$80 Billion for a Landlord? Switch's Confidential IPO and the New Math of AI Infrastructure

Every crash is just a story that hasn't finished being told. I keep that line close because it has saved me more capital than any indicator ever has. In the DeFi winter, we didn't know we were in one. We thought we were early. We thought yield was a reward for smartness. It turned out to be a transfer from slower hands. That lesson is why I am watching Switch's confidential IPO filing with something between curiosity and dread. A Nevada-based data center operator most crypto traders have never heard of wants to be valued at $80 billion. Not eight. Eighty. That is not a typo. Switch submitted its S-1 confidentially in the spring of 2025. The company is betting on one narrative that is already reshaping capital markets: AI has turned the boring business of leasing racks and selling kilowatts into the most exciting asset class on the planet. The building is not real estate anymore. It is a compute platform. This article is not a take on whether AI will save us or ruin us. I am a trader, not a philosopher. But I did spend a decade reading whitepapers that promised one thing and delivered another. And I am seeing a pattern. Switch is the kind of company that existed in the noise of the data center industry for twenty years. Founded by Rob Roy in his father's garage in Las Vegas, it grew into an operator of massive, vault-like campuses in Nevada, Michigan, and Texas. The architecture is distinctive, almost parochial: self-contained campuses designed like concrete fortresses, with cooling systems that treat water as precious as compute. Before AI, Switch was known for one thing: building ahead of demand. It hoarded land, secured power reservations, and constructed capacity that looked irrational when it was built. In 2017 that looked like ego. In 2024 it looked like a moat. The company reportedly generated around $680 million in revenue in 2022. Industry estimates put 2024 revenue somewhere in the $1.0 to $1.2 billion range. That is a strong growth story by legacy data center standards. But it is not an $80 billion story. To understand how a company with roughly one billion in revenue can target an $80 billion valuation, you have to understand the new math of AI infrastructure. The market no longer values data centers as real estate. It values them as the entry ticket to a GPU supply chain that has become more constrained than the semiconductor fabs themselves. Rack density is the key metric. A traditional enterprise rack draws 5 to 10 kilowatts. An AI rack draws 30 to 100 kilowatts. A single power-dense building can now host hundreds of racks, each running thousands of dollars of GPUs that never stop computing. Revenue per square foot dwarfs what Equinix collects from normal enterprise tenants. The problem is that the cost of power and cooling scales just as fast. The market wants to pay for future power contracts, not existing ones. And here is where my skepticism kicks in. Let's run the math. With an $80 billion target valuation, and assuming $1.2 billion of 2024 revenue, Switch would trade at roughly 65 to 70 times EV to revenue in its best case. If revenue landed at $1.0 billion, the multiple is closer to 80 times. Compare CoreWeave. CoreWeave generated about $1.6 billion in 2024 revenue and went public with a market value in the $35 to $50 billion range, roughly 22 to 30 times revenue. Equinix, the global colocation king, trades around 8 to 10 times revenue. Digital Realty sits at 7 to 9. Switch is effectively asking the public market to pay 65 to 80 times revenue for a company that has not yet proven it can operate GPU clouds, does not have a disclosed customer concentration roadmap, and has not revealed its backlog. Even on the traditional data center metric of EV to EBITDA, the implied number is severe. A well-run leased data center has EBITDA margins of 45 to 55 percent. At $1.2 billion revenue, that implies $540 to $660 million of EBITDA. At an $80 billion enterprise value, the EV/EBITDA multiple lands somewhere between 120 and 150 times. Equinix trades at 18 to 22 times. CoreWeave is loss-making but carries a growth premium. Switch would be priced as if it already has CoreWeave's growth trajectory and Equinix's asset base at the same time. That is not a valuation. That is a mandate. Mandates have a way of collapsing when the numbers slip out. There is hope, and it has a name: power capacity. In AI infrastructure, the scarcest resource today is not chips. It is grid interconnection. The queue to get transmission capacity in the United States stretches three to seven years. A data center with already secured power capacity and approved permits is worth more than a building with double the space and no electricity. Switch built its business on land banking and power reservations. If it can prove it holds 300 to 500 megawatts of shovels-ready power capacity with contracted renewable and natural gas supply, the $80 billion target moves from fiction toward a high-risk but tradeable story. But the market has not seen that proof yet. That is what confidential filings hide. And that is precisely where the smart money is waiting. Let's talk about the revenue problem. Switch's traditional business is wholesale colocation. Enterprises sign 3 to 10 year leases for racks, floor space, power, and cooling. The cash flows are predictable. Growth rates are modest. That business deserves a real estate multiple, not an AI multiple. To earn $80 billion, Switch must convince the market that the majority of its revenue is now AI-driven. That means high-value deals with large AI labs, GPU cloud providers, or hyperscalers. I have not seen any evidence that Switch has signed a CoreWeave-style contract. Wait, CoreWeave needs data centers too. There is a scenario where Switch operates as the physical layer for GPU clouds. In that case, Switch