Switch’s $80 Billion Confidential IPO: The Market Is Pricing Data Centers as AI Platforms, Not Real Estate

Switch has filed confidentially for a U.S. IPO with a reported $80 billion target valuation. That is not a pricing bet. That is a declaration that the old math of data center valuation is dead. The private filing means the market currently has zero visibility into Switch’s actual revenue, customer concentration, or backlog. No public S-1, no audited financials, no delivery schedules. Yet the $80 billion number is already being whispered around institutional desks as if it were a foregone conclusion. In my years watching capital pile into opaque infrastructure assets, this setup only ends one of two ways: early institutional alpha, or a slow-motion repricing that catches the last buyer flat. Switch is not a typical cloud company. It is a hyperscale data center operator with major campuses in Nevada, Michigan, and Texas, known for high-density racks, aggressive liquid-cooling designs, and a private interconnection network called Switch CIN. For most of its life, Switch sold space, power, cooling, and bandwidth to enterprise tenants. That was a real-estate business with an SLA contract attached. The AI explosion has forced a rebrand: the same physical assets are suddenly being marketed as AI-optimized computational infrastructure. The confidential filing lets management test that narrative without exposing quarterly churn rates or a potentially ugly backlog-to-capacity ratio. It also lets the company wait for peak sentiment before the roadshow begins. Why now? Because CoreWeave’s IPO created a valuation anchor for AI-native infrastructure. Because cloud giants are locking down multi-hundred-megawatt leases as if they are weapons. And because power utilities have become the true gatekeepers of the AI buildout. The sector is no longer capital constrained. It is power constrained. Power is the real collateral. U.S. data centers have roughly 20–25 gigawatts of total capacity in operation, with another 10 or more under construction. Wait times for new grid connections have expanded to three and sometimes seven years. Switch’s large land holdings and long-held utility relationships give it a seat in a game that is increasingly about grid interconnection queues, not cooling patents. Let’s stress-test the $80 billion figure. If Switch generated roughly $1.0–1.2 billion in revenue in the latest fiscal year—a reasonable inference based on its pre-2022 run rate and the AI-driven acceleration—the implied EV/revenue multiple is between 70x and 80x. Equinix trades near 8–9x. Digital Realty trades near 6–7x. CoreWeave, the pure-play AI cloud company, exited its IPO at roughly 22–30x forward revenue. So the market is being asked to price Switch as an order of magnitude more expensive than Equinix, and more than twice as rich as CoreWeave. The only way that math works is if Switch is no longer a data center company at all. It must be an AI asset platform with contracted capacity, locked-in power, and a customer list anchored by hyperscale cloud providers and frontier AI labs. Take a more conservative revenue estimate. If Switch’s trailing revenue is only $700 million, the implied EV/S ratio surpasses 110x, and the EV/EBITDA multiple is in the 180–200x range. That is not valuation. That is belief. It assumes revenue will compound at a 60–80 percent rate for three years, that more than 60 percent of revenue will be AI-related, and that long-term contracts with marquee clients will keep churn effectively at zero. The prior private-market valuation cross-rounds, which were presumably in the teen billions, would have to be reconciled as a sudden structural re-rating. Sometimes that happens. Usually it doesn’t. Here is the piece of analysis nobody seems to be quoting. At $80 billion, the implied enterprise value per operational megawatt is roughly $1.5 to $2 billion. That means Switch needs at least 400 to 530 megawatts of producing, leased, high-utilization capacity to justify the multiple. No public data confirms that capacity or that utilization. The last time I saw a comparable gap between narrative and disclosure, the names were asset-backed tokens trading on a promise. They did not age well. Based on my audit experience, this is the point where you stop listening to the management narrative and start demanding the S-1. The key metrics to watch in that prospectus are not vague claims about high-density racks. They are, first, the ratio of AI-related revenue to total revenue; second, the dollar value of signed power purchase agreements and their duration; third, the top-five customer concentration; and fourth, the backlog-to-total-capacity ratio. The single metric I will scrutinize is the signed backlog divided by total available capacity. If that backlog is below 70 percent, the $80 billion narrative collapses. If it is above 85 percent, the bull case is real. No other number matters as much. Now for the contrarian read. The real moat is not liquid cooling, not rack density, not the proprietary network. It is the entitlement to draw power from an increasingly constrained grid. In the American West, utility interconnection queues now stretch three to seven years. A data center with firm power capacity in hand is worth more than one with superior cooling. The market is handing Switch an AI premium, but its durable advantage is likely land, water, and long-dated power contracts. Those are assets of a utility, not a software platform. An energy asset trades around 10x EBITDA. An AI platform trades at 100x. The reported $80 billion valuation is a deliberate attempt to keep the AI label on while hiding the electricity giant underneath. Strategic pivots aren’t optional; they are forced by the capital markets. Liquidity doesn’t respect narratives. It follows cash flow and contract signatures. If Switch’s IPO succeeds, it will open the public-market doors for more private developers, pushing capital into a sector learning the difference between announced capacity and energized capacity. The true test comes in 2026 and 2027, when a wave of newly built capacity hits the grid simultaneously. If AI workload growth slows even slightly, the pricing power that supports $80 billion valuations evaporates. The bear case is not that Switch is a bad operator. It is that the whole sector is trading at perfect execution prices, and Switch is the newest, least-transparent vehicle for that trade. There is also a structural asymmetry worth naming. If Switch simply leases white space and power, its long-term economics resemble traditional data center trusts with cyclical lease rollovers. If Switch is actually building a GPU-ready platform with multi-year commitments from a frontier AI lab or a hyperscaler, the current weakness in disclosure becomes an opportunity—not because the valuation is sane, but because the scarcity of power contracts is real. I have audited enough power-heavy deals to know that the difference between a $30 million and a $300 million asset is often a single page in a power purchase agreement, and that page is usually buried in a legal appendix. What happens next is predictable. If the S-1 contains overwhelming AI revenue and long-dated power contracts, the public market will drive Switch’s valuation even higher. If the document reveals a traditional colocation company with a few high-density showpieces, the discounting will be brutal. The best trade here is not a blind buy. It is a patient wait for the actual filing, then a forensic read of the backlog, the power costs, and the customer names. You don’t need to be early in this cycle. You need to be right about the numbers. The numbers are still hidden.