---

Discussion

Red Velhouse:

Kate, take us through the market move. What makes this more than an ordinary technology-sector correction?

Kate Burvish:

The key feature was cross-asset transmission. Selling was concentrated in AI-related equities, semiconductors, and data-center operators, but investors also repriced the debt supporting that buildout. CoreWeave’s five-year credit-default-swap spread—essentially, the cost of buying default protection—rose above 1,100 basis points, and one bond’s yield reached about 16 percent before retreating. That does not prove an inevitable default, but it is a very loud warning about financing risk.

Omar Seidren:

And the market was questioning something narrower than whether machine learning is useful. It was asking whether every planned server cluster, power contract, and specialized cloud facility will earn an adequate return on the schedule investors assumed. The technology can be transformative while a particular financing model is fragile. Those are separate propositions.

Sam Dewinski:

That distinction is why the late-1990s telecom comparison helps, if we handle it carefully. The internet was real, demand for connectivity was real, and yet companies built too much capacity, forecast adoption too confidently, and financed infrastructure on assumptions that later failed. Much of the infrastructure survived. Shareholders and creditors did not all receive the future they had priced.

Red Velhouse:

So are we looking at a valuation correction, an interest-rate shock, or the first market test of an AI infrastructure cycle built on debt?

Kate Burvish:

The evidence does not cleanly separate those causes. September inflation was hotter than expected: consumer prices rose 0.6 percent in the month and 3.7 percent over the year. Normally, that would support higher yields. Instead, the ten-year Treasury yield fell sharply during the selloff. That suggests investors were also pricing a growth or risk-premium shock, not merely reacting to inflation.

Omar Seidren:

There is an important distinction within the AI industry. A company selling highly demanded chips or generating substantial cash flow has more room to absorb a delay. A newer cloud operator that buys expensive computing equipment, leases facilities, and depends on future utilization has a much narrower margin for error. If rental prices fall or customers arrive later, the servers still consume capital even when they are not earning enough.

Sam Dewinski:

History would call that a mismatch between the useful life of infrastructure and the timetable of finance. Railways, telecom networks, and electricity systems all involve long-lived assets. Debt, however, comes due on a much shorter schedule. The question is not simply whether an asset will eventually be valuable; it is whether its owner can survive until that value appears.

Red Velhouse:

Omar, the optimistic case is not just marketing. What would have to be true for this investment wave to justify itself?

Omar Seidren:

Utilization has to rise, but utilization alone is not enough. AI services must generate revenue that exceeds the cost of chips, electricity, networking, buildings, maintenance, and financing. We also need genuine applications rather than endless demonstrations: systems that reduce costs, create valuable products, or let organizations do work they could not previously do. The capability is advancing rapidly, but the business model is still being discovered in real time.

Kate Burvish:

And the incentives can push spending beyond what near-term demand justifies. Companies fear losing strategic position, so each one may rationally build more capacity even if the industry collectively overshoots. The private decision—secure chips, power, and customers now—can be sensible while the aggregate result is excess capacity and weaker returns.

Omar Seidren:

Exactly. In technology, waiting can be expensive because the frontier moves quickly. But there is a financially absurd version of that race: everyone builds the runway because they are afraid another company’s airplane might take off first. The danger is not that the airplane is imaginary. It is that too many runways get financed before anyone knows how much air traffic will exist.

Red Velhouse:

Sam, does that analogy clarify the present, or exaggerate the danger? Today’s largest AI spenders are generally more profitable than many dot-com companies were.

Sam Dewinski:

It clarifies the mechanism, not necessarily the timing or scale. The leading firms today are larger, more profitable, and technologically stronger than many late-1990s issuers. That makes a collapse of the entire technology thesis a poor historical inference. But the analogy remains strong for overbuilding, vendor interdependence, and debt service. A sound technology can still produce a destructive investment boom around it.

Kate Burvish:

The credit data also argues against treating every AI company as equally endangered. S&P Global described a widening gap between AI-exposed credits and the broader investment-grade market. Oracle’s five-year credit-default-swap spread approached 300 basis points, while other AI-linked names widened as well. The stress was especially severe among operators dependent on external financing. So far, that is sector-specific repricing, not generalized credit impairment.

Red Velhouse:

What could turn sector-specific stress into something broader?

Kate Burvish:

The connections matter. Corporate bond issuance for hyperscaler-related AI expansion exceeded 100 billion dollars in 2025, while private-credit funds and special-purpose vehicles increasingly financed data-center assets. If utilization disappoints, losses may not stay with one operator. They can reach lenders, bond investors, landlords, utilities, and funds that thought they were diversified. The Reserve Bank of Australia has warned that leverage and liquidity mismatches can amplify forced selling.

Omar Seidren:

But interconnected does not mean automatically contagious. A data center with strong customer contracts, reliable power, and equipment that can be redeployed is not the same asset as a speculative project dependent on one optimistic forecast. We need to know who bears the obligation when a capacity contract is canceled, when a guarantee sits outside a parent company, or when a private vehicle owns the facility. The technical infrastructure and the legal-financial wrapper both matter.