is the landlord and CoreWeave is the casino. It is a quality business, but it captures a small fraction of the value. The alternative scenario is that Switch becomes a GPU cloud itself. That requires purchasing tens of thousands of GPUs, building the orchestration layer, managing a pool of depreciating hardware, and doing it while maintaining a 45 to 55 percent EBITDA margin on the real estate side. I have seen exactly how hard this is from my own experience running a copy trading community. Technology is never the bottleneck. Operations and capital discipline are. And those are exactly the skills that have not been tested at scale. My audit rule has always been: if you cannot see the balance sheet, assume the worst. Right now, with a confidential S-1, no one can see anything. The only facts are the target, the sector, and the insiders' decision to file secret. Let's place Switch among its competitors, because the market is pricing it inside a specific bracket. There are two families of data center companies today. The first family is the real estate survivors: Equinix, Digital Realty, and their global peers. They have enormous revenue, mature portfolios, and occupancy rates that prove demand. Their multiples are low. The second family is the AI natives: CoreWeave, Nebius, and a handful of private GPU cloud operators. They have small revenue, massive growth, and investor narratives built around scarcity. Their multiples are high. Switch is trying to occupy both families at once. It wants the maturity of the first family and the multiple of the second. That is the core contradiction. I have seen exactly this kind of straddle before in crypto: projects that claim to be protocols while collecting rent like a SaaS company. The market figures it out within a few quarters. There is also a hidden interplay with AI chipmakers. NVIDIA is not just a chip company; it is a financing vehicle. It takes equity in GPU clouds that promise to buy its hardware. If NVIDIA or another strategic investor takes a stake in Switch before the IPO, that changes the calculus. The $80 billion target becomes partly a strategic valuation, not a market one. I would watch the insider list in the S-1 for a familiar color of logo. There is a pure crypto angle here that most crypto analysts miss. AI compute is becoming the new collateral. Already we see protocols borrowing against GPU futures, tokens backed by data center capacity, and DePIN networks trying to tokenize idle compute. An $80 billion data center IPO legitimizes the idea that physical compute capacity can be priced as a financial asset. That narrative spills directly into the market for AI-related crypto assets. But be careful. The spillover is a double-edged sword. When the S-1 lands, capital that would have chased speculative GPU cloud tokens may instead chase a regulated equity. The same money that pumps decentralized compute narratives can rotate into an exchange-listed stock the moment it looks safer. That is not a bullish sign for speculative AI-crypto. It is a maturity signal. Now, the darker layer. AI data centers are becoming a national energy policy issue. US data center power consumption is projected to rise from roughly 150 terawatt-hours in 2024 to 250 to 350 terawatt-hours by 2028. That is more than many entire countries consume. The result is a bidding war for power. Utilities are rewriting rate structures. Some states are rejecting new data centers because of grid strain. This is a problem and an opportunity. Operators with firm power contracts can charge premiums. Operators without them are stuck with unleased shells. Switch has a history of touting its green credentials and renewable-heavy power mixes. But AI workloads are intermittent and power-hungry in a way that renewables alone cannot match. The credible operators will pair renewables with natural gas peaker plants or pursue small modular nuclear reactors. Every one of those choices adds capex and regulatory risk. ESG investors will want a carbon plan that is fully funded. National security investors will want a supply chain that can survive an export ban on transformers or a geopolitical crisis. The S-1 will need to walk that line. Most S-1s do it with glossy words and hidden footnotes. I have audited enough protocols to know that footnotes are where the truth lives. The contrarian angle that nobody in the AI hype crowd wants to hear is that the supply side is coming. It is not optional. It is already under construction. Industry estimates put the US data center connected load in the 20 to 25 gigawatt range, with more than 10 gigawatts of AI-driven construction underway. Through 2026 and 2027, that capacity will hit the market. When it does, the premium for urgency collapses. Lease rates will normalize. The power-density premium will compress. The exact moment an $80 billion valuation is fully funded by revenue is the exact moment the next cycle starts to discount it. There is also electricity contract inflation. New power agreements signed today are priced higher than those signed five years ago. Operators buy power at market prices and sell it through long-term fixed leases. That makes them short volatility. If power prices spike, margins get squeezed. If they fall, the leasing market gets competitive. Either way, the balance sheet absorbs risk. And there is the government angle. If Switch hosts classified AI workloads for US defense agencies, it gains a premium revenue stream and a concentrated, politically sensitive customer base. It cuts both ways. It can be a moat, and it can write a headline that kills the IPO at the exact wrong moment. I didn't buy Terra's story because the bond mechanism was the tell. Algorithmic stabilizers always break at the exact moment you need them most. I am not saying Switch is Terra. I am saying the mechanism by which an $80 billion valuation gets confirmed, long-term