We didn't need to read Switch's S-1 to sense which way the AI infrastructure winds were blowing. The numbers had already told us. In 2022, a DigitalBridge-led consortium acquired the Las Vegas-based data center operator for roughly $11 billion, debt included. By 2025, Switch is targeting a $50 billion valuation through a US IPO, having filed confidentially in August with a November listing window and an underwriting syndicate featuring Bank of America, Citi, Goldman Sachs, JPMorgan, and Morgan Stanley. On top of that, Ben Horowitz โ a16z co-founder and one of the loudest voices in AI investing โ is taking a board seat and leading a fresh financing round.
What makes the announcement remarkable is what it does not contain. No GPU specifications. No power usage effectiveness figures. No rack density numbers. No client names. No megawatt backlog. The AI story has become so potent that a company can climb from $11 billion to $50 billion in private-market ambition without disclosing a single technical parameter. That is not a critique; it is a signal. Public markets are being asked to price a 4.5x valuation leap on the back of a narrative, and the absence of technical detail tells us exactly where the value is supposed to live: in land, in grid access, and in the permission to keep the models training.
Switch is a pick-and-shovel operation at the bottom of the AI stack. It operates multi-tenant data centers in Nevada, Michigan, Georgia, and Texas, and its business is the least glamorous part of the digital economy โ wholesale power, floor space, cooling, and connectivity sold under contracts that typically run five to fifteen years. It does not train models. It does not design silicon. It has no benchmark scores. What it has is a balance sheet that can promise a hyperscaler a hundred megawatts in a specific county, with a specific grid interconnection timeline, at a specific electricity price. In the AI gold rush, Switch sells picks and shovels, and it charges by the ton.
The details above come from a Bloomberg report relayed through a blockchain media outlet, citing people familiar with the matter. The valuation journey deserves attention regardless of the relay chain. In 2022, DigitalBridge took Switch private at $11 billion. The 2025 target is roughly $50 billion. Even accounting for new debt raised and assets built during those three years, the jump is steep. A 4.5x appreciation in a sector whose underlying asset โ concrete, power, networking gear โ still depreciates like any physical plant is, at minimum, a philosophical statement. Consider the public-market comparables: Equinix, the global colocation leader with facilities across five continents, trades around $80-100 billion. Digital Realty sits in a similar band. A regional US operator targeting $50 billion would therefore be valued in the same atmosphere as the two industry titans, with a far smaller physical footprint. That is either visionary or narrative inflation, and the next two months will tell which.
The broader wave matters here, too. Since early 2025, a stream of data center operators, contractors, and equipment suppliers have gone public in the United States. The infrastructure industry has crossed from a private-equity habitat into public-market pricing. Switch's IPO would be the largest and most symbolically charged of that wave, completing a full financial cycle: private equity buys, builds, and flips the asset to public investors who are being asked to believe the AI buildout lasts another decade. The capital that exits that flip is not pure profit. It is a bet that the AI narrative outlives the interest-rate cycle.
Now the analysis that the announcement leaves to our imagination, starting with the hardest number: valuation.
We didn't expect public markets to price a $50 billion AI story without a single disclosed megawatt in the backlog. Yet here we are. If Switch hits its target, and if it performs at the level of a top-tier US operator โ EBITDA between $1.5 billion and $2 billion โ the implied forward multiple lands between 25x and 33x. Traditional data center REITs like Equinix and Digital Realty trade between 12x and 20x. To justify roughly double the multiple of the industry's two established leaders, a company must prove one of two things: either its contract backlog is structurally better โ larger clients, longer terms, faster delivery schedules โ or the market has decided that "AI data center" is a different asset class from "data center." Both possibilities are live. Both have very different consequences for every other AI infrastructure holding.
The underwriting syndicate is the first piece of evidence pointing toward a different pricing system. Five banks of this caliber do not attach their names to an IPO without running internal models that survive weeks of scrutiny. In the 2024-2025 IPO window, this combination is reserved for companies whose financial documentation holds up. For outside observers, that means the narrative โ even without disclosed numbers โ passed a threshold of internal credibility with the very institutions that will profit from the deal. We do not get to see their models. But we can infer that the sell-side is comfortable enough to risk its own book on this valuation story.
Then there is Ben Horowitz. His board seat, combined with an a16z-led financing round, changes the deal's chemistry. a16z holds major positions in OpenAI, Anthropic, Mistral, and the broader frontier-lab ecosystem. Horowitz joining Switch's board is a channel from Switch to the companies that actually need hundreds of megawatts for training. In private-market language, this is "value-add." In practical terms, it means Switch is not just buying capital โ it is buying a customer-acquisition relationship with the most important counterparties in its sector. The move also anchors the IPO psychologically: a leading AI-insider figure has signed off, and public investors are invited to follow his pricing lead.
