When BofA, JPMorgan, and Oppenheimer simultaneously picked Palantir, Amazon, and Lam Research as their top AI stocks, the financial press cheered. A 255 target on Palantir. A 365 target on Amazon. A 400 target on Lam. The narrative was clean: AI demand is real, and these three companies are the purest proxies. But as a decentralized protocol product manager who has watched DeFi summer, NFT mania, and the 2022 crash, I see something else. The same infrastructure that Wall Street is betting on is the infrastructure that is quietly centralizing the intelligence economy. And for the blockchain industry, that is both a warning and an opportunity.
Context: The Three-Layer Stack of Centralization
The three stocks represent three layers of the AI stack. Palantir is the application layer—converting messy enterprise data into decision intelligence. Amazon (AWS) is the cloud layer—providing the compute and storage. Lam Research is the physical layer—building the machines that make the chips that power the servers. On the surface, this is a classic industrial chain. But beneath the revenue numbers lies a deeper pattern: every layer is consolidating power into a few hands. Palantir’s 653 commercial clients generate an average of $3.5 million in revenue each. That means 653 decision-makers control the AI workflow of thousands of companies. AWS’s 496 billion backlog is essentially a 2-year revenue guarantee, but it also means that the future of AI compute is locked into a single cloud provider’s roadmap. Lam Research’s 150 billion WFE forecast for 2026 implies that the world’s chip expansion is being dictated by a handful of equipment suppliers. This is not the open, permissionless future that blockchain promised.
Core: Code Betrays When We Do
During my time auditing the Zilliqa sharding implementation in 2017, I learned a hard lesson: decentralization is not a feature you bolt on later. It is a design philosophy that must be embedded from the first line of code. The current AI stack is being built with the opposite philosophy. Palantir’s growth is impressive—commercial revenue up 149% YoY—but its customer concentration is a red flag. The company’s high customer revenue ($3.5M average) means it is serving a few whales, not a long tail of users. This is exactly the pattern we see in DeFi protocols where a few whales dominate governance. Code betrays when we do. If we build AI infrastructure that centralizes data and decision-making, we will end up with a system that mirrors the very censorable, opaque systems we sought to replace.
Consider AWS’s self-developed AI chips. The analyst report highlights them as a growth driver, but from a blockchain perspective, this is a classic vertical integration move. Amazon is building a walled garden: if you want the best price-performance for AI inference, you must use Trainium chips inside AWS. This is the same logic as a centralized sequencer—it offers efficiency but at the cost of sovereignty. In my work integrating AI agents into decentralized identity protocols on Polkadot, I have seen how the need for verifiable compute clashes with the centralized cloud model. The audit trail of an AI decision is impossible to trust if the entire computation happens inside a black box. Burnout is the tax on innovation. The industry is burning out on the race to build faster models, but it is neglecting the tax of trust.
Lam Research’s 150 billion WFE forecast is perhaps the most telling signal. It means that the physical cost of AI is going to be enormous. Every new chip fab requires years of planning and billions of dollars. This is the opposite of the blockchain ethos of permissionless participation. The barrier to entry for AI compute is becoming insurmountable for individuals and small teams. Yet, the blockchain community has been building decentralized compute networks (Render Network, Akash, Golem) for years. The irony is that the same Wall Street analysts who are bullish on Lam Research are ignoring the very protocols that could democratize access to this compute. Code betrays when we do. We are building a world where the means of intelligence production are controlled by a few, and we are doing it with the same short-term thinking that led to the 2022 crypto crash.
Contrarian: The Pragmatism Test
But let me be contrarian against my own narrative. The centralized AI stack works. It works fast. It scales. Palantir’s customers are seeing measurable ROI—that is why commercial revenue is up 149%. AWS’s backlog suggests customers are making long-term commitments. Lam’s customers are building factories. The decentralized alternatives are still in their infancy. The test is not whether decentralized compute is more ethical, but whether it can deliver the same performance at a competitive price. So far, the answer is no. The market is voting with its dollars, and the dollars are going to centralized providers. The blockchain industry must face this reality: we cannot just preach decentralization; we must build systems that are 10x better, not just 10x more principled.
During the 2021 NFT explosion, I took a sabbatical in the Cordillera Mountains. I realized that the industry was prioritizing hype over substance. The same is happening now with AI. The hype is real, but the substance is being built on centralized rails. The contrarian angle is that this very centralization creates the conditions for a decentralized backlash. As AWS becomes more vertically integrated, customers will eventually seek alternatives. As Palantir’s whale concentration becomes a risk, enterprises will look for more transparent, verifiable data pipelines. As Lam’s equipment becomes a geopolitical football, the need for distributed, resilient manufacturing will grow. The window is now, but it will not stay open forever.
Takeaway: The Vision Forward
The AI stock rally is a canary in the crypto mine. It is telling us that the demand for intelligence is real, but the infrastructure is being built in a way that mirrors the old world. The blockchain industry has a choice: we can continue to build niche protocols for the crypto-native, or we can look at the AI buildout and ask: where is the verifiable compute layer? Where is the decentralized data provenance that can audit a Palantir decision? Where is the open marketplace for AI chips that can bypass Lam’s factory schedule? The answers are not yet clear, but the questions are urgent. Burnout is the tax on innovation. If we ignore the centralization of AI, we will pay that tax twice—once in the form of lost opportunity, and once in the form of a system that we cannot trust. The next bull run will not be about DeFi or NFTs. It will be about the intersection of decentralized infrastructure and artificial intelligence. The code is watching. Let us not betray ourselves.
