Global Market Strategist
The artificial intelligence (AI) buildout has become the most capital-intensive investment cycle since the telecoms boom at the turn of the century. The first phase was largely funded by the hyperscalers’ operating cash flow, but the scale of spending now requires a broader mix of funding sources.1 Cumulative AI-related investment is estimated to reach USD 5,500 billion by 2030, as companies build AI models and the infrastructure that supports them, including data centres, semiconductors, power and utilities.
Hyperscaler operating cash flow and general equity issuance are expected to cover only around a quarter of this investment. That broadens the AI opportunity set across asset classes. Investment-grade bonds, high-yield bonds, leveraged loans, securitisation and private markets such as infrastructure, real estate and private credit will all play a role.
For credit investors, this marks a clear shift: debt issuance is becoming central to funding the AI buildout. Investment-grade bonds are expected to be the largest source of external capital, accounting for USD 2,100 billion, with a further USD 700 billion coming from high-yield bonds, leveraged loans and securitisation.
In this perspective, we focus on hyperscaler debt issuance and its implications for the investment-grade bond market. We do not expect this new wave of supply to derail the market, but it is reshaping it. At the same time, hyperscaler debt offers investors a more defensive way to access the AI investment theme. Hyperscaler spreads have widened, reflecting heavier and more concentrated supply, as well as the rising scale and complexity of off-balance-sheet exposures. While we view this widening in hyperscaler spreads as warranted, we see value in the additional carry. The key test for bondholders will be the recovery in free cash flow after 2027, shaped by the trajectory of future capital expenditure (capex) and the pace of AI monetisation. Active management is well placed to add value in this environment of rising concentration, greater dispersion and increasing complexity.
Insatiable demand for capital as hyperscalers fund the AI buildout
The capital-light model that defined the largest US technology companies is fading. Rising compute demand, higher input costs and the race to secure capacity have pushed hyperscaler capex sharply higher, from USD 156 billion in 2022, the year ChatGPT 3.5 was released, to more than USD 1,000 billion expected in 2027.
Despite strong operating cash flow, capital intensity has increased so sharply that free cash flow is expected to turn negative in the near term. This creates a growing need for external funding. We estimate that hyperscalers could issue USD 300 billion in investment-grade bonds over the next year alone. That makes hyperscaler issuance one of the most important themes in investment-grade credit.
Hyperscalers are raising debt across multiple currencies, maturities and financing structures. US hyperscaler bond issuance has surged from 2% of total USD investment-grade issuance between 2022 and 2024 to an expected 9% in 2026. Year to date, hyperscalers have issued USD 219 billion of investment-grade bonds, including USD 62 billion equivalent in non-USD currencies such as EUR, CAD, CHF, GBP and JPY.
The use of project finance structures by hyperscalers is material and becoming more prominent, making credit analysis more complex. These structures are used to finance data-centre, compute-capacity and power-infrastructure assets. They typically raise debt against the cash flows and assets of a specific project, often through a ring-fenced vehicle, rather than directly against the hyperscaler’s balance sheet. This diversifies funding and helps preserve balance-sheet flexibility.
Credit investors and rating agencies therefore need to consider not only reported leverage, but also future off-balance-sheet exposures and the associated credit and liquidity needs. Disclosed data-centre lease obligations already amount to around USD 1,400 billion. Roughly USD 1,100 billion of these obligations are currently off balance sheet, as they are recognised only when the data centres become operational and the leases commence. In addition, hyperscalers have unconditional purchase commitments, which amount to roughly USD 1,500 billion and cover purchases of microchips, computing infrastructure and power. A substantial part of these purchase commitments is included in near-term capex expectations and is therefore taken into account by investors and rating agencies.
Concentration limits are unlikely to be a constraint in the near term, but will become more relevant over time as the combined hyperscaler weight in the USD investment-grade index could approach 10% by 2030. Individual hyperscaler weights are currently 1.5% or less in the USD investment-grade index, and the USD investment-grade market has historically functioned in an orderly manner with issuer concentration levels of 3-8%. This time, concentration limits could be lower, as many investors already have substantial AI exposure across equities, corporate bonds and private markets. Hyperscalers will therefore continue to diversify their funding through non-USD investment-grade bond markets, project finance structures and, increasingly, private credit.
Credit markets are repricing, not rejecting, the AI buildout
Hyperscaler bonds have lost their scarcity premium as fixed income markets adjust to this new supply regime. Hyperscaler spreads have widened by around 30 basis points (bps) year to date, compared with 2 bps widening in the broader USD investment-grade index. The widening of hyperscaler spreads gained pace in July after larger, and in some cases less well-flagged, issuance. Adjusting for duration, hyperscaler spreads are trading at around 105 bps, wider than similarly rated bonds and the broader market, which trades at around 80 bps. The move has been more pronounced at longer maturities, where issuance has increased more sharply, the buyer base is narrower and the risk premium for uncertainty around return on investment is arguably higher. Heavy long-end issuance from hyperscalers may also have contributed to the recent rise in long-dated US Treasury yields.
