The financing structure behind the artificial intelligence buildout changed in a way that is easy to miss because the numbers remain impressive on the surface. Global venture funding through the first three quarters of 2026 totaled $679 billion, the highest amount ever recorded for that period, according to Crunchbase data. AI companies alone drew $102 billion in the third quarter, roughly 64 percent of all global venture investment. Twenty-seven companies raised rounds of $1 billion or more during the quarter.
But the equity story, large as it is, now shares the stage with something structurally different. Meta sold $30 billion of bonds last October in one of the largest corporate debt sales on record, then returned for $25 billion more in April, with investors reportedly demanding more compensation the second time around. Amazon, Alphabet and Oracle have each borrowed tens of billions of their own. A joint-venture partner of Meta's, not Meta itself, raised a record $27 billion bond to finance a data-center campus in Louisiana. Morgan Stanley expects the largest AI and cloud companies to spend roughly $3 trillion on infrastructure through 2028, with only about half expected to come from their own cash flow.
The distinction between equity and debt matters less when times are good and more when they are not. An equity investor who backs a company that underperforms sees the value of that investment decline. The loss is real, but no clock is running. A bondholder is in a different position: repayment is contractual, the schedule is fixed, and the capacity being built with borrowed money has to generate enough revenue to service that debt regardless of whether the demand forecast turns out to be accurate.
That is the mechanism worth understanding. The AI buildout's internal logic, as one observer described it, runs like this: valuations support borrowing, borrowing builds capacity, and the capacity is then treated as evidence that the valuations were correct. Each claim stands on the one beneath it. As long as demand materializes at the pace the forecasts assume, the structure holds. If it does not, the bondholders are the ones holding fixed obligations against assets whose value was premised on projections that did not arrive.
This is not a novel dynamic. The telecom buildout of the late 1990s followed a similar path, with companies borrowing against demand forecasts that were widely circulated and rarely interrogated. What the available evidence cannot establish is whether AI infrastructure demand will disappoint at the same scale, or at all. The comparison is an observation about structure, not a forecast about outcomes.
What the numbers do establish is that the risk transfer has already occurred. When the AI buildout was primarily equity-financed, the people absorbing downside were venture investors who understood they were taking that risk. As the financing mix shifts toward bonds and private credit, the population carrying the downside changes. Some of those bondholders are institutional investors with sophisticated risk teams. Others are pension funds, insurance companies and retail investors in bond funds who may be less focused on what a specific Louisiana data-center campus needs to earn in order to cover its debt service. The asset is the same. The person holding the obligation to be repaid is not.
AI and cloud companies are increasingly financing infrastructure through bond issuances rather than relying primarily on equity investment and internal cash flow. Morgan Stanley expects the largest AI and cloud companies to spend roughly $3 trillion on infrastructure through 2028, with only about half expected to come from their own cash flow, meaning the remaining half depends on borrowed capital.
An equity investor backing an underperforming company experiences declining investment value with no fixed repayment deadline. A bondholder faces contractual repayment obligations on a fixed schedule, requiring that capacity financed with borrowed money generate sufficient revenue to service debt regardless of whether demand forecasts prove accurate.
According to James Harrington of Gambity, the AI buildout's internal logic runs as follows: valuations support borrowing, borrowing builds capacity, and capacity is then treated as evidence that the valuations were correct. Each claim stands on the one beneath it, creating dependency on demand materializing at forecasted pace.
The telecom buildout of the late 1990s followed a similar path, with companies borrowing against demand forecasts that were widely circulated and rarely interrogated. The comparison is an observation about structural similarity, not a forecast about whether AI infrastructure demand will disappoint at comparable scale.
Continue reading.
Fourth Circuit preemption ruling reshapes state enforcement risk
Utah attorney general charges Kalshi founders with crimes
Kalshi Donates $1 Million to New York Housing Authority
Perez's Winning Streak Ends Debate on Gaming Mentor Markets
A few more worlds
worth exploring.
From the FreeMalta ecosystem.
The AI Journey
Explore AI beyond the headlines. Tools, ideas and a place to start.
Explore the journey 02 / THE GARAGEThe Garage
The companies behind the names. Explore their stories and the people building them.
Open the garage 03 / THE DUELThe Duel
Two contenders. A closer look. Put products side by side before choosing.
Choose your duel 04 / MALTAINSIDERMaltainsider
Discover the B2B side of Malta. A different lens on business and opportunity.
Meet Maltainsider