By David Spreng, Founder, Co-CEO, and Co-CIO of Runway Growth Capital
Executive Overview
The financing architecture underpinning the artificial intelligence buildout is undergoing a profound structural evolution. For years, the unprecedented capital requirements needed to fuel the generative AI revolution were borne primarily by equity—relying on the deep pockets of venture capitalists, massive corporate balance sheets, and historical rounds of private funding.
However, a pivotal transition is now underway. Increasingly, the funding mechanism of choice is shifting from equity to debt. Tech giants, cloud providers, and specialized AI infrastructure developers are leaning heavily into the corporate bond market and private credit to sustain their astronomical capital expenditures.
This pivot from ownership-based risk to fixed obligations carries historical echoes. For market participants who weathered the telecommunications and dot-com bubbles of the late 1990s and early 2000s, the optics of this transition are strikingly familiar. In physics and philosophy, there is an old parable—famously cited in Stephen Hawking’s A Brief History of Time—about an infinite regress of supporting layers, famously summarized as "turtles all the way down."
Today, the logic of the AI infrastructure boom risks adopting that exact shape: valuations justify borrowing, borrowing constructs massive data center capacity, and the existence of that capacity is subsequently cited as empirical proof that the initial valuations were correct. Each layer of the financial stack rests entirely upon the one beneath it, postponing the day of reckoning.
While the transformative potential of artificial intelligence is undeniable—just as the internet fundamentally altered global commerce—the underlying credit dynamics demand careful scrutiny. When borrowed money builds infrastructure faster than predictable revenue arrives to service the debt, the market correction typically bypasses quiet valuation markdowns in favor of painful, debt-driven restructurings.
Detailed Chronology: From Venture Equity to Massive Bond Issuances
To understand the magnitude of the current shift, one must examine how the funding mechanisms of the AI boom have evolved over the past several years.
The Equity-Fueled Genesis
Until recently, the AI buildout was an equity-driven narrative. Backed by record levels of dry powder, venture capitalists and institutional investors poured staggering sums into the ecosystem. According to Crunchbase data, global venture funding reached a staggering $679 billion through the first three quarters of a single recent fiscal year—representing the highest aggregate total ever recorded for that timeframe.
AI-focused startups and infrastructure players commanded an immense share of this capital, capturing $102 billion in the third quarter alone, or roughly 64% of total worldwide venture funding. The concentration of capital was equally striking: a record 27 companies secured financing rounds of $1 billion or more within that single quarter, collectively accounting for about one-third of global venture dollars.
Equity financing, while fraught with valuation risks, possesses a critical safety valve: it generally does not impose fixed, mandatory repayment schedules. When high-growth investments underperform or take longer than expected to commercialize, equity investors absorb the blow through markdowns and declining valuations. Meanwhile, portfolio companies retain the operational flexibility needed to pivot, burn through cash at manageable rates, and execute long-term strategic plans without the looming threat of default.
The Arrival of Jumbo Debt
That landscape has transformed dramatically with the arrival of institutional debt at scale. Big Tech and enterprise cloud players have turned to the corporate bond market to finance capital expenditures that far outstrip even their formidable cash flows.
- Meta Platforms: In October of last year, Meta executed a $30 billion, six-part bond offering—marking one of the largest corporate debt sales on record. Seeking additional capital to feed its relentless AI spending push, Meta returned to the debt markets months later for an additional $25 billion, with institutional investors demanding notably higher yields and risk premiums on the second tranche.
- The Hyperscalers: Industry titans including Amazon, Alphabet, and Oracle have each borrowed tens of billions of dollars to fund server farms, silicon procurement, and specialized data center real estate.
- Off-Balance-Sheet Structures: The debt burden is also shifting outside traditional corporate balance sheets. For instance, Meta’s massive Louisiana data center campus was financed not via direct corporate borrowing, but through a record $27 billion bond issue orchestrated by a joint-venture partner.
Wall Street projections underscore the trajectory of this spending. Analysts at Morgan Stanley estimate that the leading AI and cloud infrastructure providers will spend approximately $3 trillion on data centers and hardware through 2028. Crucially, internal cash flow is expected to cover only about half of this outlay. The remaining trillions must be sourced externally—and increasingly, that capital is being drawn from the bond markets and private credit funds.
Supporting Context & Metrics: Echoes of the 2000 Telecom Bust
Having witnessed the telecommunications bubble from a front-row seat as a Silicon Valley venture capitalist and growth-debt lender in the late 1990s, the current market dynamics evoke powerful historical parallels.
The Fallacy of Exponential Demand Narratives
During the telecom boom, carriers borrowed astronomical sums against demand forecasts that were endlessly repeated across Wall Street yet rarely subjected to rigorous verification. The most famous mantra of the era claimed that internet traffic was "doubling every 100 days."