The timing is no accident either. A confidential filing in August, a listing window in November, after the presidential election and past the earnings-season crush, in time for institutions to make year-end allocation decisions. The banks are not picking dates; they are optimizing for a period when capital budgets are still open and the news cycle has room for exactly one big infrastructure narrative. They are also waiting for the Fed's November meeting; a hawkish surprise could make a leveraged infrastructure debut far less appetizing.
Now the uncomfortable parts, because the story's silence is itself information. The S-1, expected in late October, will answer the questions the public narrative avoids. First: net debt. The 2022 buyout was leveraged, and data center construction devours capital. If the enterprise value carries that $50 billion story but net debt is substantial โ and it will be โ then the equity value at the end of the chain is smaller than the headline suggests. Second: customer concentration. In AI colocation, the growth pattern is a small number of huge deals. One tenant at 25% of revenue is a risk that public-market screens will flag immediately. Third: megawatt backlog. The most important number in the entire document will be contracted-but-undelivered power capacity, mapped against construction timelines. Fourth: rack density. Legacy data centers were designed for racks in the 5-20kW range; AI clusters routinely demand 100kW per rack with liquid cooling. If Switch's existing portfolio was built for the old world, upgrading it will absorb enormous capital. If it was pre-built for density, it is a genuinely differentiated asset. Notice also what the coverage does not say: nothing about how a future rate cut or delay affects a leverage-heavy balance sheet, nothing about old equipment write-downs, nothing about the pricing power of hyperscale tenants. The absence of counterarguments is a feature of the genre, not an oversight.
Geography is the quiet part. Nevada, Michigan, Georgia, and Texas are not random choices. They combine low electricity prices, relatively available grid capacity, low natural-disaster risk, and cheaper land. In this industry, competitive advantage is land plus power plus speed-to-grid. The moat is procurement, not brand. Nevada's renewable mix gives Switch a sustainability narrative that ESG funds can touch, though hydropower variability in drought years is real. Georgia and Texas sit at the center of the American data center boom, where grid interconnection requests now stretch โ measured in years, not months. The bottleneck is no longer capital; it is transformers and switchgear, whose US lead times have expanded from a few months to more than two years. That equipment backlog is the true timeline of the AI buildout, and it affects every expense assumption in Switch's growth model.
There is also a quiet danger in how this story is being told. The industry is beginning to act as if "AI data center" is a different asset class from "data center," simply because of who the tenant is. The same steel, the same power, the same cooling โ but because the customer is an AI lab rather than an insurance company, the market is expected to pay a higher multiple. That distinction is not engineering. It is narrative, and narrative does not appear on balance sheets. This is exactly why the crypto world should pay attention. Our own RWA and DePIN ecosystems depend on the same trick: mapping physical infrastructure into new financial vehicles. If public markets begin systematically pricing "AI-ness" into static real estate, the door opens for on-chain representations of the same assets โ for better and for worse. Several tokenized real-world asset platforms are already structuring data-center debt products, and they desperately need a public pricing benchmark. Switch may well become that benchmark, whether or not it ever touches a blockchain.
Here is the part that keeps me up at night, because it tests what I do for a living.
We didn't build decentralized compute pilots to watch them lose to REITs. In 2024, my team and I ran a project in the Philippines using Golem's decentralized compute network for local news verification โ 10,000 data points, a 40% reduction in misinformation in our test environment. The architecture worked. The incentives worked. But when we needed production scale, we rented centralized capacity, because it was available, simple, and predictably priced. Decentralized infrastructure did not lose on principle. It lost on convenience. Switch's IPO is that lesson in macro form: centralized capital tends to win the first phase of the AI buildout because it knows how to package scarcity into financial products.
Still, let me steelman the bull case, because dismissing it would be dishonest. What if the AI premium is not about AI at all? What if the market is pricing the right to draw cheap, reliable power from specific grids? Then 25-30x is defensible, because fast-interconnection power in Nevada, Texas, Georgia, and Michigan is genuinely scarce. The AI label becomes simply the language a modern real-estate business uses to translate scarcity into a liquid security. In that world, Switch is a utility with AI-seasoned earnings, and the multiple holds as long as power remains constrained.
The bear case is equally simple. We have watched this movie before. In 2021, the metaverse label doubled multiples for companies doing mundane work. When the narrative cooled, valuations reverted to sector fundamentals. If AI capital expenditure growth slows even slightly โ if one hyperscaler trims next-quarter guidance โ a 25x+ multiple on a concentrated customer base could compress 30% or more. The S-1 is the first place where the narrative meets a spreadsheet.
We will learn most of what matters when Switch's S-1 lands and the a16z round's valuation is disclosed. Watch net debt, customer concentration, signed-but-undelivered megawatts, and the density of the existing portfolio. In this kind of market chop, these are the signals that separate positioned investors from narrative chasers: the private round's EBITDA multiple will move before the public market reacts. If the round prices at $50 billion, the IPO is effectively written. If the numbers disappoint, the AI infrastructure premium faces its first genuine corrective test โ and every inflated "AI" asset will feel it. The landlord economy is going public. The question is not whether it happens, but whether we are reading the fine print.