The $50 Billion Power Play: Switch Is Selling AI Real Estate, Not AI

A company that rents floorspace and electricity just filed for an IPO with a $50 billion valuation target. That is not a typo. It is not a hyperscale cloud provider. It is not a frontier AI lab. It is Switch, a data center operator based in Nevada. Back in 2022, a consortium led by DigitalBridge took the company private for about $11 billion including debt. Somewhere between then and now, the AI trade rewrote the price tag. Let me start with the obvious: This is not a technology IPO. This is a real estate trade dressed up in GPU clothing. Switch sells physical capacity — power, cooling, security, and floor space — to customers who want to run AI training or inference workloads. It does not build models. It does not design chips. It builds the building that holds the chips. The financial press will call it an AI infrastructure story. The term sheet will price it like a growth equity story. One of those is true. The other is the one you need to trade. This is the classic pick-and-shovel play. Switch operates facilities in Nevada, Michigan, Georgia, and Texas. That geographic split is not an accident. Those states have relatively cheap power, fast permitting, and enough land to build the kind of high-density data halls that AI training clusters demand. The company also has the story: AI training and inference need physical space. The hyperscalers need someone else to build that space. Switch will be the landlord. The capital structure tells you how serious this is. DigitalBridge and partners bought Switch for around $11 billion in 2022. Now the target is approximately $50 billion. That is a 4.5x jump in just three years. The usual suspects on the deal side are all present: Bank of America, Citigroup, Goldman Sachs, JPMorgan, and Morgan Stanley. Throw in a16z co-founder Ben Horowitz taking a board seat and leading the latest round. That is not just an IPO. That is a full-asset-class statement. Confidential filing in August. Public listing targeted for November. That window is deliberate. You skip the post-election noise, you land before year-end budget deployment, and you catch institutional investors who are desperate to find AI exposure beyond the obvious mega-cap names. It is a smart move. But smart moves and smart valuations are two different things. Let me run the numbers that the press release will not show you. Assume Switch, with a mature portfolio and ownership of its own construction pipeline, generates something in the range of $1.5 billion to $2 billion of EBITDA in 2025. That is a reasonable baseline for a data center operator on this scale. A $50 billion enterprise value against that EBITDA range gives you a multiple of 25 to 33 times. Now go look at the public comps. Equinix, the largest data center REIT in the world, has traded around 15 to 18 times EV/EBITDA. Digital Realty sits closer to 12 to 14 times. Both of them are high-quality, global, battle-tested platforms. Both are growing. Neither can touch Switch's AI narrative. That narrative premium is the entire trade. Can a data center operator justify 25 to 30 times EBITDA? Only if you believe AI capacity demand will turn those contracts into accelerating revenue for the next decade. That is possible. But it is not the same as a software product with 80% gross margin. This is steel, concrete, and power lines. It needs continuous capital expenditure. It is a heavy-asset business wearing a tech dress. Smart money doesn't pay 30x EBITDA for concrete. It pays for locked-in long-term contracts with escalation clauses and real volume commitments. If those contracts show up in the S-1, fine. If they do not, you are buying a story about land and power access, not a data center business. Let's talk about what the board seat actually means. Ben Horowitz joining Switch's board is not a mentorship play. a16z has deep positions across the AI stack — OpenAI, Anthropic, Mistral, and a dozen other model shops. When Horowitz sits in a data center board meeting, he is not just checking on cooling systems. He is connecting Switch to the AI procurement pipeline. The deal gets done because the capex budgets at those AI labs are enormous. The board seat is a relationship map and a customer funnel. It is an ecosystem hedge. The bank lineup is equally important, but read it with skepticism. Five bulge-bracket banks do not all sign a filing because the story is pretty. They sign because they believe they can sell it. That is the game. The banks are the distribution network. Their presence tells you the deal has been stress-tested enough to pass the basic underwriting bar. It does not tell you the valuation is cheap. It tells you the fees are real. There is a structural signal in the timing as well. You file confidentially in August, you shift the public offering into November, and you set yourself up for the post-Q3 window. By then, 2026 budgets are getting allocated. Fund managers who missed the AI rally in software will be looking for the next leg. A pure-play AI data center operator is an easy box to check. That is exactly why you have to be careful. The buying pressure can be enough to lift the first-day pop and leave the late buyer holding the negative carry. Here is the part the narrative will not tell you. The real bottleneck in AI infrastructure is not GPU inventory. It is not model quality. It is not even the cost of capital. It is the grid. Electrical switchgear has lead times of 18 to 24 months in many parts of the United States. Transformers are the new scarce asset. A data center with empty power capacity is effectively a brick building with no product to sell. Switch's actual moat, if it has one, is the ability to secure power capacity ahead of competitors. You cannot see that in the marketing deck. You have to see it in the utility interconnection queue. Then there is customer concentration. AI data center contracts tend to be massive — hundreds of megawatts signed with a very small number of hyperscale customers. If Switch's top tenant accounts for more than 20 or 30 percent of revenue, the market should demand a discount. That kind of concentration is not a diversified real estate portfolio. It is a lease agreement with a star tenant. The tenant has pricing power. The landlord gets headlines, not margin. Yield is the rent you pay for holding someone else's balance sheet, and this IPO is exactly that. Do not miss the environmental angle either. Data centers are the new natural gas champions. They consume enormous amounts of power and water. Local communities are starting to fight back. Nevada has renewable resources, but it also has drought cycles and grid constraints. The ESG narrative will be dressed up in green power language. The filing will give you the actual carbon and water numbers. Most people will not read that part. They are too busy watching the ticker. We don't buy narratives; we price the balance sheet. So when the S-1 drops, ignore the AI word. Look for three numbers: backlog in megawatts, net debt, and EBITDA margin. If the company can show a contract pipeline that towers over the current revenue base, the 25x multiple starts to make sense. If not, this is a classic private-equity exit dressed up as innovation. Here is my takeaway. The Switch IPO is a pricing experiment. It is the first serious test of how much public market capital will pay for the AI physical layer. If the deal clears at 25 to 30 times EBITDA, every private data center developer gets a new valuation mark. That will push secondary market prices up across the sector. If the deal stumbles, expect the whole AI infrastructure complex to trade down together. I will be watching three things between now and November: the a16z financing details, the S-1 backlog disclosure, and the 10-year Treasury. The first tells you whether the private round can anchor $50 billion. The second tells you whether there is real contracted demand. The third tells you whether the cost of leverage still allows the story to run. If the market gives you a clean IPO at a single-digit forward multiple, buy it. If it prices the AI dream at 30 times, let the dreamers hold it. The building is real. The question is whether the rent checks show up before the hype dies. I know which side of that trade I would rather be on.