Rising Yields, Record AI Capex, and What Breaks First

The 10-year Treasury yield has climbed from a 2020 low of roughly 0.57% to nearly 4.80% today, and the companies spending the most on artificial intelligence are still borrowing hundreds of billions to build the infrastructure behind it. That pairing, record borrowing costs colliding with record capital spending, has no historical precedent to lean on.

The AI capital expenditure cycle is generating enormous real-economy demand for labour, copper, and industrial materials, while depending on debt financing in the most expensive rate environment in more than a decade.

For anyone holding mining, energy, or commodity exposure, this produces two stories at once: a genuinely durable demand narrative and a fragility risk. Most coverage keeps them apart. They are not separate.

What sits below untangles the four interlocking forces driving this dynamic, giving you a grounded basis for judging which commodity exposures hold up as rates rise, and which market risks deserve more weight than current valuations imply.

Why rising long-term yields are the variable that changes everything

Start with the number that anchors everything else. The 10-year Treasury yield now sits near 4.80%, up from a 2020 trough of about 0.57%.

The distance already travelled The 10-year yield has moved from 0.57% in 2020 to roughly 4.79-4.80% today. The often-cited 6% scenario is a shorter jump than the one markets have already made.

That framing matters because a move to 6% is not a tail-risk fantasy. It has identifiable drivers, including persistent fiscal deficits and the financing demand created by AI infrastructure itself. Every hyperscaler that borrows to build a data centre adds to the supply of debt the market must absorb, which pushes yields the same direction the companies are already exposed to.

The yield has not risen in isolation; three forces behind the surge include persistent fiscal deficits, elevated oil prices feeding inflation expectations, and the sheer volume of Treasury supply required to fund both government spending and private infrastructure borrowing.

The pressure does not stay confined to Treasuries. It spreads through the entire cost of capital.

• The 10-year Treasury yield stands near 4.80%, up from 0.57% in 2020, with a credible path toward 6%.
• Corporate credit reflects the same tightening: the ICE BofA US Corporate Index effective yield is around 5.53%, and the ICE BofA US High Yield Index has reached 7.15%.
• Mortgage rates above 7% imply roughly $70,000 in annual interest on a $1 million home, a concrete measure of how far household balance sheets are stretched.

That mortgage figure is the tell. When the cost of financing a home eats $70,000 a year in interest alone, the broader economy’s sensitivity to rates stops being abstract.

Here is the read you should take from this. If you are treating 4.80% as the ceiling, you are making an assumption the data does not support, because the remaining distance to 6% is smaller than the ground already covered since 2020. The yield environment is the single variable that decides whether AI-driven commodity demand stays a durable thesis or gets interrupted by a credit-driven repricing. It deserves to be settled in your thinking before you add sector exposure, not after.

How $690 billion in AI capex is remaking commodity markets

The demand reshaping commodity markets does not begin with copper. It begins with a spending figure large enough to bend physical supply chains.

Aggregate 2026 capital expenditure guidance for the top five hyperscalers has escalated into the $600-750 billion range, with 70-75% of that spending explicitly tied to AI infrastructure. Across Microsoft, Meta, Amazon, and Alphabet alone, combined capex in just the first half of 2026 reached roughly $301 billion.

Break it down by company and the scale becomes concrete.

What this tells you is that the demand source is not a forecast. It is already committed capital, already flowing into concrete, steel, power, and metal.

From data centres to copper demand: the material math

The spending logic leads directly to copper. Research indicates AI data centres require 27-33 tonnes of copper per megawatt of capacity, well above the 21 tonnes per megawatt demanded by crypto mining facilities. That intensity gap is the mechanism: as capacity scales, copper draw scales faster than earlier infrastructure cycles would predict.

Three separate forecasts point the same direction, which matters because a demand thesis backed by one analyst is a bet, while a demand thesis backed by three independent institutions is a signal.

The 27-33 tonnes-per-megawatt intensity figure anchors the bull case, but copper demand forecasts from institutions including BHP, S&P Global, and Bloomberg Intelligence differ on timeline and magnitude, and reconciling those differences is where the investment thesis becomes precise rather than directional.

• BHP estimates AI data-centre copper demand will rise six-fold, from roughly 500,000 tonnes in 2024 to 3 million tonnes by 2050, around 9% of global copper demand.
• S&P Global Market Intelligence forecasts data-centre copper usage climbing from 1.1 million tonnes in 2025 to 2.5 million tonnes by 2040.
• Bloomberg Intelligence projects generative-AI spending driving a 3% annual increase in North American copper demand through 2035.

Copper currently trades in the $6.50-$6.70 per pound range, equivalent to roughly $13,000 per metric ton. That price is the market’s present read on this demand thesis.

For anyone holding copper exposure, the important distinction is this: the demand signal is structural, not cyclical. It does not require economic growth to beat expectations. It requires only that the AI buildout continues at its current pace, which the committed capex above suggests it will.

The overinvestment cycle already forming inside the boom

The strongest evidence that this boom carries a correction inside it comes from the spenders’ own financial statements.

Hyperscaler capex as a percentage of revenue has crossed 22%, far above the historical norm of 11-16%. That ratio is the internal metric that flags overextension, and it is drawn from the companies’ own numbers, not an outside critic’s estimate.

