Switch’s $50 Billion I.P.O. Is a Power Purchase, Not an AI Story

Hook: The Silent Filing Switch has filed confidentially with the SEC. Target valuation: near $50 billion. Underwriters: Bank of America, Citigroup, Goldman Sachs, JPMorgan, Morgan Stanley. Board addition: Ben Horowitz of a16z. Intended listing timing: November. Revenue: undisclosed. EBITDA: undisclosed. Net debt: undisclosed. Backlog in megawatts: undisclosed. Customer concentration: undisclosed. S-1 public version: not yet available. That is not a technology company approaching the public markets. That is a power-and-land arbitrage vehicle asking public investors to price a belief before the transaction log is visible. The bytecode lies; the transaction log does not. In private markets, narratives can survive without audit. In public markets, the S-1 is the transaction log, and Switch has not shown it yet. The hard facts are real enough. DigitalBridge and its consortium took Switch private in 2022 for roughly $11 billion including debt. Now the same asset base, expanded by growth capital, is targeting a valuation nearly five times that level. The jump happened in three years. The entire gap between $11 billion and $50 billion is not explained by square footage. It is explained by a single word attached to the asset: AI. Context: Compute Real Estate, Not Compute Switch operates multi-tenant data center colocation and custom data center services. It owns facilities in Nevada, Michigan, Georgia, and Texas. Those locations are selected for low power prices, relatively low natural disaster risk, and available land. The company does not train models. It does not optimize inference. It does not write kernels or design chips. It builds and leases physical environments where GPUs can be placed, powered, cooled, and connected. That makes Switch a member of the AI supply chain, but not an AI technology company. It is the pick-and-shovel seller at the edge of the gold rush. The real product is not compute. The real product is electricity capacity converted into lease payments. The business model is simple: secure long-term power contracts, build data center shells, install cooling and network infrastructure, then wholesale the space to hyperscale cloud providers and AI labs on contracts that can run five to fifteen years. Revenue visibility is structurally better than most software companies in the AI narrative trade. But the model is capital-intensive, balance-sheet-driven, and brutally sensitive to interest rates. The term “AI data center” is doing heavy lifting in this valuation. A rack of servers in Nevada does not know whether its tenant is training a large language model or running enterprise databases. The same physical asset can be marketed as AI infrastructure or traditional colocation depending on the customer list. The difference in valuation multiples between those two labels is enormous. Switch is asking the market to accept the first label while withholding the customer list. Core: The Evidence Chain Let me walk through the evidence chain that is publicly available. It is thin, but it is not empty. First, the acquisition math. The 2022 take-private was valued at about $11 billion including debt. The 2025 target is approximately $50 billion. That implies a roughly 4.5x increase in total enterprise value, even before accounting for any new debt raised to fund construction. Three years is a short window for that kind of re-rating. Traditional data center operators did not re-rate anywhere near that level over the same period. Equinix’s public market value grew, but not at this pace. Switch’s multiple expansion is therefore not a sector-wide phenomenon. It is a specific bet that AI-dedicated infrastructure deserves a different valuation system. Second, the underwriter lineup. Five top-tier banks underwriting together is not routine. It signals that Switch’s financials, contract backlog, and EBITDA projections have passed internal diligence. Banks do not put their names on a $50 billion deal without some confidence in the revenue model. But that confidence is not public evidence. It is a seal of approval from parties who will collect fees when the deal closes. The underwriter lineup is a signal of execution capability, not a proof of valuation adequacy. Third, the a16z move. Ben Horowitz joining the board and leading a new funding round is a strategic asset, not merely a capital injection. a16z has positions in some of the most prominent AI labs. That network can open doors to long-term compute contracts with model developers. It also gives Switch an AI ecosystem endorsement that Equinix and Digital Realty cannot easily replicate by buying more land. But the same fact creates a governance question. A venture capital fund with board representation and aligned incentives is not a neutral auditor. The endorsement is valuable because it is interested. Fourth, the timing. The confidential submission was reportedly made in August. The target is November. That window is after the U.S. election and before year-end rebalancing flows. It also positions the deal for institutional investors who will be allocating fresh capital at the start of the new year. The timing is professional. It suggests the company and its bankers believe the current market window can support the valuation, and that a delayed public filing would risk losing the momentum. Pressure tests expose what calm markets hide. The calm market is now. Now here is the part that bothers me. None of the metrics that actually determine a data center company’s value have been disclosed. What is the contracted but undelivered power capacity in megawatts? What is the annual delivery schedule for that backlog? What is the average construction cycle from groundbreaking to turnkey delivery? What percentage of revenue comes from the top three customers? What is the net debt position after the 2022 leveraged buyout and subsequent