AI is no longer an equity story…

AI is no longer just an equity story. It is becoming a credit story.

For much of the artificial-intelligence boom, investors have concentrated on equities. Semiconductor earnings, hyperscaler capital expenditure and expectations of transformational productivity gains have driven share prices higher and increased the concentration of global equity indices.

The next phase of the cycle, however, is increasingly being financed through debt.

The central question is no longer simply whether AI-related companies can generate sufficient earnings to justify their equity valuations. Credit investors must also ask: who is paying for the infrastructure, where does the leverage ultimately sit, what supports the borrowing and is the yield sufficient compensation for the risks being assumed?

That makes AI an increasingly important issue for fixed-income investors—and for diversified portfolios that may have more exposure to the theme than their headline equity allocation suggests.

Who pays for the AI buildout?

Craig Veysey, manager of the newly launched Guinness Global Dynamic Bond Fund, believes the most useful way for a bond investor to frame the AI buildout is to consider how it will be funded.

Current estimates suggest that approximately $5.5 trillion of capital could be required to finance AI and data-centre infrastructure through to 2030. Only around one-quarter of this is expected to come from organic cash flow and new equity. The remainder may need to be raised through investment-grade bonds, leveraged finance, private credit and structured markets, including an estimated $2.1 trillion from high-grade bond issuance alone.

Borrowing on that scale is no longer simply a question about the balance sheets of a handful of large technology companies. It has the potential to reshape the composition of global credit markets.

Combined capital expenditure among the principal hyperscalers increased from approximately $379 billion in 2025 and could rise to around $770–825 billion during 2026. By 2028, forecasts suggest annual spending could reach approximately $1.3 trillion. These projections are also continuing to move higher as companies encounter rising costs for memory, computing capacity, data-centre construction and power infrastructure.

For credit investors, that presents a fundamental underwriting challenge. Assessing a five-, ten- or thirty-year bond is difficult when a borrower’s capital-expenditure requirements are being revised almost monthly.

Strong demand for AI services may support future revenue growth, but the cash must be committed today. Data centres must be built, chips acquired and power secured well before the resulting revenues are fully visible.

Credit concentration is beginning to mirror equity concentration

The equity-market implications of AI concentration are already well understood. A small number of technology companies have come to dominate market-capitalisation-weighted indices, meaning many investors have accumulated significant exposure whether or not they made an active decision to do so.

A similar dynamic is now developing in investment-grade credit.

As the hyperscalers become some of the largest bond issuers in the market, benchmark-aware fixed-income portfolios may acquire increasing AI exposure automatically. Investors could therefore find themselves holding a large and increasingly correlated position through both their equity and bond allocations.

However, Craig argues that the largest technology borrowers should not be treated as a homogeneous group.

Alphabet, for example, has already issued heavily and funded part of its investment programme through equity. A meaningful proportion of its prospective bond supply may therefore already be reflected in current credit spreads.

Microsoft presents a different situation. It has guided towards approximately $190 billion of capital expenditure during 2026 but has yet to issue an equivalent amount of debt against those plans. Its bonds still trade among the tightest names in the sector. In Alphabet’s case, the risk may be increasingly reflected in the price; for Microsoft, a larger part of the supply pressure may still lie ahead.

The companies may appear similar from a thematic perspective, but the credit opportunity can be very different. That distinction is easily lost in a benchmark-driven approach.

Oracle provides an early warning

Oracle offers perhaps the clearest recent illustration of the tension between AI growth and creditworthiness.

The company was downgraded to BBB−, the lowest investment-grade rating, as the financial demands of expanding its AI infrastructure operations placed greater pressure on cash flow and increased its business risk. Oracle is not a distressed borrower, but it now sits only one rating notch above speculative grade.

The downgrade does not mean its strategy will fail. Its infrastructure investment could generate substantial future revenues. But it demonstrates that strong demand does not automatically translate into stronger credit quality.

There is also a concentration issue. Craig notes that Oracle’s downgrade was influenced not only by leverage, but by the extent to which one customer represents a substantial proportion of its contracted future revenues. A large order book can therefore be less reassuring when its quality depends heavily on a small number of counterparties.

