Why the AI Bubble Trade May Be Rotating Into Gold Miners

John Templeton identified euphoria as the final stage of every bull market. Multiple independent frameworks now flag the AI equity trade as approaching that phase, with leading AI companies carrying price-to-sales multiples in the 25-35x range while major hyperscalers collectively project roughly $700 billion in data-centre buildouts for 2026 alone. At the opposite end of the spectrum, precious metals miners represent well under 1% of the S&P 500 by index weight, sitting in the same pessimism phase that preceded some of the largest commodity equity runs in history. The divergence between these two trades has rarely been wider, and the institutional mechanics that eventually force capital rotation from one to the other are already identifiable. This analysis maps the late-cycle signals accumulating in AI equities against the early-cycle setup in mining stocks, explains the benchmark pressure that will eventually compel institutional rebalancing, and outlines what individual stock selection looks like at this stage of the cycle.

Why the AI trade now shows the fingerprints of a bubble

Ruchir Sharma identified four markers that appear in every speculative excess: over-investment, over-valuation, over-ownership, and over-leverage. Applied to the current AI equity complex, all four are present.

Sharma’s four bubble markers applied to AI equities: Over-investment, over-valuation, over-ownership, and over-leverage. Each is independently observable in the current AI trade.

• Over-investment: Alphabet, Amazon, Meta, and Microsoft are collectively projecting roughly $700 billion in data-centre capital expenditure for 2026, a figure that materially outpaces near-term AI revenue contributions. Major technology companies are directing all or most free cash flow toward these programmes.
• Over-valuation: Leading AI company price-to-sales multiples sit in the 25-35x range versus approximately 3x for the broader market, a stretch that depends increasingly on narrative and expectations rather than demonstrated cash flow.
• Over-ownership: Concentration in a handful of AI-linked mega-cap names has reached levels where their contribution to S&P 500 returns dominates index performance.
• Over-leverage: Capital commitments at this scale create structural rigidity; once the spending is committed, reversing course becomes a writedown problem rather than a strategic choice.

The capex intensity is the most structurally alarming indicator. The parallel to fibre optic infrastructure spending in 2000 is instructive: the technology proved ultimately useful, but the equity destruction among the companies that built it was near-total. The bubble diagnosis does not require AI technology to fail. It requires only that the capital deployed to capture AI revenues exceeds the revenues themselves for long enough to compress returns.

Gold and silver miners sit at the opposite end of the cycle

Precious metals miners occupy the structural position that historically precedes the largest commodity equity moves: under-owned, under-covered, and unrewarded for positive results.

The quantitative underweighting tells half the story. The behavioural evidence tells the rest.

Newmont reported strong Q2 2025 earnings on 24 July 2025. The stock opened lower the following day. In a favoured sector, strong results from a bellwether trigger broad buying. In a neglected sector, positive news is faded. This is the behavioural signature of a sector that generalist investors are not yet watching.

The GDX ETF (gold miners) set a record high before declining roughly 40% from peak, typical of the violent early-cycle volatility that characterises thinly owned sectors. The SILJ (junior silver miners ETF) reached a capitulation low in February 2024. Bob Thompson of Raymond James Vancouver places the mining sector at approximately the 7 o’clock position on the Mining Clock framework, indicating meaningful runway remains before cycle completion.

Index weights below 1% combined with non-confirmation of positive news represent the quantitative and behavioural signatures of a sector at the pessimism phase, precisely where the most asymmetric long-term entries historically occur.

The Templeton clock and where AI equities sit on it

John Templeton’s market cycle framework provides a diagnostic lens that applies to both equity and commodity markets. The four phases progress in sequence:

1. Pessimism: Bull markets are born here. Ownership is minimal, sentiment is hostile, and positive news fails to move prices.
2. Skepticism: Early believers take positions, but the majority remains doubtful. Performance begins to improve but is dismissed as temporary.
3. Optimism: Broad participation begins. Narratives gain credibility, capital flows accelerate, and the trade becomes consensus.
4. Euphoria: The final phase. All participants are convinced the trend will continue. Skeptics are ridiculed. Risk is highest precisely when confidence peaks.

Locating the AI trade on this framework requires examining behavioural signals rather than valuation metrics alone. The most reliable indicator is how the market treats credible skeptics.

When Dissenting Voices Are Mocked: A Late-Cycle Behavioural Indicator

The public ridicule of credible long-term cautious voices is a behavioural, not valuation, signal. It occurs because euphoria requires consensus, and consensus requires discrediting dissent.

Warren Buffett’s refusal to invest in technology stocks around 1999 was publicly ridiculed for six to eight months before vindication. The pattern repeats at cycle tops: prominent cautious voices are publicly mocked, their track records questioned, and their frameworks dismissed as outdated. This is not a coincidence of personality; it is a structural feature of how euphoria sustains itself.

The AI equity trade is exhibiting this pattern. Respected market observers with decades of experience are being dismissed in financial media as out of touch.

