November 20, 2026
शेयर करें

एआई के डेट बूम के लिए टेलीकॉम बस्ट से सीख

कृत्रिम बुद्धिमत्ता (AI) बुनियादी ढांचे का उछाल इक्विटी से डेट फाइनेंसिंग की ओर स्थानांतरित हो रहा है, जिससे 1990 के दशक के उत्तरार्ध के टेलीकॉम क्रैश की यादें ताज़ा हो रही हैं। विशेषज्ञों चेतावनी देते हैं कि यदि नकदी प्रवाह (कैश फ्लो) तालमेल बनाए रखने में विफल रहता है तो कठोर ऋण दायित्व इस क्षेत्र को बाधित कर सकते हैं।

Anthropic's delay highlights AI's dot-com bubble parallels

The funding model for artificial intelligence infrastructure is transitioning away from equity and toward debt, a pivot that veterans of the late-1990s telecom bubble may find familiar.

Stephen Hawking’s “A Brief History of Time” features an old anecdote about a lecturer discussing the cosmos, only for an audience member to claim the universe sits on the back of an enormous turtle. When challenged on what supports that turtle, she responds that it is “turtles all the way down.” Philosophers call this infinite regress—an unbroken chain of unproven assertions where solid footing is never reached.

David Spreng is the founder, co-CEO and co-CIO of Runway Growth Capital.
David Spreng is the founder, co-CEO and co-CIO of Runway Growth Capital.

That concept comes to mind when observing the current AI expansion because its underlying logic operates in the exact same manner. High valuations justify taking on debt, and that borrowed capital constructs new data center capacity. This new capacity is then cited as validation that the initial valuations were correct, creating a circular justification where each premise relies entirely on the one supporting it.

Over the past year, a larger portion of the technology stack has transitioned to debt financing rather than equity. In my view, crossing that threshold signals the moment when the telecom crash becomes a relevant historical parallel, as debt fundamentally alters how these economic cycles conclude.

Until recently, the AI expansion relied heavily on equity, with venture capital investments hitting unprecedented heights.

Data from Crunchbase reveals that worldwide venture funding reached $679 billion during the first three quarters of 2026, marking an all-time high for that timeframe. Artificial intelligence enterprises secured $102 billion during the third quarter alone, representing roughly 64% of all global venture investments. Additionally, a record-breaking 27 firms secured funding rounds of $1 billion or more within that single quarter, making up about one-third of the total volume.

While equity funding carries substantial risks, it typically lacks the rigid repayment schedules associated with debt. When investments underperform, equity stakeholders generally absorb the blow through falling valuations, leaving companies with greater operational flexibility to pursue long-term objectives.

However, large-scale debt has now entered the picture. Last October, Meta issued $30 billion in bonds in what ranked as one of the largest corporate debt offerings in history, before returning in April for another $25 billion—with investors reportedly demanding higher yields on the second issuance. Alphabet, Amazon, and Oracle have likewise incurred tens of billions in debt. Furthermore, Meta’s massive Louisiana data center campus was funded through a record $27 billion bond secured by a joint-venture partner rather than Meta itself. Morgan Stanley projects that leading AI and cloud providers will allocate roughly $3 trillion toward infrastructure up to 2028, with only about half of that sum anticipated to come from internal cash flow. The remaining deficit must be sourced elsewhere, which increasingly points to the private credit and bond markets.

I observed an early iteration of this phenomenon firsthand, initially as a venture capitalist during the late 1990s and subsequently as a lender catering to growth-stage tech businesses. Telecom companies amassed staggering liabilities based on demand assumptions that were echoed widely but rarely verified—most notably the assertion that internet traffic doubled every 100 days.

In reality, traffic was doubling roughly once a year. While still impressive, that rate was only a fraction of what debt projections assumed. By 2002, the telecommunications industry was saddled with approximately $1 trillion in debt—much of which the Federal Communications Commission chairman told Congress would never be repaid—while WorldCom filed for what was then the largest bankruptcy in U.S. history.

