CEO Who Just Sold His Data Center Firm to SoftBank for $4 Billion Is Warning of a Late 1990s Moment

Marc Ganzi—having just agreed to sell DigitalBridge Group to SoftBank in a $4 billion all-cash deal at $16.00/share—warns AI data center buildout is in a “toppy-esque” late-1990s moment. He highlights a leverage split: DigitalBridge keeps LTV near ~45% versus newer entrants at ~70%–80% LTV, arguing the latter relies on more financing “pieces” that must go right. He also points to NVIDIA’s chip-financing program raising $500B+ in third-party capital under take-or-pay/minimum revenue structures, underscoring how “priced to perfection” assumptions can leave highly levered, unrated operators most exposed if demand softens.

Analysis

The key market signal is not "AI demand is slowing"; it is that the marginal unit of growth is migrating from equity-funded hyperscale capex to leverage-dependent project finance. That shifts upside away from operating owners and toward capital providers with dry powder, while making junior equity, unrated tenants, and first-time developers the weakest links when utilization, power delivery, or refinancing terms slip. In that setup, the first casualty is usually not revenue — it is residual value and cost of capital.

Near term, the tape can still reward the biggest infrastructure enablers if reported backlog and customer commitments remain intact, but the risk is hidden in the financing stack rather than the demand narrative. If private credit conditions tighten over the next 1-3 months, the market should start differentiating between institutional-grade campuses and levered speculative builds; that is where multiple dispersion will show up first. The most exposed names are the ones relying on aggressive LTV and opaque tenant quality, while large allocators such as BLK, BX, KKR, and GS can benefit from underwriting fees, rescue capital, and eventual distressed asset purchases.

The contrarian point is that the consensus may be underpricing how much of the AI trade is being front-loaded through vendor support and structured financing. NVDA can remain fundamentally strong while still seeing a valuation reset if investors conclude that some portion of demand is effectively financed demand, not durable end-demand. The thesis is falsified if project conversion stays on schedule, spreads remain tight, and 6-12 month forward capacity gets absorbed without covenant stress; otherwise, the second-order losers show up well before any headline slowdown.

Key Decisions for Investors

• Buy BLK / KKR on weakness as a 3-6 month relative-value long against the more levered AI infrastructure complex; these franchises monetize volume and stress, while the fragile capital sits in project finance. Risk/reward improves if private credit spreads widen 50-75 bps.
• Initiate a 6-9 month NVDA put spread or collar on existing exposure to hedge multiple compression from "financed demand" concerns; keep the hedge modest because fundamentals can stay hot longer than sentiment. Falsify if backlog converts cleanly and supply obligations shrink as cash flow ramps.
• Stay underweight SFTBY versus the broader AI-capex basket for a 1-3 month window; it is more exposed to sentiment reversal and balance-sheet scrutiny if the market decides the cycle is too dependent on leverage. Cover if its AI allocation is funded with limited incremental balance-sheet stretch.
• Avoid chasing DBRG at/near takeout economics; the better trade is not the equity but the read-through to who survives the next funding tightening. If the deal price is the anchor, there is little asymmetric upside left.
• Set an alert on private-credit data-center loan spreads and covenant language rather than headline AI spend; a widening in that market is the cleanest early warning for second-order losers in power, cooling, and networking.