The CEO of a recently sold data center firm warns that current AI infrastructure spending mirrors the speculative excesses of the late 1990s dot-com era.
Marc Ganzi has spent three decades transforming digital infrastructure into an institutional asset class. This week, as SoftBank’s acquisition of his firm approaches completion, Ganzi told CNBC that the AI infrastructure market is experiencing a “toppy-esque moment” that feels “very similar to the late 1990s.” During that era, companies borrowed heavily to build fiber-optic networks before demand materialized. Today, Ganzi sees parallel dynamics in AI infrastructure, though he maintains that the current buildout remains “another iteration in the evolution of technology.”
On December 29, 2025, Ganzi agreed to sell DigitalBridge Group (NYSE:DBRG) to SoftBank Group Corp. for $16.00 per share in an all-cash transaction valued at $4 billion. Shareholders approved the deal in April 2026. DBRG shares last traded at $15.96, closely tracking the acquisition price and reflecting a 42.48% gain over the past year. With Ganzi’s exit secured, his recent warnings are directed at investors still navigating the market.
Ganzi outlined a bifurcated data center landscape. DigitalBridge typically maintains a loan-to-value (LTV) ratio near 45% across its portfolio, whereas newer competitors are advancing into the 70% to 80% range—a risk profile Ganzi likened to altitude sickness. He drew a clear distinction between investment-grade tenants locked into long-term leases, which he termed “islands of safety,” and unrated operators relying on aggressive private credit structures.
These concerns were not new. During DigitalBridge’s Q3 2025 earnings call, Ganzi cautioned that the upcoming cycle would be “marked by a lot of amateurs and a lot of tourists in the next 24 to 36 months.” He noted that “these gigawatt projects are really tough,” emphasizing that clients can easily distinguish between first-time developers and established firms with “over 400 data centers and 11 different companies.” DigitalBridge’s scale underscores this advantage. As of Q1 2026, the firm reported $40.80 billion in fee-earning equity under management. In Q3 2025 alone, it leased a record 2.6 gigawatts, accounting for roughly one-third of U.S. hyperscale leasing during that period. Additionally, DigitalBridge-backed Vantage Data Centers is constructing Frontier, a $25 billion, 1.4-gigawatt Texas campus contracted to Oracle and OpenAI’s Stargate program—factors that clearly attracted SoftBank’s attention.
Ganzi reserved his sharpest critique for the financing architecture surrounding NVIDIA (NASDAQ:NVDA). He described the capital backing NVIDIA’s chip-financing initiative as “priced to perfection,” meaning every underlying assumption must hold for the structure to remain viable. On its most recent earnings call, NVIDIA disclosed partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise over $500 billion in third-party capital. Under this model, NVIDIA provides “a take or pay commitment on a portion of the facility’s capacity, a minimum revenue guarantee that gives lenders the confidence to underwrite the project.” CFO Colette Kress acknowledged the controversy, noting that “some will call this circular financing,” while maintaining the company views it differently.
The scale of NVIDIA’s operations necessitates this financing approach. In its Q2 FY2027 report, NVIDIA posted revenue of $96.22 billion, a 105.8% year-over-year increase, with data center revenue reaching $89.02 billion, up 117%. Supply obligations surged to $279.00 billion. Trading at approximately 45 times earnings and up 896.2% over five years, NVIDIA’s valuation metrics provide further context for Ganzi’s late-1990s comparison.
Ganzi’s analysis clarifies where risk actually resides within the AI infrastructure sector. A campus leased for years to an investment-grade tenant and financed with moderate debt already possesses a customer and a predictable revenue stream. Conversely, a first-time developer borrowing 70% to 80% of a project’s value for an unrated operator requires significantly more variables to align correctly. The power, cooling, and networking suppliers supporting these campuses represent a distinct industry, detailed in a separate analysis of AI infrastructure beyond semiconductor manufacturers. Ganzi is not suggesting that every data center represents a bubble. Rather, he warns that leverage will ultimately determine which projects can withstand shifting market conditions.
For questions or corrections, contact [email protected]. Gerelyn Terzo is the author of the dividend investing handbook *Dividend Investing Strategies: How to Have Your Cake & Eat It Too.* A veteran financial journalist, she covers agri-finance for outlets such as Global AgInvesting and broader stock market and personal finance topics for 24/7 Wall Street. She began her career at CNBC and later helped launch Fox Business in New York. Gerelyn currently resides in Woodland Park, Colorado, where she pursues nature photography as a hobby.