contracts with a handful of AI customers, locked-in power prices, and a believable construction pipeline, is a mechanism that can break just as fast as an algorithmic peg. In crypto, I learned to watch who is selling the shovels. In 2017, the ICO issuance machines sold tokens by the crate. The people who made money were the infrastructure providers, not the token buyers. In 2020, the liquidity providers subsidized yield farms that promised 1000% APY. The winners were the protocol founders and early insiders. The same pattern is repeating in AI infrastructure. The real financial winners of the buildout may not be the model companies at all. They may be the power equipment manufacturers, the cable makers, and the data center landlords, but only if the landlords are not forced to over-invest in speculative, unleased capacity. And this is my battle-tested rule: infrastructure without committed revenue is just a very expensive story. The sharpest operators will avoid the IPO pop and buy either the suppliers like Vertiv, Eaton, or Cummins, or the developers that can demonstrate backlog before they fill it. The retail narrative will funnel money into equity at $80 billion valuations based on AI optionality. The institutional capital will wait for the S-1 and ask one question: what is the renewal rate and what is the building cost per megawatt? Let's game out three scenarios. Optimistic: revenue reaches $1.5 billion and the market accepts a 53 times revenue multiple, still above CoreWeave, but justified if AI revenue grows at triple digits. Base: revenue is $1.1 billion, the multiple is 70 to 80 times revenue, and the market will demand a very long runway. Pessimistic: revenue is $700 million and the multiple is 100 to 114 times revenue. That is not a valuation; that is a religious experience. Let's talk about the numbers the S-1 must reveal. First, revenue mix. If AI-related revenue is more than 50 percent of contracted bookings, the valuation has a base. If it is less than 20 percent, this is a colocation company trying to wear a GPU cloud costume. Second, backlog. Accounting rules require disclosure of remaining performance obligations. For data center operators, this is rental committed but not yet invoiced. If backlog is more than $3 billion with weighted lease terms of five years, the story starts to hold. If it is a few hundred million, the $80 billion target is a mirage. Third, customer concentration. AI workloads come from a very small club: Microsoft, OpenAI, Meta, xAI, Oracle, Anthropic, plus the GPU cloud brokers. If Switch's top five customers account for more than 70 percent of revenue, churn is existential. Fourth, power procurement. Does Switch control enough electrical capacity to double its portfolio over three years? I want to see long-term power purchase agreements that lock in 70 to 80 percent of projected energy costs. Anything less is margin risk. Fifth, balance sheet. The construction pipeline needs $50 to $75 billion in capital over three years if the company wants to add 500 megawatts of new capacity. How much is debt, how much is equity, and who is buying the equity? This is the checklist I use for every protocol I audit. And before the S-1 drops, nobody can tick these boxes. The confidential filing also tells us something about timing and confidence. Switch could have filed a public S-1 immediately. It chose the confidential route. That means management wants to control the narrative, clean up the numbers, and pick a moment when AI sentiment is maximally supportive. It also means they have left themselves a path to withdraw. The target of $80 billion is likely pre-marketed. There is probably a pilot fish in the water, an anchor investor, a strategic partner, or a sovereign fund that has signed a preliminary subscription. In crypto, this looks like a pre-IPO round at a premium valuation with a lock-up and a promise of listing. It is a coordination mechanism, not a price discovery mechanism. If the S-1 reveals disappointing financials, the bankers will quietly lower the range. If it reveals a backlog that supports the target, the IPO will be oversubscribed. The market will believe whatever the numbers tell it to believe. I also want to flag listing venue choice. NYSE or Nasdaq is not just a technical detail. A Nasdaq listing will signal tech growth. A NYSE listing will signal industrial infrastructure. The choice tells me which investor base Switch is trying to attract. I have sat through two complete boom-bust cycles in crypto, and I have watched a third form in AI. The pattern is always the same. First there is a real bottleneck. Then money floods in to solve the bottleneck. Then capital overshoots and builds far more supply than demand can absorb. Then the price plunges. The long-term question is not whether Switch can build data centers. It has proven that. The question is whether data centers will remain a luxury real estate product or become a utility-like service with commodity rents. If AI compute demand is real and permanent, the current scarcity prices can justify heroic valuations for a while. If another AI winter arrives, all the power contracts in the world will not defend an $80 billion multiple. The headline is $80 billion. The real story is a test of whether capital markets can price physical scarcity with software-era multiples. If Switch prints after the S-1 reveals even a half-believable AI revenue mix, the entire data center sector will be repriced. If the offering lands flat or below the target range, the AI infrastructure trade will face a serious correction. Either way, my advice is the same as it was in the DeFi winter. Price is just a story. The balance sheet is the cash register. Watch the filing date. Watch the pre-IPO strategic investors. And above all, watch the megawatt count. Because in the next cycle, the collateral is not a token. It is a substation. t saying.