Switch, the Las Vegas-based data center operator, filed a confidential S-1 with the U.S. Securities and Exchange Commission in the second quarter of 2025. The target valuation is $80 billion. The last audited public revenue figure for Switch was approximately $680 million, reported for fiscal year 2022. If the company grew at a heroic 50% compound annual rate for the next two years, trailing twelve-month revenue would reach about $1.5 billion. That still yields an implied EV/Revenue multiple above 50x. Equinix generated roughly $8.7 billion in 2024 revenue and carries an enterprise value near $80 to $90 billion. Its multiple is approximately 10x. Switch asks the public market to pay five to eight times that multiple on a revenue base that is one-seventh the size. Audit gap confirmed.
Confidential submission is not a neutral procedural choice. It is a timing instrument. Management can prepare the financial narrative, test institutional demand through pre-marketing conversations, and withdraw without the reputational scar of a failed public offering. Switch wants maximum flexibility in a market that has priced AI infrastructure as a new asset class. The anchor for this narrative is CoreWeave. The GPU cloud provider went public in 2025 with a valuation of $35 to $50 billion against roughly $1.6 billion in 2024 revenue. That generated a 22 to 30x revenue multiple, far above the traditional data center sector. Nebius, another European AI cloud player, trades near 20 to 30x revenue. Switch wants to be evaluated in this cohort, not alongside Equinix and Digital Realty.
Switch was founded in the early 2000s and developed large campuses in Nevada and Michigan, including sites in Las Vegas, Reno, and Grand Rapids. It marketed proprietary cooling designs and concentrated power models before the AI era. It also operates its own carrier-neutral network, Switch CIN, connecting its facilities. These assets are real. They pre-date the AI narrative. What the source material does not disclose is how much of this capacity is AI-grade versus legacy enterprise colocation. That is the information gap at the center of the valuation.
Confidential submission also implies that Switch, or its advisers, believe the AI infrastructure window may not stay open indefinitely. The company has the option to delay, withdraw, or refile. The target valuation of $80 billion might be a ceiling request rather than a realistic expectation. Market participants should treat it as an anchor for negotiation, not a point estimate. The source article itself observes that Switch may seek strategic investors before the IPO, which would provide a soft validation of the number. Without that validation, the target is merely a number in a press cycle.
The public record does not yet demonstrate that Switch's growth profile is closer to a GPU cloud than to a real estate trust. The S-1, assuming it is ever made public, will resolve the ambiguity. Until then, an auditor can only test the target valuation against three datasets: revenue, physical capacity, and competitive positioning.
Valuation Scenarios: $80 Billion Against the Ledger
Let us walk through the arithmetic precisely. The table below compares Switch's target valuation with credible public comparables. All figures are approximations based on the most recent disclosed data.
| Company | Enterprise Value (est.) | 2024 Revenue (est.) | EV/Revenue | Business Core | |---------|-------------------------|---------------------|------------|----------------| | Switch (target) | $80B | $1.0-1.2B | 67-80x | Colocation + AI high-density space | | CoreWeave | $35-50B | $1.6B | 22-30x | GPU cloud, NVIDIA-tied | | Equinix | $80-90B | $8.7B | 9-10x | Global interconnection + IDC | | Digital Realty | $50-60B | $5.7B | 9x | Global wholesale IDC | | GDS Holdings | $5-6B | $1.5B | 4x | China/Southeast Asia IDC |
Now hold these numbers next to the assumptions. If Switch's most recent fiscal year revenue is $1.5 billion, the implied EV/Revenue ratio is approximately 53x. If revenue is $1.0 billion, the ratio is 80x. If revenue is $700 million, the ratio approaches 114x. Each scenario requires a different narrative. Each collapses if the revenue figure disappoints. The EBITDA side is worse. A leased data center operator typically posts EBITDA margins of 45-55%. At $1.2 billion revenue, that is $540 to $660 million. An $80 billion enterprise value implies an EV/EBITDA multiple of 121 to 148x. Equinix trades at 18 to 22x. Digital Realty trades at 16 to 18x. To justify a 130x multiple, EBITDA must quadruple within three to four years. That requires revenue to compound at 60% or higher, and it requires the margin structure to hold while the company constructs capital-intensive assets at scale. I have audited this exact pattern in both DeFi and traditional infrastructure. The trajectory is not impossible. It is highly improbable without contractual revenue visibility that has not yet been disclosed.
