Thematically, a lot of my pre-CPI coverage this week revolved around the impact of the AI buildout on debt and equity supply.

That was both intentional and not. I don’t get up every morning with an editorial plan, nor do I pursue any manner of agenda. I hit the macro high points and try to say something meaningful about whatever that day’s “top” geopolitical news happens to be (unless the overarching narrative hasn’t shifted, in which case I spare you the recapitulation), and leave the rest to whim and circumstance.

That said, if I know the macro’s going to predominate during the latter half of a given week, I try to do more market-specific color early on, that way there’s a decent balance. This is one of those weeks when the macro’s back-loaded, so I front-loaded the market color. As noted, a lot of that color revolved around how companies are “bankrolling the dream,” as I put it Monday.

The estimated cost of that dream rises every, single week. The five hyper-scalers alone are set to spend nearly $2.5 trillion across 2027 and 2028 on the AI arms race, and that’s after $800 billion in 2026. Those outlays will almost surely exceed 100% of cash flow in 2027, and likely in 2028 as well.

The table above, from Goldman, shows you different scenarios for hyper-scaler capex and cash flow next year. As you can see, consensus expects capex to exceed cash flow by $150 billion in 2027.

As discussed in the linked article above, company analysts guesstimate that Alphabet, Amazon, Meta, Microsoft and Oracle will fund about 35% of capex in 2027 with debt. That’d be about $400 billion in new supply. The same’s expected for 2028.

Since September of 2025 (i.e., since Oracle kicked off this binge 11 months ago), the global IG market’s absorbed around $350 billion of new supply from the hyper-scalers plus Nvidia and SpaceX. Most of that ($290 billion or so) is USD IG.

There’s the “since September” chart, updated with Alphabet’s latest $25 billion deal, which was almost five times oversubscribed.

For 2026, the YTD, hyper-scaler plus SpaceX (but not Nvidia) figure inclusive of high-grade sales and other financing (e.g., loans for data centers) is $270 billion. As Nomura’s Charlie McElligott noted on Tuesday, total corporate debt supply’s up by two-thirds in 2026 versus this time last year, at nearly $867 billion through August 10.

This is coming at a cost for tech firms. The 40-year tranche of the above-mentioned Alphabet deal, for example, promised a 1.3ppt premium over 30-year Treasurys. Although 20bps tighter versus the initial chatter, that compares to a 95bps premium for the longest-tenor portion of the company’s February deal.

The figures above, from McElligott, compare an index of hyper-scaler spreads to the broad IG universe: We’re 30bps+ wide to the market now. Note from the table on the right that at the height of the COVID panic, the hyper-scaler proxy was 35bps rich to the broad high-grade complex reflecting, as noted in parentheses, the flight to quality zeitgeist.

Although the AI buildout’s a boon to growth (“all that capex has to go somewhere,” as the refrain goes), the funding “implications across both debt and equities have led to concerns about legacy tailwinds turning to headwinds,” McElligott wrote, in the same Tuesday piece.

In corporate credit, the read-across is straightforward: The more supply, the more it’ll cost to sell the debt. There’s some risk of spillover there for Treasurys. And vice versa. Hold that thought.

On the equities side, it’s not so much a matter of the stock effect on buybacks, but rather the flow effect. In other words: Yes, overall buybacks may continue to ht records, but incremental demand from a critical cohort, namely four of the mega-caps, will slow. The figure below’s rather stark in that regard.

Net buybacks — i.e., buybacks less issuance — are negative to the tune of almost $150 billion YTD once you incorporate Alphabet’s equity raise and SpaceX’s IPO.

The cash flow burn from the AI buildout may have a “negative impact on US equities as a function of reduced demand, at the very least from a flow perspective as the ‘corporate buybacks as a perpetual equities demand mechanism’ dynamic is likely set to pause or even inflect lower,” McElligott went on.

Coming back to the interplay between IG debt sales to fund AI expenditures and Treasury’s attempts to optimize issuance such that taxpayers get the best deal, it’s worth asking if Scott Bessent’s feeling any pressure to ensure the US government doesn’t accidentally make it more expensive for the country’s tech titans to stay ahead in what’s widely viewed as a race with existential national security implications.

On that point, McElligott quoted his colleague Jon Cohn. “Layering on AI / data center issuance in 2025-26 shows a sizable shift, one that adds back much of the duration supply that Treasury offset with its bill-heavy lean,” Cohn wrote. “If this AI supply is indeed structural, Treasury may decide that increasing issuance of long-end high grade supply into a well- or over-supplied market is not in the best interest of the taxpayer or the country, given AI national security considerations, and reducing issuance is the better course of action.”

Now that’s food for thought, and it adds a new dimension to the ostensibly innocuous language tweak in last week’s QRA.