A vast amount of the fiber-optic cable laid during that era sat unused for years before being acquired by buyers at pennies on the dollar through bankruptcy proceedings. This occurred amid a decade of corporate restructurings and private mergers that completely reshaped capital structures. Some enterprises survived that period, while many others collapsed.

The demand narrative

The parallel that raises the most concern is the prevailing demand narrative. Today’s conventional wisdom insists that data center capacity is fully leased out before construction even begins, and that the only true risk is underbuilding. That perspective may ultimately prove accurate, but it is repeated with the exact same uniformity that accompanied the 1999 traffic forecasts, and it frequently relies on projections generated by the very companies constructing the facilities. Recently, however, investors have begun pushing back for hard answers.

With Anthropic now public, enterprises of that magnitude must disclose financial results quarterly, enabling investors to test the demand thesis against verified financial statements.

Some critics may fairly contend that this comparison is flawed, noting that the corporations taking on the most debt rank among the most profitable enterprises in history, boasting cash flows that telecom carriers never matched. Furthermore, certain AI companies are scaling their revenue faster than any previous businesses.

The internet did indeed transform society, and the anticipated data traffic eventually materialized. All of that may be true, yet it bypasses the primary question a lender must evaluate: whether cash flow arrives according to the strict timeline mandated by debt obligations. Equity can generally outlast an inaccurate timeline in a way that debt cannot. When borrowed capital builds infrastructure at a pace that outstrips incoming revenue, corrections tend to manifest as formal restructurings rather than quiet asset write-downs.

Enduring lessons

None of this implies that the AI expansion is destined to replicate the exact trajectory of the telecom bust. The most enduring takeaway from the telecom crash relates to capital structure, which ultimately determined which entities remained solvent once demand finally matched capacity—long after the underlying technology had already proven its worth. I have spent years asserting that growth enterprises are excessively fearful of debt, a belief I still hold, but that view remains conditional.

Debt functions effectively for borrowers whose liabilities align with visible, reliable revenue streams, and for lenders who underwrite fundamental business metrics rather than marketing narratives. For any other scenario, debt proves unforgiving. The debt accumulated underneath the AI boom will ultimately be measured against that standard. Every current assumption may prove correct, but the primary lesson from past cycles is that once investors begin questioning the foundational support beneath the market, they rarely relent until they encounter absolute stability.


David Spreng is the founder, co-CEO and co-CIO of Runway Growth Capital. He has more than 30 years of experience as a Silicon Valley venture capitalist and growth debt lender, has been involved in the formation and development of nearly 50 technology companies, and has appeared four times on the Forbes Midas List of top venture capitalists. He co-founded Decathlon Capital Partners in 2010, served on the board of the National Venture Capital Association and has participated as a panelist at the World Economic Forum in Davos.

अक्सर पूछे जाने वाले प्रश्न

01How is the funding model for the AI buildout changing?

The financing of AI infrastructure is shifting away from equity investments and toward heavy reliance on debt and corporate borrowing.

02How much infrastructure spending is expected through 2028?

Morgan Stanley projects that major cloud and AI companies will spend approximately $3 trillion on infrastructure through 2028, with only about half expected to be covered by internal cash flow.

03What lessons does the author draw from the 1990s telecom bubble?

The author highlights that the telecom sector amassed around $1 trillion in debt based on overhyped demand forecasts that failed to materialize on schedule, leading to massive bankruptcies and restructurings.


04अस्वीकरण

Note: This article expresses the author’s views as of the publication date and is provided for general informational and educational purposes. It is not investment advice or a recommendation to purchase, sell, or hold any security or to enter into any financing transaction. References to companies and transactions are illustrative and do not reflect a conclusion regarding their investment merits or creditworthiness. Historical market comparisons are inherently limited, and future conditions may differ materially. The author’s views may change without notice.

चित्रांकन: डोम गुज़मैन

शेयर करें

प्रतिक्रिया दें

आपका ईमेल पता प्रकाशित नहीं किया जाएगा। Required fields are marked *