Markets say AI-crypto convergence is a straight line to growth, but the data tells a different story. Over the past 90 days, total value locked in decentralized physical infrastructure networks (DePIN) has surged 240%, yet the underlying protocols are bleeding liquidity. Filecoin’s storage utilization rate hit 89% in Q3 2026, while its token price dropped 12%. This is not a contradiction—it is a supply chain warning.

Infrastructure providers in crypto are the MKS Instruments of the digital asset world. They do not issue tokens or run DeFi applications; they supply the subsystems—oracle feeds, data availability layers, compute networks—that make the rest of the ecosystem functional. The market treats them as commodities, but they are the first to feel strain when demand inverts.

Context: The MKS Parallel

In semiconductor manufacturing, MKS Instruments supplies RF power, pressure controllers, and vacuum systems to equipment OEMs like Applied Materials and Lam Research. It is a critical but invisible link. When AI chip demand spiked, MKS reported an 86% EPS jump, but margins shrank. The reason: it had to accept lower-margin orders to secure volume, and its customers—the OEMs—passed cost pressure downstream.

Crypto’s infrastructure layer is structurally identical. Chainlink, for example, provides oracle data to hundreds of DeFi protocols. Arweave and Filecoin offer decentralized storage. Celestia and EigenLayer supply data availability and restaking. These projects are not the front-end dApps that users see; they are the subsystems that enable the front-end. And right now, they are experiencing the same squeeze.

Core: The Seven Dimensions of Infrastructure Stress

1. Technical Throughput vs. Demand

Filecoin’s network storage capacity is 24 EiB, but active deal storage is only 12% of that. The headline number is misleading. The real bottleneck is retrieval speed and proof cost. For AI training data, high-frequency retrieval is essential—Filecoin’s proof-of-replication mechanism adds 30-second delays per read. This is the equivalent of MKS’s pressure controllers not being fast enough for new EUV lithography. The technology is not obsoleted, but it is not optimized for the new demand profile.

2. Supply Chain Concentration

Crypto infrastructure projects depend on a handful of centralized providers for their own inputs. Chainlink’s oracle nodes rely on AWS for execution. Arweave’s storage depends on Intel’s SGX enclaves for secure computation. If AWS or Intel faces a supply chain disruption (e.g., chip export controls, power outages), the entire crypto subsystem fails. This is analogous to MKS’s dependence on high-precision sensors from Japan or specialty metals from China. The 2022 bear market revealed that when liquidity dries up, the first to break are the dependencies, not the front-end.

3. Revenue Concentration Risk

Top 10 DeFi protocols account for 80% of Chainlink’s oracle data requests. A single protocol—like a major lending market—can represent 15% of a node operator’s revenue. If that protocol suffers a hack or regulatory action, oracle revenue disappears. MKS faces the same risk: its top three customers (Applied Materials, Lam Research, TEL) generate over 50% of its revenue. One customer’s inventory adjustment can wipe out a quarter.

4. Cost of Integrating New Demand

AI agents need verifiable inference. This is driving demand for zero-knowledge proofs and tokenized compute. But integrating these into existing infrastructure is expensive. Filecoin is adding FVM (Filecoin Virtual Machine) to support smart contracts—a $40 million engineering investment. Chainlink is building a new standard for AI oracle data (CCIP for AI). These costs are upfront, and margins suffer. MKS’s own integration of Atotech (a $5.1 billion acquisition) depressed its GAAP margins for two years.

5. Tokenomics as a Liquidity Trap

Unlike MKS, which has a real P&L, crypto infrastructure projects fund their growth through token emissions. When demand spikes, they need to spend more tokens on incentives (e.g., storage deals, oracle rewards). This inflates supply and dilutes the token price. The 86% EPS growth at MKS was real dollars; the equivalent in crypto is often fake growth—revenue in native tokens that are then sold for stablecoins. The market is only now realizing that token-based revenue is not the same as dollar-based earnings.

6. Regulatory Arbitrage Gap

MKS navigates semiconductor export controls by maintaining a compliance team and applying for licenses per shipment. Crypto infrastructure projects face a different regulatory patchwork: the EU’s MiCA requires oracle data providers to register as data service providers; the US SEC’s proposed rules on crypto custody affect storage networks. Most projects are not prepared. The regulatory arbitrage that MKS exploits (e.g., Nordic banking frameworks) has no equivalent in crypto yet. This creates a latent liability.

7. The AI-Crypto Convergence Tipping Point

MKS’s growth is driven by AI chip demand. In crypto, the AI-crypto convergence is real but overhyped. The data shows that only 5% of AI training data is stored on-chain today. The rest is on centralized cloud. The infrastructure projects that will survive are those that can serve as a complement to cloud, not a replacement. Celestia’s data availability layer, for example, is being used by rollups to publish transaction data, but the actual compute happens off-chain. This is a hybrid model—similar to how MKS supplies subsystems to ASML’s EUV tools, which are themselves a hybrid of laser and vacuum systems.

Contrarian: The Decoupling Thesis is Wrong

Most analysts argue that crypto infrastructure will decouple from traditional markets as AI-driven demand grows. I disagree. The decoupling narrative ignores the fact that crypto infrastructure’s inputs—chips, cloud services, bandwidth—are still priced in fiat dollars and subject to the same macro liquidity constraints. When the Fed tightens, the cost of capital rises for all hardware-dependent projects. The crypto subsystem is not a hedge; it is a leveraged bet on the same global supply chain that MKS operates in.

Furthermore, the move to “dedicated data availability” (DA) layers is overhyped. 99% of rollups do not generate enough data to need dedicated DA. They are buying a product they don’t need, driven by VC narratives. This is like MKS’s customers buying a high-end pressure controller for a mature process node—it adds cost without benefit. The result will be a shakeout: only the top 3 rollups will survive, and the DA layer will consolidate into a duopoly (Ethereum’s blob space and Celestia).

Takeaway: Position for the Contraction

Survival is the first metric of success. The next 12 months will see a liquidity crunch in infrastructure tokens as real AI demand exposes the gap between hype and capacity. Projects with high token dilution, low dollar revenue, and concentrated customer bases will see 60-80% declines. The ones that survive will be those that pivot to a subscription model for node operators, akin to MKS’s recurring service revenue.

I am currently shorting infrastructure tokens with high valuation and low utility—specifically those that rely on the “AI-crypto” narrative without a clear integration path. Markets lie, but liquidity tells the truth. The truth is that the infrastructure layer is the most fragile part of the ecosystem, and the AI boom is accelerating its fractures, not healing them. We do not predict; we position. And the position is to be underweight infrastructure until the supply chain corrects.