growth trajectory relative to consensus. If NVIDIA and peers continue to beat and raise, the multiple remains defensible. If guidance softens, the primary support for current valuations weakens.
2. Ten-year yield direction. At 4.66-4.68%, rates are already a headwind. A move higher intensifies multiple compression. A move lower provides relief.
3. Hyperscaler capex revision trajectory. The difference between upward revisions (the current trend) and plateauing guidance is the difference between sustained demand visibility and a cycle approaching its peak.

If all three variables remain supportive, current multiples are defensible but not compelling at these yield levels. If any one deteriorates, the absence of a multiple cushion creates rapid re-rating risk. Hartnett’s four-signal cluster (price momentum, concentration, volatility compression, equity-to-bond pressure) provides a real-time monitoring framework for when the balance shifts.

By at least one concentration metric, AI chip market capitalisation now rivals the Mississippi Bubble and exceeds the dot-com Nasdaq peak. That does not mean a crash is imminent. It means the margin for error is narrow.

Technology winner does not equal equity winner

Investors who bought railroad equities at peak 1840s valuations largely lost money even as railroads reshaped the global economy within a generation. The technology succeeded; the equity speculation did not. The question for AI equity investors is whether they are buying the technology or buying the speculation around it. The answer depends on the price they pay.

For investors wanting a structured foundation before tracking the three monitoring variables, our dedicated guide to AI bubble valuation frameworks applies the Shiller CAPE ratio, Minsky financing stages, and index concentration data to place the current AI cycle among the most extreme valuation episodes in 155 years of market history.

The verdict will take longer than one earnings call to write

The weight of evidence, from concentration metrics and yield headwinds to institutional short positioning and historical analogues, justifies heightened scrutiny of AI chip valuations. The strength of demand fundamentals, visible in $650-$725 billion of hyperscaler capex commitments and NVIDIA’s earnings trajectory, means the bubble thesis is not yet confirmed. Both observations are simultaneously true.

The three monitoring variables remain the practical takeaway: earnings growth relative to consensus, yield direction, and capex revision trajectory. These are the signals that will determine which narrative prevails over the coming quarters.

NVIDIA’s 20 May earnings and guidance language are the next data point, not the final answer. Readers tracking this question should focus on management commentary around Blackwell demand visibility and order backlog duration as the most forward-looking signal available before the next valuation inflection.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors.