Where Capital Goes When the AI Bubble Runs Out of Fuel

The same machinery that inflated the AI trade is now the mechanism quietly taking it apart. Private credit built the towers, and private credit is writing the demolition notice.

That is the structural irony of late September 2026. The financing apparatus that funded the AI infrastructure boom, private credit funds, junk bond syndicates, and vendor balance sheets, is now the exact channel through which the unwind is transmitting. And most investors remain positioned for the trade that has already turned.

The conditions are visible if you know where to look. AI-linked junk bond yields sit in the high single to low double digits. Major private credit funds have gated redemptions. Semiconductor stocks peaked in June and have not made new highs since.

This is what a late-stage credit cycle looks like when it collides with a technology bubble running out of fuel. The question is no longer whether capital rotates. It is where capital goes when it does, and whether the case for commodities after AI bubble exhaustion is strong enough to act on now.

Here is the analytical frame for making that call: the historical commodity precedent, the honest limits of that precedent, and what the current evidence does and does not confirm.

What the financing collapse tells you that the equity prices do not yet show

Private credit grew roughly 50% between 2023 and 2026, becoming the dominant source of new credit in the U.S. economy. Traditional commercial, industrial, and consumer lending played a comparatively minor role over the same stretch.

That matters because when the marginal credit engine slows, it removes the primary fuel source for the entire leveraged asset complex. Neither housing nor commercial lending is positioned to fill the gap.

The redemption data is the first place the strain becomes visible. These are not isolated fund events. They are the leading edge of liquidity being pulled out of illiquid structures.

Combined redemption requests across Blackstone, Apollo, Ares, BlackRock, and Morgan Stanley’s retail vehicles reportedly exceeded $10 billion in Q1 2026, though this figure has not been independently confirmed. Standard 5% quarterly withdrawal caps prevent an outright collapse. What they do instead is stretch the pain out, concentrating it in mark-to-market losses and a slow erosion of institutional confidence.

From redemption queues to rising defaults: the transmission mechanism

The gating is only half the story. The other half is what happens to the loans inside these funds.

Fitch Ratings private credit default data confirmed the U.S. private credit default rate reached 6.0% on a trailing 12-month basis through April 2026, up from 5.7% in March, with subsequent reporting placing the figure at 6.3% through August 2026.

Fitch Ratings reported the U.S. private credit default rate reached approximately 5.8% on a trailing 12-month basis through early 2026, rising to around 6% by end-April 2026, with smaller issuers hitting as high as 10.9% (figures not independently verified).

When funds face redemptions and rising defaults simultaneously, the response is mechanical. They tighten underwriting, cut new lending, and sell loans at discounts. That widens credit spreads and raises the cost of capital for the middle-market borrowers who depended on cheap private credit in the first place.

So the cost of capital for leveraged companies is rising precisely as their access to fresh credit narrows. Rising consumer credit card delinquencies corroborate the broader stress. This is how credit cycle turns transmit from financial plumbing into real corporate distress, and it typically happens months before equity markets reprice. The read for you: the equity repricing is still ahead, not behind.

Private credit systemic risk compounds the fund-level gating problem because the structures underpinning non-traded vehicles were designed for orderly markets, not the simultaneous redemption pressure and rising default rates now arriving together.

How the AI infrastructure trade turned into a junk bond problem

In the fourth quarter of 2025, Morgan Stanley projected that private credit markets would fund roughly 50% of the AI infrastructure buildout. That makes the credit channel the structural backbone of the AI capital cycle, not a footnote to the equity story.

The pricing of that debt has now shifted in a way that reframes the entire AI narrative. A data-center developer tied to CoreWeave reportedly priced $1.1 billion in five-year notes at a 9.25% yield in September 2026, roughly 2.7 percentage points above comparably rated debt, though the specifics remain unverified. The original reporting placed CoreWeave-linked yields closer to 13%.

Then there is SoftBank.

SoftBank completed an approximately $11.1 billion junk bond offering in late September 2026, described as the largest in the history of that market, with a 7-year tranche yielding as high as 9.75% (subsequent specifics not independently confirmed).

That is not growth-stage speculative paper being priced. That is the market treating AI-infrastructure debt as high-risk credit. And the demand side explains why the revenue meant to service that debt is under pressure from several directions at once:

The concentration risk is not confined to specialist private credit vehicles: AI debt in bond portfolios has migrated into broadly held fixed income products, meaning the repricing pressure reaches investors who have no direct exposure to data-center lending facilities.

• Open-source competition offering comparable capability at lower cost, eroding the pricing power of proprietary frontier labs.
• Enterprise in-housing, as large customers build their own AI systems rather than buy from frontier providers.
• Regulatory lobbying, with a coordinated AI-safety messaging campaign around 8 September characterised as an effort to secure favourable regulation that disadvantages open-source rivals.

The most telling signal sits on Nvidia’s balance sheet. As traditional credit tightens, Nvidia is expanding its own balance sheet to finance its customers. When the primary vendor starts lending to its own buyers, the read for you is that the capital cycle has reached its late stage, where creditworthy demand is exhausted and the seller absorbs the financing risk to keep volume flowing. Semiconductor stocks, having peaked in June and failed to recover, may already be pricing it.

The post-dot-com playbook: what history says about where capital goes next

The obvious comparison is the dot-com bust, and it is worth understanding the actual mechanism rather than the headline. When the technology bubble collapsed after 2000, capital was released from stranded fibre optic infrastructure just as China’s industrialisation created an enormous, largely unanticipated demand shock. That combination produced the commodity supercycle of 2002-2008.

The pattern that followed is instructive. Technology stocks underperformed for roughly five to eight years after the collapse. Value investors holding low-cost equity positions generated positive returns even through the 2000-2001 downturn.

The reversal mechanism matters here. Passive investment flows that built self-reinforcing momentum behind growth stocks can reverse just as forcefully when sentiment shifts, and that reversal is what pushes capital toward neglected real assets.

The structural bull case for commodities today rests on four pillars:

1. Underinvestment: Years of low capex in energy, metals, and mining since 2014 left supply capacity thin.
2. Energy transition: Electrification raises structural demand for battery metals and copper.
3. Fiscal expansion: Post-pandemic infrastructure and defence spending echo earlier public-investment waves.
4. Dollar dynamics: Dedollarisation narratives and a softer dollar support dollar-priced commodities.

Here is where honesty is required. The China demand shock was a once-in-a-generation structural event. Energy-transition demand is real but narrower in scope than building an entire industrial base, and current global growth is slower and less commodity-intensive.

Warren Buffett’s elevated cash position offers a contemporaneous reference point for patient capital. The analogy is strong enough to frame the opportunity. It is not strong enough to treat as an automatic roadmap. Before you act on it, identify which of these specific conditions are actually present today.

What the commodity thesis actually requires you to believe

The historical case is compelling. Now stress-test it against what has to hold for it to pay off.

Start with the supply side. Higher spot prices do not automatically translate into mining volume growth. When permitting, financing access, and institutional mandates all work against new project development, the link between price and volume breaks. ESG scrutiny adds a valuation discount, forces divestments, and restricts financing even in a bull market.

Copper supply constraints illustrate the structural underinvestment argument most precisely: mine development timelines of 10-15 years mean that even a sharp increase in spot prices today cannot produce new volume for most of this decade, locking in the supply side of the bull case regardless of near-term demand variability.

Commodity exposure vs. commodity equity exposure: they are not the same trade

This is the distinction that catches investors out. Owning a commodity index and owning a mining equity give you very different return profiles in a risk-off credit cycle turn.

Producer margins do not move one-for-one with spot prices. Hedges and cost inflation create basis risk, and in a risk-off environment, which a credit cycle turn manufactures, equity multiples can contract even as spot prices rise. That distinction matters most during the transition period, when credit stress is peaking but commodity demand is only beginning to assert itself.

The credit backdrop sharpens the point. Bank strategists warn overall defaults could reach 8%, and up to 13% for AI-disrupted software borrowers, against roughly 4% for standard speculative-grade corporate bonds. That is the risk-off environment in which commodity equity multiples would be compressing.

Three further risks can sever the connection between commodity prices and shareholder returns:

• Geopolitical concentration: Resources sit in unstable regions, where nationalisation or disruption can lift spot prices while destroying individual corporate earnings.
• Management execution: Missteps and value-destructive M&A can offset a broad commodity boom.
• Over-leverage: Debt amplifies downside in a credit tightening cycle.

Apple’s deliberate decision to sit out the initial AI capex wave is the discipline analogy worth holding onto: in a high-capex cycle with genuine stranded-asset risk, restraint is a position. For you, getting the commodity call right is necessary but not sufficient. The equity structure through which you access the exposure adds a separate layer of risk that can negate a correct macro thesis.

Where patient capital actually sits in this cycle

Put the threads together and a positioning decision emerges, not a prediction. This is the accumulation phase before the cycle turn, not the breakout phase.

The distinction is practical. The Bloomberg Commodity Index (BCOM) reportedly posted a 2.2% total return in Q1 2024, a modest bounce rather than a breakout. The historical precedent favours patient positioning at discounted entry points, not momentum-chasing into already-elevated prices.

Warren Buffett’s elevated cash position serves as a contemporaneous institutional signal of patient-capital strategy: preserve capital through the repricing, retain the optionality to deploy into dislocation.

Elevated cash does double duty in this environment. It protects you through the equity repricing that credit conditions are already flagging, and it gives you the optionality to buy commodity-linked assets at prices that reflect dislocation rather than enthusiasm.

The structural case for why this eventually becomes consensus is straightforward. Passive flows that sustained AI and growth stocks will hunt for a new leadership narrative. Just as Google, Amazon, Facebook, and Apple emerged to capitalise on overbuilt fibre optic infrastructure, new commodity-adjacent winners are likely to emerge from cheapened AI compute.

Watch for three confirmation signals before treating the rotation as underway:

• Credit spreads peaking rather than widening further.
• AI equity multiples contracting materially, not just drifting.
• The commodity index breaking convincingly above its Q1 2024 bounce range.

For you as a U.S. investor, the implication is timing. The commodity thesis is a second-half-of-cycle trade. The window to accumulate at reasonable valuations is now, before credit-driven equity repricing triggers the broader rotation that makes commodity exposure consensus and expensive.

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.

Making a well-calibrated bet in a cycle that has not yet turned

Three threads point in the same direction. AI credit is deteriorating, with junk bond yields in the high single to low double digits and semiconductor stocks failing to recover their June peaks. Private credit is contracting, with default rates approaching 6% and rising. And the historical pattern shows capital rotating from a deflating technology bubble toward real assets.

The structural case is compelling and well-supported. The timing case is still early. BCOM’s 2.2% Q1 2024 bounce, supply underinvestment since 2014, and building energy-transition metal demand describe a thesis that is credible but not yet confirmed.

That distinction is the whole game. Credibility and current underpricing in consensus positioning are not the same as confirmation, and the gap matters for how you size a position and time an entry.

So reframe your decision. The useful question is not whether commodities win the next decade. It is this: what would have to be true for them not to, and do those conditions exist right now. A well-calibrated frame lets you update as the confirmation signals arrive, rather than committing before the cycle has turned.

For investors wanting a systematic framework for identifying when the rotation becomes actionable rather than speculative, our dedicated guide to capital rotation triggers examines the specific market signals, credit-spread thresholds, and equity multiple contractions that have historically marked the transition from accumulation phase to confirmed cycle turn.

Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors. These statements are speculative and subject to change based on market developments.