The Equity/Debt Divide in AI

Programming note: ARPU will return on 23 February and examine the market narratives dominating the hyperscalers.

Century of Faith

The standard timeline for the AI economy is usually measured in months. We wait for the next model iteration, the latest GPU shipping estimates, or the next quarter's cloud revenue growth. But this week, Alphabet decided to test a different time horizon: a century.

Here's Reuters:

Alphabet on Tuesday sold a rare 100-year bond [as] part of a $31.51 billion global bond raise...[The company's] century bond was met with demand nearly ten times the one billion pounds sought, according to IFR data. It comes with a 6.05% yield.

Some analysts said Big Tech's greater use of debt reflects a pivot from asset-light models toward long-term infrastructure.

"...this deal shows that, at least for now, investors are willing to take on very long-dated risk tied to AI investment," said Lale Akoner, global market analyst at eToro.

Historically, century bonds belonged to sovereigns and railroads—entities that expected to outlast the century because they owned either the law or the land. By issuing one, Alphabet is adopting the financial posture of a permanent utility. The signal to the market is that the AI race is no longer just about algorithms; it's about owning the physical estate on which those algorithms run, and acting like a landlord to service the debt.

But if you are going to act like a sovereign landlord, you need investors who believe in the long-term value of your kingdom. It turns out that faith is currently split into two very different camps.

The AI Schism

There is a profound clash between the narratives being told in the equity and debt markets these days. The Bank for International Settlements (BIS) recently identified what it calls a "schism," which is a tale of two very different bets:

• Equity Bet: Equity investors are paying for the miracle. They price AI companies at a massive premium, assuming a structural labor displacement by AI will arrive and justify the multi-trillion dollar build-out.
• Debt Bet: Lenders are pricing the risk as if AI is a stable, utility-like business. They charge AI firms the same interest rate spreads they charge non-AI firms, treating a data center with the same indifference they apply to a shopping mall.

This is the schism in a nutshell: the market is simultaneously pricing in both a moonshot and business-as-usual. The BIS notes that one of these narratives must be wrong:

This schism suggests that either lenders may be underestimating the risks of AI investments (just as their exposures are growing significantly) or equity markets may be overestimating the future cash flows AI could generate.

The Debt-Fueled Cycle

If the equity market is right about the miracle, the logical move for hyperscalers is to spend as much as possible, as fast as possible. But there is a physical limit to how much of this can be funded out of a company's own wallet.

Historically, Big Tech companies operated with almost no debt because they were "asset-light." But the capital requirements for a gigawatt of compute are so vast ($40-$50 billion/GW) that even the largest hyperscalers are hitting their limits.

As Goldman Sachs’ Kash Rangan noted in a recent report, we are now witnessing the birth of a "debt-fueled capital cycle":

Most of the capital deployed to fund AI projects has so far come from hyperscalers utilizing cash flows from their core businesses. But now, entities are being funded with 80% debt and 20% equity, with the equity portion often backed by collateral from the sponsoring entity.

Oracle recently completed an $18bn bond sale to fund its AI ambitions, and non-hyperscalers like CoreWeave have also secured significant debt financing. So, leverage is starting to emerge in the system, making it even more vital that the firms driving the need for capital hit their revenue and earnings targets.

(That last part is a polite way of saying that for three years, AI was a "cool research project," but now that there are interest payments due, it had better start making money).

Crowding Out the Treasury

This shift to debt is reaching a scale that is starting to warp the broader financial system. Earlier this week, the Dallas Fed analyzed how AI is impacting the interest rate market, and the numbers begin to rival government spending.

Wall Street estimates suggest that AI-related investment-grade bond issuance could hit $300 billion this year. To put that in perspective, that is roughly one-eighth of the entire duration supply of U.S. Treasury issuance.

The Dallas Fed identifies a few duration supply channels to watch:

• Crowding Out: AI firms are now competing so aggressively for long-term capital that they may crowd out financial firms that usually dominate the bond market.
• Synthetic Duration: Because private credit lenders (the "shadow banks" of the AI boom) prefer floating-rate loans, data center operators are frantically using interest rate swaps to lock in their costs.

In other words, AI is not just a tech trend; it is a major macroeconomic force that is competing with the U.S. Treasury for the world's savings. We have moved from disrupting industries to disrupting the yield curve.

One Hundred Years of No Promises

A telling detail of Alphabet's $31.5 billion raise is the lack of covenants.

Normally, if you lend someone money for a century, you include "restrictive covenants"—legal promises that the company won't take on too much debt. But Alphabet's bonds have no meaningful restrictive covenants. They aren't even guaranteed by Alphabet's subsidiaries.

Alphabet's pitch to the credit market is essentially: We are building the future of intelligence, and that future is going to cost up to $5 trillion over the next few years. You can either fund it on our terms—with no rules and a 100-year wait—or you can find someone else to lend to.

Lenders looked at that pitch and offered ten times more money than Alphabet asked for. Whether the productivity gains ever show up to pay back the debt is a question for the year 2126. For now, the chips are being delivered, the concrete is being poured, and the check is in the mail. It's just a very long delivery route.

More on AI Financing:

• How AI debt financing impacts duration supply and interest rates (Dallas Fed)
• Oracle's monster $25 billion debt financing points to anxieties around AI funding (Morningstar)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

Google Gains EU Antitrust Approval for $32B Wiz Acquisition

• The Story: The European Commission has granted unconditional approval for Google's $32 billion acquisition of cybersecurity firm Wiz, concluding the deal does not threaten competition given Google's third-place position in the cloud market. (Reuters)
• The Investment Implication: Unconditional EU approval for Google's largest-ever deal is a major strategic win that significantly de-risks its cloud expansion strategy. For investors, this confirms that Google can still use major M&A to close the competitive gap with AWS and Microsoft, while the lack of imposed remedies means Google can fully integrate Wiz's technology without operational restrictions, accelerating its roadmap for high-margin cloud security services.

Cisco Launches Silicon One G300 to Challenge Broadcom and Nvidia in AI Networking

• The Story: Cisco unveiled its Silicon One G300 switch chip, a 3nm processor designed to manage the massive traffic spikes in AI data centers, claiming it can complete AI computing jobs 28% faster by automating data re-routing. (Reuters)
• The Product Implication: Cisco is making a direct bid to recapture relevance in the AI infrastructure stack by targeting the networking bottleneck. By focusing on "end-to-end efficiency" and microsecond-level automated re-routing, Cisco is attempting to differentiate its hardware from Broadcom’s Tomahawk and Nvidia’s networking solutions, positioning itself as the superior choice for hyperscalers managing clusters with hundreds of thousands of chip links.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.