The Bill Arrives

Since our earliest issues, Monard's investment committee has held a consistent view: the capital intensity of the AI and electrification build-out would eventually collide with the limits of corporate balance sheets, and that collision would show up first in the price of debt rather than in the headlines about capex. Issue V set out the mechanics of that thesis in detail, arguing that even the most profitable companies in history could not indefinitely fund compute, power and water infrastructure out of operating cash flow alone.

This issue returns to that thesis and tests it directly against the bond market — both sovereign and corporate. The short answer is that the data supports our position, and more emphatically than we expected twelve months ago. Corporate bond issuance tied to AI infrastructure has reached levels without precedent, government yields have moved higher on the back of fiscal and inflation pressure, and the price investors demand to hold hyperscaler debt has risen across every maturity bucket we examined.

Government Yields: The Risk-Free Rate Is No Longer Free of Pressure

The starting point for any credit analysis is the risk-free rate, because every corporate spread is priced on top of it. Twelve months ago the case for a benign government bond backdrop was reasonable. It no longer is. The US 10-year Treasury yield has traded in a 4.0 to 4.5 per cent range through most of 2026, and by early August had pushed through 4.6 per cent — close to its highest level of the year.

The move is not simply noise. Core inflation has remained stubbornly above the Federal Reserve's target, and the Fed under Chair Kevin Warsh has signalled it would consider a rate rise in September if incoming inflation data stays firm — a marked change in tone from the rate-cutting narrative that dominated markets through 2024 and 2025. Markets were pricing a roughly even chance of a September hike at the time of writing. Layered on top of monetary policy is a fiscal picture that is doing its own work on yields. National debt has continued its climb past US$38 trillion, and last year's tax and spending package, while supportive of near-term growth, has widened the deficit trajectory that bond investors must absorb through new issuance every quarter.

The practical effect for our thesis is straightforward. A corporate borrower today is not just contending with a wider spread over the risk-free rate. It is contending with a risk-free rate itself that sits meaningfully higher than the environment in which the AI capex cycle began. Both components of the cost of debt are moving in the same direction, and both are moving against the borrower.

Corporate Issuance: A Borrowing Binge Without Modern Precedent

Overall US corporate bond issuance is forecast to reach approximately US$2.46 trillion in 2026, up nearly 12 per cent on 2025's US$2.2 trillion, according to Barclays. The more instructive figure sits inside it. The five hyperscalers at the centre of the AI buildout — Amazon, Alphabet, Meta, Microsoft and Oracle — issued around US$28 billion a year in US corporate bonds on average between 2020 and 2024. In 2025 that jumped to US$121 billion. Through the first half of 2026 alone the group had already issued more than that full-year 2025 total, and forecasters including Goldman Sachs, UBS and Bank of America have lifted full-year 2026 estimates for the group into a US$230 billion to US$250 billion range — with some estimates for total AI-linked global debt issuance running as high as US$570 billion once financing structures beyond the hyperscalers themselves are included.

The individual transactions illustrate the scale better than the aggregate figures. Meta's US$30 billion offering in October 2025 was, at the time, the largest single-issuer high-grade bond sale on record outside a merger financing. Amazon followed in March 2026 with an approximately US$54 billion multi-tranche deal, large enough that Bank of America revised its full-year forecast for the group upward on the announcement alone. By October 2025, AI-linked debt had already become the largest single segment of the US investment-grade market, at roughly 14 per cent of the benchmark high-grade index, surpassing the US banking sector for the first time on record according to M&G Investments.

None of this reflects distress. Every one of the hyperscalers involved retains a strong investment-grade rating, and analysts are consistent in describing the borrowing as a choice rather than a necessity. Hyperscaler capital expenditure is now consuming close to 100 per cent of operating cash flow for the group, up from a historical average closer to 40 per cent, and management teams have chosen to fund the gap through debt markets rather than cut spending, slow the buildout or sell assets. That is precisely the dynamic Monard flagged in Issue V. Borrowing at this scale preserves flexibility on buybacks, acquisitions and pace of construction — but it does so by drawing directly on a finite pool of investor capital that must now be shared with every other issuer in the market.

