According to Bloomberg Markets, CreditSights Global head of credit strategy Winnie Cisar is now providing dedicated outlooks on technology company bonds — a signal that fixed-income markets are beginning to price risk into a sector that equity investors have, until recently, treated with considerable optimism. That shift in credit market attention is worth examining carefully. In the late 1990s, the bond market was among the last institutional constituencies to turn sceptical on internet companies — and when it did, the equity collapse followed within months.

The Mechanism, Not the Asset

The popular account of the dot-com bubble focuses on irrational exuberance, sock puppets, and Super Bowl advertisements purchased by companies with six months of runway. That account is accurate but incomplete. The deeper mechanism was a capital cycle: cheap, abundant funding enabled companies to operate at sustained losses while building user bases, infrastructure, or market share, with the implicit promise that monetisation would eventually arrive. Investors accepted this promise because the technology — the internet — was genuinely notable, and because enough early winners (Amazon, eBay) seemed to validate the thesis.

That mechanism — abundant capital masking structural insolvency, followed by a credit tightening that exposes the gap — is precisely what credit analysts now appear to be examining in the AI sector.

AI's Capital Cycle in 2025–2026

The parallels between the current AI investment boom and the dot-com cycle are structural, not merely atmospheric. In both cases, a genuinely significant technology attracted capital at a pace that outran demonstrated revenue. In both cases, new valuation metrics emerged to justify prices that conventional analysis could not support. In the late 1990s, analysts substituted price-to-eyeballs and price-to-clicks for price-to-earnings. Today, AI infrastructure spending is frequently justified by reference to total addressable market projections and model capability benchmarks rather than near-term cash flow.

The scale of capital deployment into AI infrastructure — data centres, chips, power infrastructure, and foundation model training — has been extraordinary. Major technology companies have disclosed capital expenditure plans running into hundreds of billions of dollars collectively over multi-year horizons. The central question, which credit analysts are now beginning to formalise, is whether the revenue generated by AI applications will be sufficient to justify that infrastructure spend — and on what timeline.

In credit markets, the question is never whether a technology works. It is whether the cash flows arrive before the debt does.

This is precisely the question the bond market failed to ask aggressively enough between 1998 and early 2000. When it finally did — when high-yield spreads on internet and telecommunications companies began to widen in late 2000 — the equity market's last source of cheap capital evaporated. Companies that had been funding operations through secondary equity offerings and convertible debt found those windows closing simultaneously.

The Pets.com Problem and the Amazon Question

The dot-com collapse did not destroy the internet. It destroyed the companies that could not survive long enough to reach sustainable economics. Pets.com lasted only a short time after its IPO. Amazon fell dramatically from its peak — by some estimates more than 90% — before recovering to become one of the most valuable companies in history, a process that took many years. The technology was real in both cases. The difference was the underlying business model's capacity to generate cash before capital markets lost patience.

The AI sector faces an identical sorting problem. Some companies building on AI infrastructure will prove to be Amazons: genuinely durable businesses that can survive a capital cycle contraction and emerge dominant. Others will prove to be Pets.coms: companies whose unit economics never close, whose burn rates exceed their revenue growth, and whose survival depended entirely on the continuation of cheap, abundant funding.

Identifying which is which in advance is, by definition, difficult — the dot-com era demonstrated that even sophisticated institutional investors got this wrong systematically. For macro context on how capital cycles interact with technology investment booms, the pattern across multiple historical episodes suggests that the sorting process typically begins not with an equity market correction but with a credit market reassessment — exactly what appears to be underway now.

For readers tracking current valuation signals across sectors, the Bubblepedia Bubble Watch indicators provide a continuously updated read on where concentration and valuation risk is accumulating.

New Metrics, Old Warning Signs

One of the most reliable leading indicators of a bubble's mature phase is the proliferation of novel valuation frameworks designed to justify prices that conventional metrics cannot. In 1999, analysts at major investment banks published research introducing metrics like price-to-page-views and price-to-registered-users as replacements for earnings-based analysis. These metrics were not entirely irrational — user engagement does have economic value — but they served a secondary function: they allowed analysts to maintain buy ratings on companies whose fundamentals, under any conventional framework, could not support their valuations.

The AI era has generated its own suite of novel metrics. Model parameter counts, inference speed benchmarks, and synthetic capability scores have all been cited in contexts that blur the line between technical assessment and investment justification. This is not to say these metrics are meaningless — they are not — but their deployment in investment contexts echoes the dot-com era's creative accounting for attention and traffic.

The Dot-Com Bubble vs. AI Infrastructure comparison on Bubblepedia examines these parallels in greater structural detail, including the role of semiconductor and infrastructure spending in both cycles.

What the Credit Market Is Telling Us

The emergence of dedicated credit analysis on technology bonds does not, by itself, signal an imminent collapse. Credit analysts cover sectors continuously, and scrutiny is not the same as deterioration. What it does signal is that the fixed-income market — historically more conservative and more sensitive to cash flow realities than equity markets — is beginning to formalise its assessment of AI-related credit risk.

None of this is to predict an equivalent outcome for AI. The current AI investment cycle has meaningful differences from the dot-com era: the leading companies have substantial existing revenue bases, the infrastructure being built has broad applicability beyond a single use case, and the capital markets environment — while generous — has not yet replicated the late-1990s condition of essentially zero standards for public market access.

But the mechanism is the same. Capital abundance masks the gap between burn rate and revenue. Credit markets begin asking questions that equity markets have deferred. The answers determine which companies are Amazons and which are Pets.coms. History suggests the market's ability to identify the difference in advance is considerably worse than participants believe at the time. The full Bubblepedia bubble encyclopedia documents how consistently this pattern has repeated across asset classes and eras.