In reality, global internet traffic was doubling roughly once a year—still an extraordinary rate of organic growth, but a mere fraction of the hyperbolic projections that justified the trillions in capital expenditures. By 2002, the telecommunications sector was saddled with roughly $1 trillion in debt. The chairman of the Federal Communications Commission (FCC) later testified before Congress that a vast majority of those borrowings would never be repaid. The fallout culminated in massive corporate collapses, including WorldCom filing for what was then the largest bankruptcy in U.S. history.
The structural legacy of that crash is instructive. Much of the fiber-optic cable laid during the frenzy sat dark for years. Eventually, it was put to productive use, but only after new buyers acquired the assets for pennies on the dollar through bankruptcy proceedings, multi-year restructurings, and sweeping corporate reorganizations that completely wiped out existing equity holders. Some pioneers survived the purge; many did not.
The Modern Parallels in AI Infrastructure
Today’s AI buildout features a strikingly similar demand narrative: the prevailing consensus holds that data centers are fully leased before their foundations are even poured, and that the only existential risk is under-building.
While this thesis may ultimately prove correct, the uniformity with which the narrative is repeated mirrors the 1999 telecom traffic forecasts. Much of this optimism relies on internal projections supplied by the very enterprises constructing the facilities. However, market dynamics are shifting. As foundational AI players like Anthropic mature and transition into public markets—or face regular financial disclosures—institutional investors are increasingly demanding hard, audited numbers to substantiate the demand thesis.
Skeptics of this comparison argue that today’s major borrowers are fundamentally different from late-90s telecom carriers. Companies like Meta, Alphabet, and Amazon represent some of the most profitable enterprises in human history, generating organic cash flows that dwarf anything the telecom sector ever approached. Furthermore, leading AI companies are scaling revenues faster than virtually any cohort of businesses in commercial history.
The internet did, ultimately, change everything, and the projected data traffic did materialize. Yet, from a lender’s perspective, these historical truths miss the core vulnerability: whether cash flows arrive on the strict schedule demanded by fixed-income obligations. Equity can endure a prolonged timeline and absorb strategic miscalculations; debt cannot. When borrowed capital constructs infrastructure faster than paying customers arrive to service the debt, corrections rarely manifest as gentle valuation markdowns. They materialize as formal restructurings.
Official Statements & Industry Perspectives
The structural transition from equity cushions to debt-laden capital stacks has elicited intense debate among institutional financiers, credit strategists, and policy observers.
Industry analysts emphasize that while the revenue potential of generative enterprise software is vast, the sheer velocity of capital expenditure creates unique credit exposure. As Morgan Stanley and other major financial institutions project trillions in infrastructure spending over the coming years, credit underwriters are scrutinizing the debt-servicing capabilities of second-tier AI players who lack the multi-billion-dollar cash reserves of Big Tech oligopolists.
Venture and private credit veterans have consistently cautioned against the indiscriminate use of leverage during technological inflection points. While growth-stage companies historically harbor an institutional fear of debt, seasoned lenders argue that leverage is a powerful tool only when strictly calibrated to visible, near-term cash flows rather than speculative, long-term market capture.
Furthermore, regulatory bodies and market watchdogs have begun monitoring the systemic linkages between private credit markets—which are increasingly stepping in to fund tech infrastructure—and traditional banking institutions. The concern centers on whether over-leveraged joint ventures and off-balance-sheet special purpose vehicles (SPVs) could introduce hidden contagion risks if data center utilization rates fall short of aggressive underwriting models.
Future Outlook: Navigating the Leverage Stack
None of this analysis should be interpreted as an inevitable prediction that the AI infrastructure boom will mirror the catastrophic demise of the telecom sector. The underlying technologies powering modern machine learning are driving tangible productivity gains across healthcare, logistics, software engineering, and enterprise workflows in ways that far exceed the utility of early fiber networks.
Instead, the most enduring lesson of the telecom bust concerns capital structure. Long after a transformative technology has been fully vindicated by history, it is the underlying capital stack that determines which enterprises survive the intervening turbulence.
For the modern AI economy, the road ahead will be defined by three critical considerations:
- Underwriting Discipline: Borrowers and lenders must ensure that debt obligations are sized strictly to observable, contracted revenues rather than speculative adoption curves. Lenders must evaluate fundamentals rather than relying solely on macro growth narratives.
- Transparency and Audited Metrics: As AI native players scale and face public market scrutiny, transparency regarding unit economics, inference costs, and customer retention rates will separate sustainable business models from capital-burning anomalies.
- The Resilience of Capital Stacks: When investors inevitably begin questioning what supports the layers of the financial stack—checking what the "turtles are standing on"—market resilience will depend entirely on whether equity cushions remain deep enough to absorb the friction of adoption delays.
Debt remains a formidable instrument: it accelerates growth for those who manage it prudently, but it treats overextended balance holders with absolute severity. The ultimate legacy of the AI debt boom will not be judged by the brilliance of the underlying algorithms, but by the structural discipline of the balance sheets built to support them.