The Yield Trap Pattern
In 2020, I tracked the liquidity flows of a yield farming protocol that advertised a 10,000% APY. I mapped the token emission schedule on-chain and concluded the incentive model was mathematically unsustainable. My report predicted insolvency within 45 days. The protocol collapsed in 41 days. The core mechanic was simple: the reward rate exceeded the protocol's ability to generate real revenue, so the system required an infinite supply of new capital to maintain the facade. I see a structural echo in the $80 billion valuation. The multiple assumes that AI data center demand will grow faster than supply for at least five years. It assumes that cloud providers and AI labs will continue signing leases at premiums, that power prices will remain manageable, and that the largest customers will not renegotiate terms when the 2026-2027 construction wave delivers new supply. Yield trap detected. Ledger does not lie.
The source article's own analysis reaches a similar conclusion with different language. It notes that if Switch's revenue is $1.0 billion, the multiple is 80x; if revenue is $700 million, the multiple is 114x. It also notes that existing IDC leaders are rapidly upgrading their own high-density capabilities. The competitive window for Switch's 'AI-ready' design advantage is closing. What was once proprietary thermal management is becoming a commodity feature in every new build.
What 400 Megawatts Says
Data center transactions are priced in megawatts. My experience auditing infrastructure deals, in crypto and in traditional data centers, tells me the first question is always: what is the operating capacity, what is the construction pipeline, and what is the reserved power pool? The source material does not answer any of these questions. That is not a documentation failure. It is a strategic omission. Market precedent from 2024 AI data center transactions puts the per-megawatt enterprise value at roughly $15 to $20 million for operating high-density capacity. At $80 billion, Switch would need between 400 and 530 MW of operating, leased capacity to match that comparable range. If Switch operates only 200 MW today, the implied per-MW valuation exceeds $400 million, an absurd premium unless the company holds multi-gigawatt reserves of permitted power and land. If Switch operates 500 MW or more, the valuation begins to look less insane, but still requires a substantial forward growth premium.
To put this in perspective, Equinix operates more than 250 data centers globally. Digital Realty operates over 300. Switch's campus model concentrates capacity in fewer locations, which can be operationally efficient but creates geographic concentration risk. If a single campus loses grid access or suffers a disaster, the revenue impact is not diversified. Constructing AI-grade data centers costs $10 to $15 million per MW, excluding IT hardware. A 500 MW expansion program, the minimum credible size for an $80 billion growth story, requires $5 to $7.5 billion in capital. The company will need debt and equity, which means dilution, which means the public market's return expectations must be met even after the massive capital raise. The S-1, if published, will reveal the true CapEx plan and the free cash flow trajectory. Until then, any claim that the $80 billion is 'reasonable' is an act of faith, not analysis.
Power Is the Moat, Not Cooling
The source article correctly identifies which asset is truly scarce: electricity. U.S. grid interconnection queues are running three to seven years. A data center operator with already-approved power allocations and land permits owns something that cannot be rebuilt quickly. This is the real barrier to entry in 2025. NVIDIA's latest silicon can be bought by anyone with capital. Transformers, switchgear, and grid access cannot be expedited with a purchase order. I have seen this dynamic in private market audits. A facility with 100 MW of existing load and another 150 MW of approved capacity is worth meaningfully more than a facility with identical IT load but no expansion rights. The market began pricing this gap in 2023. Switch's Nevada and Michigan campuses carry regional advantages, from solar resources to lake cooling. But the valuation-relevant question is not historical geography. It is the quantity of reserved but undeployed power capacity. The source article does not disclose it. This is precisely the data point that will separate a fair valuation from a speculative one.
Switch's carrier-neutral interconnection platform, Switch CIN, does add genuine value. Large GPU clusters require high-bandwidth connectivity between data centers, and an operator that controls its own inter-campus fiber can offer lower latency than a tenant that must purchase transit from third parties. This is an underappreciated asset. It is also difficult to value without knowing the traffic volume, the number of connected networks, and the revenue contribution of the network segment. The source material mentions none of these. An auditor cannot assign a premium to an unmeasured quantity.
The Landlord-Compute Binary
Switch's position between Equinix and CoreWeave is the central strategic question. If Switch remains a wholesale and retail colocation landlord, its legitimate valuation anchor is the traditional IDC framework, perhaps nine to twelve times revenue, not fifty to eighty times. If Switch transitions to GPU-as-a-Service, it must build an entirely different operational organization, compete with CoreWeave's NVIDIA relationship, and absorb the balance-sheet risk of owning tens of thousands of GPUs. If Switch tries a hybrid model, leasing shells to CoreWeave and others while also running its own compute stack, it must manage channel conflict with its own tenants. The source article's competitive table highlights this precisely: CoreWeave's GPU compute capability is strong, Switch's is unverified. The market will price the hybrid only after seeing direct evidence of AI revenue share and the margin mix.