capital expenditures? What is the EBITDA margin on existing facilities? None of those numbers appear in the public evidence chain. Without those numbers, the $50 billion valuation is not an analysis. It is an assertion. Based on my experience stress-testing DeFi lending protocols in 2020, I learned that collateral quality matters more than collateral volume. A liquidation mechanism can look robust until the moment a single large position moves. The same principle applies here. Contracted megawatts are not the same as delivered megawatts. A signed lease with an AI lab is not the same as a paid invoice. The transaction log, not the tweet, will tell us which part of the vision is real. Let us test the valuation range with industry benchmarks. Traditional data center REITs such as Equinix and Digital Realty have traded in roughly a 12x to 20x enterprise value to EBITDA range in recent cycles. If Switch reaches a $50 billion enterprise value, and if it is generating something like $1.5 billion to $2 billion in EBITDA in 2025, the implied multiple is approximately 25x to 33x. That is a meaningful premium to the traditional infrastructure cohort. A premium can be justified if growth is structurally higher, but the burden of proof is on the company to show that its existing lease portfolio and contracted pipeline support that growth. In the absence of S-1 data, the 25x to 33x multiple is a hope wearing a spreadsheet. There is also the debt layer. The 2022 deal at $11 billion was leveraged. Data center construction is cash-hungry. Switch has likely taken on additional debt to fund expansion since then. Enterprise value includes debt, and equity value is what remains after subtracting it. A $50 billion valuation narrative can conceal a net debt position large enough to make the equity story far less attractive than the headline number. The article reporting the IPO target did not disclose net debt. That silence is not neutral. Silence in the logs speaks louder than tweets. Contrarian: AI Premium Is a Liability Here is the counterintuitive angle. The AI label that justifies the $50 billion valuation is also the structural weakness that could cause it to collapse. If Switch is priced as an AI infrastructure pure play, then its fair value will depend on the capital expenditure plans of a handful of large technology companies. Those spending plans are cyclical, concentrated, and outside Switch’s control. A single decision by a major cloud provider to slow GPU cluster expansion can reduce demand for new data center space faster than a traditional enterprise colocation cycle would. Correlation is not causation. The fact that AI training demand has grown is used as evidence that data center capacity will keep growing. But data center demand is ultimately a function of financed construction, not merely of model training appetite. If capital costs rise, or if the largest customers renegotiate terms, the backlog becomes a negotiating liability rather than a guaranteed revenue stream. In crypto, we call this a squeeze. In commercial real estate, it is called lease rollover risk. Customer concentration is the hidden fault line. AI data center orders tend to come from a small number of large counterparties. If one customer accounts for more than 20% or 30% of revenue, the public market usually assigns a discount, not a premium. The entire $50 billion narrative may depend on contracts that have not been disclosed. High-value deals signed in a private market moment may not survive public scrutiny. Volatility is noise; structural flaws are signal. The structural flaw here is the mismatch between a long-duration physical asset and a fast-moving demand narrative. The second contrarian point is that the true moat for data centers is not the AI narrative. It is the ability to obtain electric power. Transformers and switchgear now have lead times stretching many months or even over two years. Grid interconnection queues are congested. Local opposition to high-power facilities is growing. In that environment, control of power capacity is the real competitive advantage. Switch’s geographic footprint in states with relatively favorable power markets is a genuine asset. But that asset is not an AI asset. It is an energy asset. AI demand is the marketing wrapper that determines whether the market values it as a utility or as a growth technology. That leads me to a practical caution. If the a16z-led private round closes before the IPO at a valuation near $50 billion, it will create an anchor for the public offering. That is not necessarily a positive signal. It means the private round is being used to legitimize a price that the public market has not yet tested. I have seen this pattern in crypto projects that raise large private rounds before listing. The private round creates a price floor, but it also creates a psychological ceiling for early public investors. The disciplined move is to wait for the S-1, not the press release. Takeaway: The Next Signal Data does not dream; it only records. The next serious data record will be the public S-1 filing, expected in the weeks before the November launch. That document will reveal revenue, EBITDA, net debt, customer concentration, contracted megawatts, and the actual pace of construction. Every serious investor should read those pages before accepting any valuation above 20x EBITDA. If the S-1 shows a meaningful backlog, a manageable debt load, and a diversified customer base, the AI premium may be justified. If it shows a small group of customers and a high net debt position, the 500 percent increase from the 2022 buyout will look like a transfer of risk from private owners to public shareholders. The safest trade is not to predict the IPO price. It is to measure the gap between the narrative and the transaction log. Trust the hash, verify the execution path. Switch has not yet given the market enough hash to verify.