The framework is not a precise timing instrument. Greenwood and Guo argue that AI remains in the early stages of a potential bubble, noting that valuations are elevated but have not reached the extremes of the late-1990s dot-com period. Fidelity’s research, as of early 2026, treated AI as potential bubble territory but did not identify all classic late-cycle warnings at the index level. The honest reading is that multiple indicators point to bubble-like conditions and rising late-cycle risk, even if the exact phase remains genuinely debated. What the framework does identify with confidence is asymmetry: the closer a trade moves toward euphoria, the worse the risk-reward ratio for new capital entering.

How institutional benchmark pressure eventually forces the rotation

The rotation from AI equities into mining stocks is not a question of whether institutional capital eventually moves. It is a question of what conditions trigger the sequence. The mechanics are specific and observable.

1. AI underperformance begins. The contribution to S&P 500 returns from top AI names and their suppliers (chips, data centres, infrastructure) starts to fade. This does not require a crash; a period of sideways performance or modest underperformance is sufficient.
2. Mining outperformance creates benchmark drag. As precious metals equities outperform over several quarters, the underweight position in miners becomes a measurable source of relative performance drag for managers benchmarked to broad indices.
3. Institutional buying begins regardless of sector enthusiasm. Portfolio managers do not need to become enthusiastic about gold miners. They need only to recognise that underweighting the sector is costing them performance relative to their benchmark. The buying that follows is mechanical, not conviction-driven.

This is not theoretical. Eric Sprott’s hedge fund, launched around 2001-2002, employed precisely this strategy: long gold miners, short technology and financial stocks. The trade worked because the rotation mechanics were structural, not dependent on sentiment shifting first.

The key rotation signal to monitor is whether the contribution to S&P 500 returns by top AI names begins to fade while energy, infrastructure, and commodity equities gain. That divergence is the trigger condition. Everything else follows from benchmark arithmetic.

What individual stock selection looks like at this stage of the cycle

Early-cycle sector positioning does not eliminate stock selection risk. Even in deeply under-owned sectors, individual outcomes diverge dramatically. The historical case studies illustrate the return distribution available.

These returns originated in the same structural conditions that exist today: a sector with minimal generalist ownership, suppressed valuations, and non-confirmation of positive news. Kirkland Lake Gold rose approximately 500% during the difficult 2017-2018 period before acquisition by Agnico Eagle. Great Bear Resources moved from approximately $1.50 to a $29 acquisition. Snowline Gold appreciated from below $1 to approximately $15 with further upside speculation.

Framing the portfolio trade correctly

The actionable positioning is a relative trade, not a binary crash call. Lower AI and growth exposure paired with higher miners and energy exposure captures the rotation without requiring a specific prediction about when AI equities peak.

The genuine risks are real: miners can remain cheap or get cheaper before rotation occurs. Timing is unknowable. Basis risk applies to any relative value framing. The cycle can extend further than any framework predicts. Framing the trade as relative value, reducing late-cycle exposure while increasing early-cycle exposure, is consistent with institutional best practice and intellectually honest about what can and cannot be known.

The asymmetry that makes this rotation worth taking seriously now

The two-sided asymmetry is the core of the thesis. AI equities carry late-cycle downside risk from multiples of 25-35x price-to-sales with limited upside from current levels. Mining equities carry early-cycle upside potential from structurally depressed starting weights well below 1% of the S&P 500. The rotation does not require a crash call on AI. Even a normalisation of AI multiples toward the broader market’s approximately 3x price-to-sales would represent capital displacement at scale into other sectors.

Three signals indicate the rotation is beginning:

• AI names’ contribution to S&P 500 index returns begins fading over consecutive quarters
• Mining sector outperformance sustains over multiple quarters rather than reversing quickly
• Generalist fund inflow data into GDX or GDXJ shifts from persistent outflows to measurable inflows

When the downside case for one trade and the upside case for another share the same trigger, the relative trade becomes unusually compelling. The trigger here is AI performance leadership fading; the beneficiary is a sector that represents less than 1% of the benchmark it could grow into.

The mining sector’s benchmark weight below 1% means that even a return to historical norms represents substantial capital reallocation in absolute dollar terms, given the trillions of dollars benchmarked to the S&P 500.

Capital rotation is not a prediction, it is a positioning framework

Multiple late-cycle signals in AI equities and structural early-cycle positioning in precious metals miners create an asymmetric opportunity for capital rotation. That is the core thesis, and it is supported by independent lines of evidence on both sides of the trade.

The central risk is equally clear: the cycle can extend further than any framework predicts, and miners can underperform for longer than the evidence suggests they should. Conviction about the direction of travel does not translate into precision about timing.

The actionable framework is specific. Reduce late-cycle AI and growth exposure. Increase early-cycle miners and energy exposure. Monitor the three rotation signals, AI contribution fading, sustained mining outperformance, and generalist institutional inflows, over rolling quarters. Adjust positioning as the evidence accumulates rather than front-loading a single trade on a timing prediction.

The framework does not require being right about when the AI trade peaks. It requires only that the asymmetry between the two sides of the trade is correctly identified, and that the positioning adjusts as confirmation arrives.

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.

—