Switch filed confidentially with the SEC. Target valuation: $80 billion.
Do the arithmetic before the bankers do. Equinix, the world's largest data center REIT, generates $8.7 billion in annual revenue. Its enterprise value: roughly $85 billion. That's about 10x revenue. Digital Realty: 9x. Switch, by reasonable industry estimates, likely pulls in $1.0-1.2 billion. The $80 billion target implies a 70-80x revenue multiple — and an EV/EBITDA around 130-160x, assuming the standard 45-55% EBITDA margin for leased facilities.
Switch doesn't sell GPUs. It doesn't operate an AI cloud. It rents floor space, power, and cooling. CoreWeave, the NVIDIA-backed GPU cloud with $1.6 billion in sales, trades around 22-30x revenue. Switch asks for nearly three times that multiple while selling the warehouse, not the brain inside it.
This is not an IPO. It is a narrative stress test.
The Liquidity Map Has Shifted
AI infrastructure has become the liquidity sponge of this cycle. Pension funds, sovereign wealth vehicles, private equity — everyone is pricing compute scarcity as the new oil. American data center capacity sits near 20-25 gigawatts. Under construction: another ten. Grid interconnection queues stretch three to seven years. Power, not silicon, is the binding constraint.
Switch understood this early. It spent a decade building "over-engineered" facilities — proprietary thermal management pools, power density pushing 150kW per rack — when enterprise demand hovered at 5-10kW. That looked like capex madness. Then the GPU era arrived, and the madness became foresight.
But here's the uncomfortable fact the $80 billion narrative omits. Land and power contracts are not software. They are not defensible platforms. They are regulated, depreciating, capital-intensive structures with construction cycles that outlive hype cycles. The "AI infrastructure as platform" thesis is a story the sell-side tells to justify pricing a landlord like a SaaS company.
I audited 14 ICO whitepapers in late 2017. The pattern was clinical: wrap a real asset in a growth narrative, seek a multiple seven times the incumbent's, watch the story carry the price until the tokenomics — or the power bill — arrive. The underlying economics rarely move. Only the story does.

Three Conditions for the Multiple
For $80 billion to hold, Switch must prove three things. The first is revenue composition. AI-related income must exceed 60% of total. The company's legacy lies in enterprise colocation — compliance-grade cages for banks, insurers, and government agencies. These are stable, low-growth cash flows that belong in a REIT, not in a platform narrative.
Second, revenue must compound at 60-80% annually for three to five consecutive years. That's hyper-growth software math, not data center math. With ten gigawatts of new supply under construction during 2025-2027, the market is moving toward abundance, not scarcity. Rent per kilowatt will compress. Pricing power decays as the largest build-out since the interstate highway system comes online.
Third, the S-1 must show anchor wholesale contracts. Backlog is the only metric that matters. If a hyperscaler — Microsoft, Oracle, or an AI lab — has signed multi-gigawatt commitments, the valuation gets a floor. Rumors won't.
The margin profile compounds the risk. "Inflationary power contracts," the industry's euphemism for supply agreements signed amid grid congestion, have compressed forward project IRRs by 200-400 basis points across the sector. Several US states have already rejected new data center projects outright. Switch's locked-in power portfolio is its moat. That moat loses water every quarter electricity prices rise.
I stress-tested Compound and Aave in October 2020, modeling oracle failures and cascading liquidations three weeks before the market cracked. The tell was the same: thin depth beneath an inflated surface. Institutional AI infrastructure trades on identical mechanics — the wedge between narrative and cash flow eventually becomes a cliff.
The Decoupling Fiction
The market will call this a decoupling moment: AI infrastructure, newly independent of traditional REIT frameworks, deserves its own asset class. Consensus is fragile. The upgrade cycle that produced this multiple can produce its reversal just as quickly.
Here's the contrarian angle most analysts miss. Bubbles don't pop; they deflate slowly. The AI infrastructure premium won't vanish in a single crash. It will erode quarterly, as financial disclosures fail to match the multiple and the "platform" narrative collides with property tax assessments. Liquidity is a mirage in high heat — and the S-1 heat is considerable.
But if Switch succeeds — if the public market absorbs an $80 billion landlord — the exit door swings open for the entire compute ecosystem. Tokenized GPU networks, decentralized AI marketplaces, every layer-1 claiming to decentralize artificial intelligence suddenly holds a comparable asset. Their scarcity thesis gets validated by a Las Vegas data center operator. Their failure mode, however, follows the same curve: real infrastructure, speculative pricing, eventual reversion.
Position for the Reveal
Watch the S-1's backlog section and power acquisition disclosures. Anchor tenants make the narrative credible. Their absence makes $80 billion a mirage — visible at a distance, gone on arrival.
Code is law, until the chain forks. And the fork here arrives at the first earnings release.