Why James Durance Is Avoiding AI-Financed Tech Bonds

Fidelity International portfolio manager James Durance is taking a deliberately small position in technology credit as large technology companies borrow heavily to finance artificial-intelligence infrastructure. Across income strategies overseeing about $14 billion, he has kept technology exposure below 2%, arguing that the sector’s growing size can become a risk in itself. 1

His caution is fundamentally about whether bond investors are being adequately paid for a crowded, capital-intensive and increasingly correlated credit trade.

A supply-and-valuation call, not an immediate default forecast

Durance’s stance does not require a prediction that major hyperscalers will default. The concern is that a fast-expanding technology allocation can leave credit portfolios exposed to the same underlying assumptions: sustained AI demand, productive capital expenditure and eventual monetization of the infrastructure being built.

At current prices, Durance has said he does not like the sector. His strategies are not forced to follow a benchmark, allowing him to avoid technology debt rather than buy it simply because the sector’s weight in bond indexes is rising. 1

That distinction matters. A benchmark-aware manager can become relatively underweight as a sector issues more debt and takes up a larger share of an index. Durance’s approach is to preserve the option not to participate when he judges the risk-reward balance unattractive.

Why AI bond supply changes the market backdrop

The volume of financing is central to the thesis. Reuters reported that AI-related debt was close to 15% of 2026 investment-grade issuance by midyear, with borrowing supporting data centers and chip supply chains. 2

The composition of that supply is also important. The Federal Reserve Bank of Dallas noted that recent issuance by AI-related firms has been both large and concentrated in long maturities. Its cited Wall Street estimates centered on roughly $300 billion of AI-related investment-grade issuance for 2026, potentially adding up to $360 billion in 10-year-equivalent duration supply. 15

More long-dated corporate debt requires buyers to absorb more interest-rate and credit-spread risk. Even companies with strong balance sheets can see their bonds cheapen when supply is heavy or investors demand extra yield to hold more duration. That is the market-structure issue behind Durance’s caution: the risk is not limited to one issuer’s creditworthiness.

Concentration is the systemic-risk concern

Durance’s warning is best understood as a concentration argument. If a handful of large issuers, credit indexes and investment portfolios become increasingly tied to the same AI capital-spending cycle, disappointment in that cycle could affect a broad segment of the market at once.

That does not establish that such an outcome will occur. It does mean that a portfolio manager must ask whether spreads reflect the possibility of weaker-than-expected AI cash generation, higher leverage, or a prolonged period of issuance. In a concentrated trade, those risks can be transmitted through index weights and investor positioning rather than remaining isolated to a single borrower. 115

The more optimistic “long compute” view

The opposing investment case is not that long-term government bonds are risk-free. It is that AI infrastructure—compute capacity, data centers and related service providers—may offer a more attractive opportunity if demand for computing remains durable.

This is the logic behind Goldman Sachs executive Anshul Sehgal’s preference for being “long compute” rather than simply extending into long-dated Treasuries: investors who accept AI-infrastructure exposure may be seeking growth potential alongside income. But that opportunity is inseparable from the risks Durance identifies. Strong compute demand must ultimately translate into cash flow sufficient to support the capital spending and financing behind it.

The disagreement, then, is mainly about price and durability. The bullish case sees sustained demand for scarce infrastructure. The cautious case asks whether bondholders are being paid enough for a long-lived buildout whose utilization and returns are still being tested.

What investors should monitor next

Fidelity identifies several practical indicators for evaluating whether the AI buildout is becoming self-financed or increasingly reliant on capital markets:

• Free cash flow after capital spending: whether operating cash generation remains after AI investment.
• Net debt and interest coverage: whether incremental borrowing is reducing financial flexibility.
• Credit spreads: whether yields adequately compensate investors for supply and uncertainty.
• Long-term Treasury yields: a key input into the attractiveness and pricing of long-dated corporate bonds.
• Externally funded AI investment: whether spending is being funded internally or through new borrowing and other outside capital. 7

Durance’s sub-2% technology exposure is therefore a portfolio-construction decision: avoid letting a rapidly expanding AI-financing theme dominate an income strategy before valuations and spreads offer a clearer margin of safety. The bullish “long compute” view may prove right if infrastructure demand and cash flows remain strong, but the cautionary case is that size, duration and common exposure can become risks of their own. 1715