Credit analysis must examine not simply how much revenue has been contracted, but who ultimately stands behind it.

Where does the leverage actually sit?

The financing architecture is also becoming more complex.

An increasing share of AI investment does not appear as straightforward corporate borrowing. It is being arranged through special-purpose vehicles, private placements, long-term leases, corporate guarantees and structured securities.

Craig highlights a recent AI-compute financing involving Broadcom and Anthropic. Under the structure, a special-purpose vehicle acquired Broadcom chips and leased them to Anthropic. Senior notes supported by a Broadcom guarantee priced at around 5.75%, while a smaller, unguaranteed and more junior tranche priced at 8.5%.

The underlying pool of AI hardware was broadly the same. What changed was the position within the capital structure and the strength of the corporate backstop.

Once the Broadcom guarantee was removed and the investor moved further down the structure, the required yield increased by well over two percentage points. That demonstrates that the perceived investment-grade quality of the senior securities depended as much on the guarantor as on the assets or tenant supporting them.

It also shows how risk can increase without appearing in headline corporate debt figures. Broadcom may not have borrowed the money directly, but its guarantee represents a real financial commitment. Rating agencies can treat such obligations as economically similar to debt when assessing effective leverage and financial flexibility.

A company can therefore appear conservatively financed at the corporate level while still supporting substantial risks through guarantees, lease commitments and project-finance structures elsewhere in the ecosystem.

Financing long-term debt with short-lived assets

The collateral itself introduces another risk.

Much of the borrowing is ultimately supporting specialised technology hardware that may depreciate rapidly. Craig notes that a high-end AI chip could lose approximately half its resale value within three years, while the secondary market for such assets remains relatively undeveloped.

Yet the associated debt may run for ten, thirty or even longer periods.

Financing an asset with a relatively short technological life using very long-term borrowing creates an obvious mismatch. If computing technology develops faster than expected, today’s premium infrastructure may become less valuable well before the debt issued against it matures.

This does not make every AI-related financing unattractive. Strong guarantees, durable leases and well-capitalised counterparties can provide substantial protection. But the name on the bond, its headline rating and the broad AI narrative are not sufficient.

Investors must understand the underlying asset, the tenant, the guarantor, their position in the capital structure and the recoverable value of the collateral.

Why flexibility matters for MGTS Qualis Defensive

These developments reinforce the importance of active credit selection within the MGTS Qualis Defensive Fund.

The purpose of the fund’s fixed-income allocation is not simply to own a bond benchmark or to reproduce the concentrations developing elsewhere in financial markets. It is to provide diversified sources of income, capital stability and defensive return potential across changing economic conditions.

As AI-related issuers become larger parts of credit indices, a passive or benchmark-constrained approach may increasingly absorb this exposure by default. That does not mean every AI bond should be avoided. It means each security must earn its place in the portfolio.

This thinking helped inform our decision to provide seed capital to the newly launched Guinness Global Dynamic Bond Fund.

Craig’s unconstrained, research-led approach allows him to move across government bonds, investment-grade credit and other fixed-income opportunities without being forced to replicate benchmark weightings. He can distinguish between companies whose future issuance may already be reflected in spreads and those where the market may be underestimating the financing still to come.

Just as importantly, he can examine guarantees, leases and special-purpose structures to establish where the underlying risk genuinely resides.

AI remains a potentially transformative technology and many of the companies funding its development remain financially strong. The conclusion is not that an AI credit crisis is inevitable.

It is that the buildout is becoming too large, too capital intensive and too structurally complex to analyse through equity prices alone.

Increasingly, the success—or excess—of the AI cycle will be visible through bond supply, credit spreads, guarantees, collateral values and balance sheets. For fixed-income investors, detailed credit work and the freedom to step away from benchmark-driven exposure are likely to become more important, not less.

This article is for information only and does not constitute investment advice or a personal recommendation. Capital is at risk, and the value of investments can fall as well as rise.