Hyperscalers on Tour

Programming note:

• Our field report on enterprise AI adoption is now available, based on interviews with three operators with experience at Anthropic, OpenAI, Palantir, Salesforce, ServiceNow and SAP. [Read the report]
• ARPU will return next Tuesday to look at Anthropic's IPO filing.

AI Is a Bond Market Problem Now

There is a standard rite of passage for global rock stars: when you have thoroughly exhausted your domestic market, you take the show overseas. America's hyperscalers have officially embarked on their world tour—except instead of selling concert tickets, they are selling hundreds of billions of dollars in corporate debt across four continents.

The roadshow was born out of necessity. In the United States, the corporate bond market has begun showing signs of acute indigestion. After tech giants poured roughly $220 billion of new bonds into the domestic market over the past year—pushing total AI-related financing past $500 billion in 2026—the domestic pool of capital simply wasn't deep enough to satisfy their appetites. To keep the concrete flowing around Nvidia chips without completely overwhelming Wall Street, Big Tech had to pack its bags and go hunting for foreign currency.

The resulting overseas borrowing binge has shattered financial records wherever the tech giants have landed. In Canada, Amazon smashed the country's all-time corporate borrowing record with a C$14 billion bond deal. In Australia, Alphabet floated an A$5.5 billion package, more than doubling the nation's previous debt record, before turning around and setting a new high-water mark in the Swiss franc market. In London, Alphabet even managed to sell a rare 100-year bond in British pounds, while Meta is preparing to tap European bond markets for the first time this autumn.

This global borrowing spree illustrates how completely the conversation around artificial intelligence has mutated. The debate is no longer an equity analyst's thought experiment about when AI models will generate adequate enterprise returns. It has become a fixed-income crisis about financial plumbing: can the global debt markets absorb trillions of dollars in tech financing without breaking everyone else's cost of capital?

Tiptoeing Around Silicon Valley

The traditional pecking order of global credit markets is very simple. Sovereign governments set the baseline, and everyone else borrows around them. If you were a corporation preparing to issue debt, one of the first things your bankers checked was the calendar: when was the Treasury bringing government bonds to market? You generally didn't want to compete for investors on the same afternoon.

Today, that dynamic has quietly inverted.

Across Europe and Canada, corporate treasurers and sovereign debt managers are now actively tiptoeing around Silicon Valley. In Canada, domestic companies have begun postponing bond sales or shortening their debt maturities specifically to avoid clashing with mega-deals from Amazon and Alphabet, fearing that bond buyers will hoard cash to deploy in the tech deals instead.

Even sovereign officials are voicing uncharacteristic frustration. Federal Reserve Chair Kevin Warsh and Treasury Secretary Scott Bessent have both argued that Big Tech's relentless debt issuance is directly competing for capital with the $31 trillion US Treasury market, adding upward pressure to long-term interest rates.

Across the Atlantic, staff at the European Central Bank published a warning asking whether European financial markets can smoothly handle such concentrated inflows of tech debt without crowding out domestic industrial companies.

The sheer scale of tech's dominance over capital markets explains the anxiety. According to data from S&P Global, technology companies this year have accounted for 60% of all convertible bond issuance, 50% of equity follow-on offerings, and more than a quarter of all US investment-grade corporate bonds.

Silicon Valley is no longer just a sector inside the global capital markets; it is rapidly threatening to swallow them whole.

The Hundred-Year GPU

To understand why the tech giants are fanning out across the globe like sovereign borrowers, you have to look at what has happened to their cash flow.

For decades, the foundational dogma of Big Tech investing was that these companies were asset-light cash-printing machines. They didn't need debt because their monopolies generated more free cash flow than management knew what to do with.

That sacred cow has officially been put to pasture. S&P Global now projects that the six largest hyperscalers will collectively generate negative free operating cash flow through 2026 and 2027, with recovery not expected until 2029 at the earliest. Capital expenditure across the group is projected to exceed $1.3 trillion over the next three years. Even the most prodigious software monopolies cannot fund $800 billion in annual capex purely out of their quarterly earnings.

And so, they borrow. But as the debt pile swells, the nature of the borrowing has begun to look increasingly surreal.

Consider Alphabet's 100-year sterling bond. To fixed-income investors, a century-long bond is a vehicle historically reserved for sovereign empires, ancient universities, or transcontinental railroad trusts. It is an instrument designed for assets that endure across generations.

Alphabet, by contrast, is using that capital to finance an AI infrastructure arms race where the underlying silicon hardware depreciates over five years.

There is an inherent financial curiosity in taking out a 100-year loan to fund a technology that measures its hardware cycles in single digits. The data centers will presumably endure and continue humming along, but by the time that bond matures in the year 2126, the facility will have had to cycle through twenty generations of replacement silicon and cooling retrofits just to keep up with the frontier.

When Big Tech's borrowing begins reshaping sovereign debt auctions in Ottawa, lifting corporate yields in London, and consuming the lion's share of the convertible bond market in New York, the nature of the AI trade has fundamentally changed.

AI is no longer an equity bet on software adoption. It is a macro credit event—and the ultimate pace of the technology may end up being decided less by the limits of silicon, and more by how much debt the global financial system can safely swallow.

Signal Stack

The operating reality beneath the headlines.

• Nvidia Turns to Insurers to Spread the Risk of AI Build-Out (FT) – Nvidia now expects a quarter of next year's revenue to come from AI labs it supports with its own balance sheet, and is approaching insurers to cover losses when neoclouds default and the pledged chips cannot be resold for enough to repay lenders.
• SoftBank Takes On Junk-Bond Debt at Record Yields to Fund OpenAI Ambitions (The Japan Times) – SoftBank raised roughly $11 billion at the highest yields it has ever paid—9.75% on the longest dollar tranche—becoming the world's largest corporate junk-bond borrower to fund commitments approaching $65 billion to OpenAI alone.

You received this message because you are subscribed to ARPU newsletter - you may unsubscribe anytime at the bottom of this email.

If a friend forwarded you this message, sign up here to get it in your inbox.