In 2020, I spent a week manually tracking $2.5 million in cross-exchange flows between Ethereum and Ethereum Classic after the fork. My colleagues were chasing ICO narratives; I was trying to understand which chain actually had liquidity pretending to be consolidated. It taught me something that has aged well: when infrastructure becomes a story, the spread between narrative and fees becomes a trade.
Today, the same play is running in AI. Cloud providers no longer sell compute; they rent outcomes. Inference prices are falling, capex is shifting toward maximizing rental yield, and the infrastructure supply chain is feeling its first real compression. For crypto, it's familiar. After the merge, Ethereum's Layer 2 ecosystem crossed the same line: property owners started charging rent, and hardware suppliers started bleeding.
The context matters. In AI, cloud providers moved from selling resources to selling platforms. Revenue shifted from one-off projects to subscriptions and per-token billing. Recurring revenue earned higher multiples, so the market rewarded landlords and quietly repriced shovel sellers. In crypto, the same shift is happening in the data availability stack. Rollups rent settlement and data space from Layer 1, then repackage it as cheap execution. The catch: their "rent" is largely subsidised by token treasuries. Stop the emissions and you stop the tenants.
When you look at the underlying unit economics, the story becomes sharper. I have audited rollup transaction flows since 2023; the data point that matters: the top ten rollups by total value locked generate, on a busy day, roughly 40–200 MB of compressed rollup data. That is a folder of holiday photos. There is no law that says "blobs must be scarce because we want to sell DA." The DA layer is overhyped because the narrative assumes scarcity. The reality is that efficiency will destroy the landlord's pricing power, just as inference optimization is destroying GPU rental margins in AI. This is not a bearish opinion; it is an arithmetic fact.
When a layer's core function is to be a cost center, its margin ceiling collapses to the cost of capital. In AI, quantization, speculative decoding, and model distillation lower compute per token. In crypto, byte code compression, data availability sampling, and zk-rollup aggregation will lower data per transaction. The infrastructure that remains essential — high-reliability power, cooling, interconnect — supports the landlord's uptime. The rest will watch margins grind down to manufacturing levels. If a product cannot price above its replacement cost, it is a commodity.
What does this mean for investment logic? AI investors have stopped valuing clouds by how many GPUs they bought; they now ask how much recurring AI revenue the clouds collect. That is a shift from capex worship to cash-flow discounting. Crypto is late, but coming. A Layer 2's valuation cannot be anchored to "total value locked" if that TVL is rented with liquidity mining incentives. Liquidity mining APY is not user demand; it is a subsidy paid to make TVL look real. Stop the incentive and the user vanishes. The same logic will force rollups to report net revenue, not gross deposits. That repricing is already visible in private term sheets: rollup teams get dilution based on revenue run-rate, not token price. Every landlord hates overcapacity; rentier economics is a starvation diet for unused assets.
From my audit work, the most misleading number in both worlds is capacity. AI data centers full of GPUs aren't assets if utilization stays below 40%. In crypto, a DA layer with huge throughput isn't valuable if no one uses it. The rentier era punishes idle capacity, not missing innovation. The strongest counterweight to landlord power is state-backed alternatives. China's clouds are building a parallel stack with domestic chips and their own toolchain, because the AI rentier depends on imported GPUs. In crypto, sovereign chains and appchains serve a similar function: they let users build on a foundation no single landlord controls. The long-term winners are not the best technology platforms but the most resilient rent collectors.
The second-order effect is a shift in hardware procurement. AI hyperscalers are designing silicon in-house because they can no longer accept the dominant vendor's margin. In crypto, the equivalent is application-specific sequencers and custom rollup frameworks. The infrastructure chain isn't just compressed; it's being vertically integrated by the landlords. When the landlord can produce the shovels, the independent shovel-maker becomes a leasehold. The 'infrastructure pressure' narrative misses the key shift: pricing power is not disappearing, it is migrating up the stack. The independent hardware maker becomes a tenant of the platform; the platform becomes a tenant of energy markets. In both worlds, valuable infrastructure is no longer sold by the pound.
Now the contrarian angle. Most people read the cloud rentier story as a warning: if infrastructure gets squeezed, crypto infrastructure will follow. I think the opposite. The pressure on rents is the market punishing inefficiency. For AI, the GPU bubble starts deflating from the middle. For crypto, the next cycle belongs to applications with actual net revenue, not protocols with the largest emissions budget. The landlords with durable moats — Bitcoin as a settlement layer, Ethereum as a settlement and liquidity anchor — will survive. Bitcoin, though, has already been transformed by the ETF into Wall Street's rent-collection vehicle. The management fee an ETF issuer charges is a rent payment, and it makes Satoshi's "peer-to-peer cash" vision more legacy than living. Value is the illusion we agree to sustain, and we now agree to sustain it through 0.25% management fees.
The final piece is resource physics. AI has one binding constraint that crypto also has: electricity. While chip prices fall and DA costs compress, power remains the only "super-rentier" in both worlds. Companies that own power assets, or can relocate computation next to abundant energy, are the true landlords. Everyone else is a tenant waiting for a cheaper room. This is why the infrastructure fear is partially misplaced: the pressure is real, but not evenly distributed.
The next phase will be defined by a single metric: net revenue per unit of end-user workload. For a rollup, that is revenue per transaction after DA and execution costs. For a cloud GPU tenant, it is gross margin per dollar of rented compute. When that number rises without subsidies, a new cycle begins. Until then, chaos is just liquidity waiting for a narrative. History doesn't repeat; it rhymes through margin compression. Liquidity is the only truth in a world of noise — and right now, liquidity says respect real landlords and ignore illusion-renters.