The Price of That Debt Is Rising — Across Every Rating and Maturity

Our second test was the one Monard set out to answer directly: is the market simply absorbing this supply at an unchanged price, or is it beginning to charge more for it? The evidence points clearly to the latter. A Reuters analysis of LSEG pricing data published in late July found that for Amazon, Alphabet, Meta and Oracle, the median new-issue spread over the risk-free rate on 2 to 4-year bonds rose to 40 basis points in 2026 from 30 basis points in 2025. On 5 to 7-year debt the median spread rose to 60 basis points from 50, and on bonds maturing beyond 20 years — the segment most exposed to long-run uncertainty about AI returns — the median spread widened to 118 basis points from 108.5.

Secondary market pricing tells the same story with more force. Of 91 hyperscaler bonds issued in 2026 with comparable pricing data, 78 were trading at wider yields on 28 July than at their issue date — a median widening of roughly 22 basis points. Bid-to-cover ratios on hyperscaler paper fell from close to 5 times in February 2026 to below 2 times by July, a far sharper decline than the roughly half-point slippage seen across the investment-grade market as a whole over the same period, according to analysis cited by Fortune. JPMorgan strategists described the widening as the market rationally pricing in an accelerating pace of issuance rather than any deterioration in credit quality — a framing we broadly accept. Whether the driver is credit risk or simple oversupply, the marginal cost of every new dollar of hyperscaler debt is rising.

There is a second-order effect worth noting. Apollo's academy has highlighted a growing divergence between hyperscaler and industrial spreads, with hyperscaler spreads widening even as industrial spreads have tightened over the same window. As hyperscalers absorb an increasingly large share of investment-grade issuance capacity and investor attention, smaller or lower-rated issuers are being crowded toward the edges of the market, competing for a shrinking share of demand at a time when the largest borrowers in the index are also the newest and least tested at this scale.

The Second-Order Problem: Rising Rates Lift the Hurdle for Every Future Project

There is a further consequence of this repricing that sits above the bond desk and lands directly on the boardroom table. The weighted average cost of capital (WACC) blends a company's cost of debt and its cost of equity, weighted by how much of each it uses to fund itself. We have just shown that the cost of debt component is rising — both because the risk-free rate underneath it has moved higher and because spreads over that rate have widened. Cost of equity tends to move in the same direction for a related reason: investors who can now earn 4.6 per cent doing nothing in Treasuries expect a correspondingly higher return for taking on equity risk. When both inputs rise together, WACC rises — and WACC is not an abstract accounting figure. It is the discount rate companies use to test whether a proposed investment is worth making at all.

The 2026 AFP Cost of Capital Survey, which benchmarks nearly 300 corporate treasury and finance practitioners, found that 62 per cent of organisations use their calculated cost of capital as their standard hurdle rate outright, while the remaining 38 per cent set the bar higher still to reflect project-specific risk. Roughly half said they adjust their hurdle rate in response to market conditions, interest rates and macroeconomic shifts — precisely the conditions in play through 2026. For capital-intensive sectors such as utilities and industrials, PwC's deals research puts typical published hurdle rates in a 10 to 14 per cent range.

The mechanical effect is unforgiving. A project's own cash flows do not need to deteriorate at all for it to stop clearing the bar; the bar simply has to move. An asset that comfortably cleared a 9 per cent hurdle a year ago can destroy value on paper at 11 per cent, with nothing about the asset itself having changed. That creates a genuine governance dilemma for any board sitting on a finite pool of capital — every dollar allocated to a marginal infrastructure upgrade is a dollar not allocated to core revenue-generating investment.

That is precisely the category of project most at risk of being starved of capital under a rising WACC: not the discretionary ones, which are easy to defer, but the critical ones that do not generate a standalone commercial return. Grid connection upgrades, backup power, water treatment capacity and emissions compliance infrastructure sit in exactly this bracket. They are essential to a company's licence to operate, yet they rarely clear a 10 to 14 per cent hurdle on their own economics, and they now have to compete for capital against higher-yielding core investment at a moment when the cost of capital itself has moved against them.