The Second Act of AI Capital Expenditure: When the Bond Market Becomes the Scoreboard of the Computing Power Race

The Second Act of AI Capex: When the Bond Market Becomes the Scoreboard of the Compute Race

Over the past three years, the market’s discussion of artificial intelligence has revolved around almost two questions: where the ceiling of demand lies, and who is leading in the technology race. As the scale of capital expenditure began to be measured in hundreds of billions of dollars, a more financial question surfaced—who is footing the bill for this buildout, and how the manner of payment will in turn affect interest rates, credit spreads, and equity valuations.

Fidelity Institutional’s Capital Markets Strategy team, in its Market Signals Weekly, offered a judgment worth incorporating into a macro analysis framework: the next phase of the AI trade is no longer merely a question of “exposure,” but of “capital discipline.” Demand remains strong, but the next market signal may come from how this buildout is financed, and what that financing means for rates, credit, and stock valuations.

1. The Two Sides of Capital Expenditure: Current Growth and Future Liabilities

Hyperscale technology companies are making enormous commitments in semiconductors, data centers, networking equipment, and power infrastructure. These expenditures support upstream suppliers and also support current economic activity. During a period when the macro environment was not easy, capital expenditure in fact played an important pillar of private-sector demand.

But the temporal structure of capital expenditure determines its two-sided nature. It occurs before revenue: as the cycle matures, the gap between “capital invested” and “cash recovered” will gradually replace the demand narrative and become the market’s core variable. What companies get at this stage is higher depreciation, higher interest expenses, and refinancing needs they must face further in the future. These costs do not suddenly appear on a single day; rather, they gradually become visible in income statements and cash flow statements over the coming quarters.

This is also the key to understanding the current macro contradiction: capital expenditure is both a tailwind for growth and a point of financial fragility. While it pulls current activity forward, it lays the groundwork for subsequent debt service and refinancing, and the time gap between the two is precisely where market pricing is most prone to error.

2. The Bond Market Is a Scoreboard

Looking back at history, transformative infrastructure always first attracts capital and only later demonstrates its economics. Railways, telecommunications networks, and internet backbone networks all followed a similar path: the technology was reshaping the economy, but the returns ultimately delivered by some financing structures were not ideal. AI is not necessarily an exception.

The real transmission mechanism lies in financing costs. When capital is abundant, companies can keep investing even when the return cycle is not yet clear; once long-term bond yields rise, credit spreads widen, or debt investors demand stricter covenants, hurdle rates change. Projects that looked reasonable under easy financing conditions quickly lose their appeal after the cost of capital rises.This creates a cross-market feedback loop: large-scale borrowing itself can put upward pressure on yields; higher yields compress equity valuation multiples while raising debt-servicing costs; and weaker equity prices in turn make new financing more expensive. In other words, the AI trade cannot be evaluated in isolation from interest rates and credit markets.

It is worth noting that even if financing conditions tighten, demand may remain strong. The divergence in this scenario would be stark: the beneficiaries would be companies with scarce capacity, pricing power, strong balance sheets, and a clear path from capital expenditures to recurring cash flow. The distinction that truly matters may be the difference between “using AI investment to expand an existing competitive advantage” and “borrowing to stay in the game”—the former can compound value; the latter may find that the cost of staying in the game has exceeded the economic return itself.

3. Concentration Is Spreading from Equity Indexes to Bond Indexes

The AI trade has long been about more than equity indexes. Record investment-grade debt issuance by hyperscalers, related data centers, and chip-financing vehicles is changing the composition of the corporate bond market.

According to estimates from Fidelity and public company disclosures, this group has issued more than $300 billion of investment-grade bonds so far this year, compared with $136 billion for full-year 2025; issuance is expected to approach $500 billion in 2026 and rise further to about $550 billion in 2027.

When this supply is included in bond benchmark indexes, bond portfolios gain greater exposure to this AI capital cycle—the same cycle that has already dominated parts of the equity market. According to Bloomberg data, as of the end of August 2026, hyperscaler and AI-related bonds accounted for about 6.2% of the Bloomberg U.S. Corporate Bond Index, up from 4.2% in 2025; their share of the Bloomberg U.S. Aggregate Bond Index rose from 1.1% to 1.5%. Individually, Amazon’s weight in the corporate bond index rose from 0.80% to 1.40%, Oracle from 1.10% to 1.30%, Meta from 0.80% to 1.00%, and Alphabet from 0.40% to 0.90%.

More noteworthy than the weights is how risk is measured. Measured by “duration × spread,” the group’s aggregate credit risk contribution has already surpassed that of the six largest banks. This implies a concentration that is even less conspicuous than in the equity market: a seemingly diversified bond fund may simultaneously hold multiple technology issuers, data center financing entities, utilities, and capital goods companies, while the underlying dependencies of their credit outcomes are highly aligned—continued AI spending, open capital markets, and eventual monetization.

As bond spreads and new-issue concessions widen, the importance of issuer limits and security selection will rise markedly. Diversification may be far weaker across these correlations than it appears from sector labels.## IV. Circular Financing and Opacity

AI financing structures are becoming increasingly interwoven and increasingly difficult to assess. Chipmakers and hyperscalers may invest in AI labs or specialized cloud providers, and these investee companies in turn commit to purchasing chips and compute over the next several years. Meanwhile, data center development is increasingly being outsourced to third-party operators, which lease the facilities back to tech companies under long-term agreements, sometimes with exit clauses.

These arrangements often fall within private markets, making it difficult for investors to identify unintended risks: how assets are actually financed, and how much demand within the ecosystem is actually supported by financing itself. Such structures can amplify growth in the present, but they can also amplify downside risks later. The key question is not whether the accounting treatment is compliant, but whether the boundary between demand signals and financing signals remains clear.

V. From “AI Exposure” to “AI Capital Discipline”

The opportunity in AI remains considerable, but the focus of investment discussion is shifting away from “adoption rates” and “market leadership.” The next phase will test whether companies can convert earlier heavy investment into returns that exceed their rising cost of capital. In this framework, balance sheet quality is as important as earnings growth.

Two companies with similar AI exposure may end up with completely different outcomes: one can self-finance, while the other depends on the bond market and optimistic forecasts of future cash flows. The core question is therefore not just “who will win the AI race,” but “who can finish the race without impairing returns to shareholders and creditors.”

VI. Implications for Long Cycles and Global Macro

Viewed over a longer time horizon, AI capex is becoming a key node connecting the real economy and financial conditions. It contributes growth on the demand side, while on the financing side it may become a transmission channel between interest rates, credit spreads, and equity valuations.

This means the next round of stress testing may not necessarily occur within the tech sector. It may first appear in the segments of the market that fund compute expansion: the investment-grade credit market, data-center-related financing structures, and diversified fixed-income portfolios that hold these bonds. The fiscal and debt backdrop is equally important—in an environment where the long-term rate anchor is rising and term premiums are being repriced, any large-scale, long-cycle, back-loaded private investment cycle will expose its sensitivity to the cost of capital earlier.

For macro research, the variables that need to be tracked are also changing: long-term bond yields and term premiums, investment-grade credit spreads and new-issue concessions, the duration and spread contribution of AI-related issuers in credit indices, and the coverage of capex by corporate free cash flow. The technology narrative still matters, but what shapes the cycle may increasingly be the financing structure itself.

---Source and analytical basis: This article is reconstructed and analyzed based on publicly available content from Fidelity Institutional's "Insight & Outlook: Fidelity Market Signals Weekly." Original link: https://institutional.fidelity.com/advisors/insights/series/fidelity-market-signals-weekly