We view the hyperscaler spread widening as warranted and primarily driven by the need to absorb high and concentrated debt supply, rather than by a deterioration in hyperscaler business models. With hyperscalers now offering a spread premium to the broader index, we see value in earning the additional carry. We also expect greater dispersion among hyperscalers as investors scrutinise projections for future capex and signs of AI monetisation, similar to the pattern visible in equity markets.
Balance sheets are starting from a position of strength, with gross and net leverage well below the average for the broader investment-grade industrials index. The AI investment cycle is unusually large and front-loaded in nature, but hyperscalers have considerable additional debt capacity. We estimate that hyperscalers could add USD 1,500 billion of debt before their lease-adjusted leverage ratios reach the average for the USD investment-grade industrials index. Fundamentals are further supported by strong margins and solid earnings growth. The recent rise in the cost of debt is likely viewed as a marginal headwind to overall funding costs and is therefore not expected to materially change capex plans.
At the index level, we do not expect hyperscaler issuance to fundamentally derail the investment-grade credit market. Hyperscalers represented only around 3.5% of the USD investment-grade index at the start of 2026, so their spread widening has added only a few basis points to the broader index. Demand for new issuance outside of technology has remained resilient, and there is no evidence so far that hyperscaler issuance is crowding out other sectors in the investment-grade market. The asset class continues to be supported by high all-in yields and solid corporate fundamentals, which have helped sustain inflows into USD investment-grade credit.
A wider AI opportunity set, with hyperscaler credit hinging on free cash flow
As the AI investment cycle progresses, investors are shifting focus from the creation of the technology to its adoption. We expect the next phase to bring further bouts of volatility as investors assess end-user demand, monetisation and where profits accrue across and within each layer of the AI value chain.
For investors already carrying significant AI exposure in equity portfolios, hyperscaler debt offers a more defensive expression of the same investment theme. Equity investors need AI capex to translate into higher revenues, margins and market share. Bondholders mainly need sufficient cash flow to preserve credit quality and repay debt.
The key driver of hyperscaler spreads will be the path of free cash flow after 2027. Consensus expectations point to improving free cash flow from 2028 onwards, as the data-centre buildout matures and as capex growth is expected to level off towards 2030. That suggests gross debt issuance will continue to rise into 2027 and could peak around 2028. If these consensus expectations are realised, the improvement in free cash flow would be supportive for hyperscaler spreads.
Credit investors are therefore currently pricing a risk premium relative to the baseline scenario. This reflects elevated debt supply, rising off-balance-sheet exposures and considerable uncertainty around the scale and timing of future capex and AI monetisation. As noted earlier, we see this risk premium as warranted, but it also allows investors to earn higher carry. If the recovery in free cash flow comes under pressure or is delayed, hyperscaler spreads could widen further.
AI monetisation is becoming more tangible. In their recent second-quarter earnings reports, the largest cloud providers Alphabet, Amazon and Microsoft reported a combined 48% year-on-year increase in cloud revenue. More of this compute demand is also coming from end users, as inference requests rise relative to demand for training new AI models.
Backlogs for compute demand have increased further and are concentrated among frontier model builders Anthropic and OpenAI. The hyperscaler business model will be resilient in a scenario where overall compute demand meets expectations and pricing power for cloud services remains supported, even if today’s frontier model builders do not turn out to be the long-term winners. From an operational perspective, hyperscalers should be able to reallocate compute capacity relatively easily to other model builders that gain market share. Equities would capture more of the upside in this scenario as returns on earlier investments are realised, but bondholders would still benefit through stronger credit quality.
In a negative scenario where demand or margins for compute services disappoint, hyperscalers’ capital allocation flexibility should mitigate pressure on bondholders to some extent. In this environment, hyperscalers would face lower pricing power, greater scrutiny of return on investment and sunk costs linked to previous investments and off-balance-sheet exposures. Project finance terms are complex and vary materially by transaction. Even so, we expect hyperscalers to retain some room for manoeuvre to protect free cash flow by scaling back planned capex. That flexibility matters for bondholders because it would help limit pressure on credit quality. The downside would likely be greater for equities in this scenario, given their higher sensitivity to future returns.
Market dynamics call for an active approach
In our view, active management can add value within fixed income. Passive fixed income portfolios mechanically drift towards the largest debt issuers, regardless of whether the underlying corporate fundamentals or technicals justify the exposure.
The surge in hyperscaler and AI-related issuance further strengthens this case. Active managers can manage exposures relative to the broader index, position across curves, assess supply-demand technicals and diversify across hyperscalers based on balance-sheet strength, profitability and capital discipline.
Active management also creates scope to assess opportunities beyond traditional corporate bonds. Project finance structures are complex, with terms that need to be analysed on a case-by-case basis. For investors able to underwrite the risks, these structures may offer attractive spread pick-up. On average, hyperscaler-related project finance structures offer spreads around 100 bps wider for investment-grade-rated structures and around 200 bps wider for high-yield-rated structures, relative to equivalent public hyperscaler bonds.