Sam Dewinski:

That is where history offers a warning about labels. In earlier booms, risk often looked contained because obligations were distributed among suppliers, lenders, and special entities. Distribution can reduce concentration, but it can also make the final exposure harder to see. Off-balance-sheet financing is not inherently deceptive; the concern is that economic obligations may be larger than the headline balance sheet suggests.

Red Velhouse:

The currency move was also unusual. Why did the dollar weaken during a U.S.-originated risk-off episode instead of attracting the usual safe-haven demand?

Kate Burvish:

Markets may have been marking down U.S. growth and future capital investment, while falling Treasury yields reduced the dollar’s rate advantage. There had been a view that U.S.-centered AI spending would attract foreign capital and support demand for U.S. assets. If that expectation softens, the dollar can weaken. But this is not exclusively an AI story; oil, geopolitical developments, European fiscal concerns, and central-bank expectations were also moving currencies.

Omar Seidren:

And a weaker dollar would have mixed consequences for the buildout. It might help U.S. exporters, but much of the hardware and capital equipment involved in AI infrastructure is globally sourced. Imported chips, energy-related inputs, and specialized equipment could become more expensive. The currency can cushion some businesses while raising costs for the investment cycle itself.

Red Velhouse:

Beyond investors, who bears the downside if this becomes an overbuilding cycle?

Kate Burvish:

The distribution depends on the financing chain. Equity holders take losses first, but lenders and private-credit investors can be exposed next. Utilities and local governments may have planned grid or power investments around data-center demand. Regions can receive construction activity without receiving many permanent operating jobs. And if public subsidies helped secure power, land, or infrastructure, taxpayers may share risks that were initially presented as private investment.

Sam Dewinski:

The historical pattern is mixed. Infrastructure booms often leave behind useful systems and new economic capacity, even after financial losses. That does not make the losses harmless. It means society can inherit the benefit while the original investors bear—or try to shift—the cost. The political question becomes who gets the durable infrastructure and who pays during the correction.

Red Velhouse:

Let’s make this practical. What should listeners watch over the next one to three months to distinguish a temporary repricing from deeper deterioration?

Omar Seidren:

Watch operating evidence rather than headlines: data-center utilization, customer cancellations, pricing for computing capacity, and whether new AI services generate repeat revenue. Also watch whether the most capable firms keep investing from cash flow while debt-dependent operators pull back. A one-day equity rebound tells us very little about those fundamentals.

Kate Burvish:

Financially, watch whether spreads widen among strong issuers, whether companies can issue debt on reasonable terms, and whether refinancing becomes difficult. Rising borrowing costs alone are not a crisis. Failed issuance, forced asset sales, or stress moving from specialized operators into broad investment-grade credit would indicate a more serious transmission mechanism.

Sam Dewinski:

And compare promises with completion. Earlier infrastructure booms were full of forecasts about future demand; the decisive evidence arrived when projects had to operate, earn revenue, and service debt. The historical lesson is not that AI is simply another bubble. It is that transformative technology does not exempt investors from arithmetic.

Red Velhouse:

The unresolved issue is whether October’s selloff marks a temporary separation between strong AI businesses and highly leveraged operators, or the beginning of a broader reassessment of how quickly AI investment can earn its way through debt and infrastructure costs. Watch credit spreads, refinancing access, computing utilization, customer demand, and the treatment of guarantees and private financing. The technology may prove transformative even if parts of the investment cycle prove excessive. Sources and references for this discussion are available with the episode at Factolio.com.

---

Sources and References

These sources supported the factual material used in this discussion. Factolio’s panel discussion is AI-generated from researched evidence and is written in original language.

1. Associated Press — Stocks rebound after tech sell-off and Indonesia shares sink on market downgrade warning (NEWS)
2. Reuters, republished by AOL — Dollar heads for weekly fall as investors gauge AI selloff impact (NEWS)
3. Reuters, republished by Investing.com — AI credit worries send default insurance to record high (NEWS)
4. S&P Global Market Intelligence — US IG Credit: No Sector Is an Island? (ANALYSIS)
5. MSCI — Agentic AI Widens Tech Spreads, but Hardware Holds Firm (ANALYSIS)
6. Bank for International Settlements — Financing the AI boom: from cash flows to debt (PRIMARY)
7. Bank for International Settlements — Markets recalibrate amid shifting currents (PRIMARY)
8. U.S. Bureau of Labor Statistics — Consumer Price Index — September 2026 (PRIMARY)
9. MUFG Research — FX Focus — How AI is reshaping FX markets (ANALYSIS)
10. Board of Governors of the Federal Reserve System — Federal Reserve issues FOMC statement — September 16, 2026 (PRIMARY)
11. Reserve Bank of Australia — Financial Stability Review — The Global Macro-financial Environment, March 2026 (PRIMARY)
12. Federal Reserve Board — Speech by Vice Chair Jefferson on financial stability (PRIMARY)
13. Amundi Research — AI Boom or Bubble? Lessons from the Dot-Com Period (ANALYSIS)