The clearest single marker Capex-to-revenue at 22% against a historical norm of 11-16% means these firms are spending at a pace their revenue base has not previously supported.

The dot-com parallel is not decoration. The mechanism is specific: infrastructure spending that overshoots near-term demand creates future capacity that compresses returns, which eventually forces a pullback. Research firm Futuriom forecasts a 20-30% capex pullback in 2026, invoking exactly this pattern.

The AI investment cycle moves through distinct phases, from infrastructure buildout through platform consolidation to application monetisation, and the commodity demand thesis is concentrated in the infrastructure phase, which makes the duration and pace of that phase the key variable for mining and energy investors.

The scale keeps rising even as the warning signs accumulate. Aggregate cash capex for the top five hyperscalers is expected to exceed $690 billion in FY2026 and reach $900 billion by FY2028, much of it reliant on external borrowing, which loops straight back to the rate sensitivity from the first section.

Play the cycle forward and the sequence has a recognisable shape.

• Boom: capex surges past revenue growth, funded increasingly by debt.
• Oversupply: data-centre and semiconductor capacity overshoots near-term demand.
• Deflationary bust: excess capacity compresses returns and triggers a correction.
• Central bank expansion: policymakers respond with aggressive monetary easing.
• Commodity reflation: loose policy restarts inflationary pressure, concentrated in materials.

Here is what this changes for you. The overinvestment signal is already visible in the capex-to-revenue ratios, so waiting for the bust to confirm the cycle means acting on lagging information. The copper demand thesis can be durable over a decade while the path there still runs through a correction, and that correction could hand mining and energy investors materially better entry points than today’s prices offer.

Where financial market fragility meets the AI trade

The link between equity valuations and bond yields is not a vague worry. It is a measurable threshold, and markets are already behaving as the theory predicts.

Once nominal yields push above a certain level, the discount rate applied to future earnings starts competing directly with the return available from risk-free Treasuries. That is the point at which equities decouple from their growth stories.

Goldman Sachs analysis on rising yields and growth stocks identifies the compression of equity risk premiums as the primary transmission mechanism, reinforcing why the yield threshold data in the table above carries practical weight for investors holding long-duration tech positions.

The behaviour is already surfacing. Broadcom reported strong earnings that failed to lift its share price, an early sign of diminishing market responsiveness. Meta shares fell after the company raised its 2026 guidance, despite strong underlying performance, as investors fixated on the scale of its AI spending and the strain on free cash flow.

Professional investors have named the same fears. A recent Bank of America fund-manager survey ranked the two largest current risks to equities as follows.

What fund managers fear most A “disorderly rise in bond yields” and an “AI bubble” ranked as the two largest current risks to equities in Bank of America’s recent survey of professional money managers.

If you hold AI-adjacent equities and the 10-year yield moves above 5%, the correlation data suggests your position faces a structurally different risk environment than the one in which it was built.

Late-cycle signals investors tend to miss

Two markers tend to appear late in a cycle, and both are worth reading as position-sizing information rather than exit signals.

The first is complacency. When investors stop pricing the implications of rising long-term rates, that is a structural feature of late-cycle positioning, not a moral failing. It simply tells you where in the cycle sentiment sits.

The second is fraud risk. Periods of financial mania have historically concentrated fraudulent behaviour, and its surfacing tends to shift sentiment abruptly rather than gradually. Neither marker argues for abandoning exposure. Both argue for sizing positions with the assumption that sentiment can turn faster than fundamentals.

For mining and energy investors specifically, a broad AI-equity correction would compress risk appetite across the whole market. That could create short-term commodity price headwinds even where the underlying demand thesis stays fully intact.

Reading the terrain ahead for commodities and capital allocation

The structural copper demand case and the financial fragility risk both hold at the same time. Neither cancels the other. The way to carry both is through position sizing and clarity about time horizon: a decade-long demand thesis can survive a twelve-month correction, provided you have not sized as though the correction cannot happen.

The structural commodity bull case extends beyond copper to gold and energy, where the same rate-environment and AI-demand dynamics interact differently depending on whether the asset functions as an inflation hedge, a monetary alternative, or a direct input to the buildout.

Three variables will decide whether the commodity thesis plays out on the optimistic or the pessimistic path, and they are worth monitoring in order.

1. The trajectory of the 10-year yield. A move toward the 6% scenario should be treated as a stress test for commodity equity valuations, not a forecast, but the closer yields climb, the more it matters.
2. The pace of hyperscaler capex normalisation. The Futuriom 20-30% pullback forecast for 2026 is the leading indicator to watch for early confirmation that the overinvestment cycle is turning.
3. The timing of any central bank policy reversal. A shift back toward easing is what would restart the commodity reflation leg of the cycle.

The most useful question is not whether the AI boom continues. It is at what price and yield level your commodity position stays attractive regardless of what happens to AI equities.

Anchor the volatility against the long horizon. BHP’s six-fold copper demand projection is the durable backdrop; the near-term swings are noise around it. Investors who separate the commodity demand story from the AI equity story, and watch these three variables, are positioned to act on dislocations rather than react to them.

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. Forward-looking scenarios discussed here are speculative and subject to change based on market developments.