The actual pricing mechanism in this industry is dollars per kilowatt per month for powered racks. AI racks at 30-100 kW command far higher unit prices than legacy racks at 5-10 kW, but they also consume far more electricity. The unit economics improve only if utilization remains high and power costs stay locked. If new supply floods the market in 2026, the price per kilowatt-hour of occupied data center space will compress. This is why the standard IDC multiple, 9 to 12x revenue, exists. It reflects a business with stable but modest growth, long lease durations, and significant capital reinvestment. Switch must demonstrate that its business escapes this gravitational pull.
The 2026 Supply Wall
The U.S. data center construction pipeline is measured in gigawatts. A substantial fraction of that capacity is scheduled to hit the market in 2026 and 2027. The law of this cycle is identical to every infrastructure cycle I have audited: when a wave of homogeneous assets comes online, pricing power migrates to the tenant. The customer concentration magnifies the risk. A handful of buyers, Microsoft, Meta, OpenAI, Anthropic, xAI, Oracle, and a small group of NVIDIA-linked clouds, account for most AI infrastructure demand. If one or two of these entities decelerate their buildout, the effect on the entire sector is enormous. I have verified this dynamic before. The 2024 ETF custody analysis I published showed that a single custodian's multisig setup created a systemic risk that the broader market ignored. In this case, the systemic risk is not a wallet. It is the concentration of demand on a handful of balance sheets. When the supply wave lands, the marginal asset loses pricing power. Mathematical collapse verified across every capital cycle I have audited since 2017.
ESG and the Electricity Floor
The S-1 will also contain a risk factor that headline coverage ignores: power price inflation. New utility contracts signed in 2024 and 2025 carry materially higher tariffs than contracts signed in 2018 and 2019. If Switch's existing power agreements expire and must be renegotiated during a grid-constrained period, margins compress precisely when the capital expansion plan demands more cash flow. The environmental angle is not separate from the financial one. Data centers are the fastest-growing category of electricity demand in the United States. State regulators are already rejecting or delaying permits for new sites due to grid reliability concerns. Switch's ESG profile, its renewable energy percentage, and its PUE metrics will be scrutinized by institutional investors who increasingly treat these factors as credit risks, not activist marketing. In a market where power is the binding constraint, energy strategy is valuation.
There is also a national security angle. If Switch hosts government or defense compute workloads, those contracts carry higher margins and strategic value. They also introduce supply chain review under U.S. executive orders affecting AI infrastructure. Any dependence on foreign-manufactured transformers, cooling equipment, or electrical components becomes a compliance risk. The source article scrubs this topic to a footnote. The S-1's risk factors section will tell a different story.
What the Bulls Got Right
The bull case is not empty. Switch has been operating since before the AI wave. It likely has positive EBITDA and real cash flow, which distinguishes it from loss-making AI clouds like CoreWeave. If Switch has already signed long-term build-to-suit leases with hyperscalers or GPU cloud providers, the revenue backlog provides genuine visibility. The source article itself notes that the 'AI narrative' may be less important than the 'power allocation' asset, and in that dimension, Switch's years of land acquisition and grid connection work may give it a durable advantage.
There is also a terminal-value argument. If Switch builds several large campuses at today's power prices, the long-term contracts lock in five, ten, or fifteen years of cash flow. In a period of rising electricity costs, these locked-in contracts become inflation-protected assets. That is a genuine economic advantage. But it only works if the landlord has actually locked in the power price on the expense side, and if the tenant has committed to take-or-pay leases on the revenue side. Both conditions must appear in the S-1.
A precedent is also on the bulls' side. CoreWeave's IPO, despite its losses, established that the market will pay AI-infrastructure multiples for scarce real assets. If Switch can prove that its AI-related revenue is above 60% of total revenue, and if the growth rate is in the 50-80% range, then 50-60x revenue is a defensible, if aggressive, multiple. Strategic investors like NVIDIA, Microsoft, or a sovereign fund could also pre-anchor the valuation by purchasing a stake before the listing. Confidential filing, in this context, is simply rational timing. The market is paying high multiples for anything with a credible AI story. Management would be negligent not to exploit this window.
The S-1 Will Decide
The $80 billion number is a request, not a fact. The S-1 will provide the ledger. Watch for three figures: trailing revenue, AI revenue share, and contracted but undelivered capacity. If revenue is under $1 billion, the rational value range is $35 to $50 billion, implying a 40-60% cut from the target. If revenue is above $1.5 billion with strong AI composition, the target becomes credible. The difference between those two outcomes will not be settled by narrative. It will be settled by the document Switch has, as of this writing, kept hidden. I have audited enough capital markets to know that the most dangerous sentence in a prospectus is 'we believe.' The second most dangerous is 'pro forma.' When the S-1 arrives, read the actual numbers. Do not read the slides. The ledger does not lie. The question is whether the market insists on reading it before writing the check.
