Tracing the code back to its chaotic genesis, you will find that every market cycle—whether crypto or AI—begins with a single, unsustainable narrative. This week, Crypto Briefing dropped a headline that made me spit out my coffee: “AI infrastructure boom drives Anthropic valuation toward $1.2T by year-end.” Let that sink in. One point two trillion dollars. That is more than the combined market cap of every Layer 1 blockchain outside Bitcoin and Ethereum. It is roughly the GDP of South Korea. And it is assigned to a company that, by its own admission, still loses money on every API call, whose “Constitutional AI” is a branding exercise more than a technical moat, and whose primary competitive advantage is being bankrolled by Google and Amazon.
Where logic meets the absurdity of market hype, we find the same pattern that wrecked portfolios in 2017, 2021, and 2022. A macro tailwind (AI infrastructure) is conflated with the success of a single player (Anthropic) to justify a valuation that has no basis in revenue, profit, or technological superiority. As an open source evangelist who spent 2017 explaining why Ethereum wasn’t just “Bitcoin with apps,” and 2020 dissecting the illusion of DeFi yields, I see the same conflation happening now. The AI infrastructure boom is real—NVIDIA’s data center revenue quadrupled, cloud providers are building clusters at hyperscale, and every enterprise is scrambling to adopt LLMs. But the value accrual is not linear. The “picks and shovels” (compute, networking, energy) are capturing most of the upside, not the “gold miners” (model companies). Anthropic, like most model layer firms, faces a brutal commoditization trap. Their best models are already being replicated by Meta’s open-source Llama and Google’s Gemini at a fraction of the cost. The valuation of $1.2T assumes Anthropic will not only maintain a lead but also monetize it at multiples higher than OpenAI—which is itself struggling to justify its $80B valuation.
The Infrastructure Paradox
In the silence between the block hashes, I have watched the same drama play out in crypto. During the 2021 NFT boom, projects like OpenSea were valued as if they would own the entire digital art market forever. Three years later, OpenSea’s valuation has collapsed by 90% as Blur, LooksRare, and decentralized marketplaces ate their lunch. The infrastructure (the Ethereum blockchain, IPFS, and the ERC-721 standard) remained valuable, but the application layer was a battleground of zero-sum competition. The same is true for AI. The infrastructure—data centers, high-bandwidth interconnects, GPU clusters—is the bottleneck, and its owners (TSMC, NVIDIA, Equinix) are the real winners. Anthropic, on the other hand, is a tenant. They rent compute from AWS and Google Cloud, pay for electricity, and have no proprietary hardware advantage. Their differentiation—safety research and model alignment—is a cost center, not a revenue driver. When I audited 50+ institutional reports in 2024 for my series “Beyond the ETF,” I found that 80% of analysts missed this basic fact. They treated “AI company” as a monolith, ignoring the stark differences between infrastructure providers and model builders. The same mistake led to the ICO mania: “blockchain” was the buzzword, but 90% of projects had no defensible tech.
Why the Crypto Playbook Applies
Crypto markets have already priced in the lesson that “infrastructure booms” are not uniform. During the DeFi summer of 2020, liquidity was abundant, but it fragmented across dozens of protocols. VCs pushed narratives like “Layer 2 scaling will solve Ethereum’s gas problem” to juice valuations, yet the reality was that most L2s were redundant. The same is happening in AI. Every week a new “decentralized AI” project raises millions, promising to democratize compute. But the market is already splitting. Centralized providers (OpenAI, Anthropic, Google) gobble the enterprise contracts, while open-source models (Llama, Mistral) eat the long tail. The only players with sustainable moats are those that control the supply side—NVIDIA for chips, AWS for cloud, and soon, decentralized compute networks like Akash or Render that offer verifiable, permissionless compute.
Based on my audit of 15 AI-decentralized compute projects, I have seen a recurring flaw: they copy the centralized model’s business logic but ignore the incentives. In crypto, we learned the hard way that supply-side incentives (miners, stakers) must be carefully calibrated. AI compute networks that reward GPU providers with native tokens risk the same liquidity fragmentation problem that plagues DeFi. The narrative that “decentralized AI will replace AWS” is as naive as the 2017 claim that “blockchain will replace the internet.” It won’t. It will complement it, and only for specific use cases where trustlessness matters—like verifiable inference for financial auditing, anonymous compute for private data, or censorship-resistant training.
The Contrarian Angle: Why the $1.2T Narrative Is Bullish for Crypto
Logic fails, but the narrative persists. The very absurdity of the Anthropic valuation says more about the market’s hunger for a new story than about the company’s fundamentals. And that hunger is exactly what crypto needs to break out of its sideways chop. In a consolidation market, the biggest risk is narrative fatigue. The “ETF approval” story is played out. The “Bitcoin as digital gold” story is half-baked. But “AI infrastructure is the new internet” is a narrative with legs. If the financial press is willing to assign $1.2T to a single company, imagine what they will assign to the decentralized alternative. That is where the contrarian opportunity lies.
Consider the following: the AI boom will create an insatiable demand for verifiable compute. Why? Because no enterprise will fully trust a black-box model from a single provider after the first scandal (a biased loan algorithm, a hallucinated legal citation, a leaked trade secret). The solution is not better centralized models; it is open-source models run on decentralized hardware, with every inference recorded on a public ledger. The same logic that drove the shift from on-premise server to cloud will now drive a shift from centralized cloud to decentralized cloud—but only for workloads that require transparency. In 2022, when I analyzed the collapse of FTX, I wrote that “trust is a bug, not a feature.” The same applies to AI: if your business depends on a model you cannot inspect or audit, you are not sovereign.
An evangelist who doubts his own gospel: here is the dark side. The decentralized compute narrative is already being co-opted by the same VCs who manufactured the liquidity fragmentation myth. They are funding 50 “AI-on-blockchain” projects that will compete for the same finite pool of GPUs and developers. Post-Dencun, Ethereum’s blob data will be saturated within two years, and rollup gas fees will double. The same pattern will hit decentralized AI networks: as demand for verifiable inference grows, the cost of storing proof-of-inference on-chain will skyrocket. The projects that survive are those that abstract the blockchain layer—like EigenLayer’s proof-of-work for AI—rather than shoving every calculation onto a smart contract.
Takeaway: The Real Value Is in the Middleware
So where does a blockchain evangelist place his bet? Not on the $1.2T myth, and not on the decentralized AI clone armies. The value is in the middleware that bridges centralized AI capability with decentralized trust. Think of zero-knowledge proofs for model inference, oracles that feed real-world data to LLMs, or token economic games that incentivize honest computation. In the same way that the 2020 DeFi boom was won by the composable layers (Uniswap, Aave, Chainlink), the 2025-2026 AI-crypto convergence will be won by those who build the pipes, not the pools.
When the AI hype cycle corrects—and it will, because all hype cycles correct—the projects that survive will be the ones that solve a real coordination problem. Not “AI on the blockchain” as a slogan, but “verifiable computation at scale” as a product. The $1.2T valuation is not just wrong; it is dangerous because it distracts from the actual opportunity. The infrastructure boom is real, but the value accrues to the infrastructure’s owners, not its renters. Decentralized compute networks, properly designed, can become the owners. But only if they avoid the same narrative traps that inflated OpenSea, then Anthropic.
In the silence between the block hashes, the machines are learning. But the ledger still whispers the same lesson: code is not law, incentives are. Verify, then doubt. And above all, don’t bet the farm on a narrative that sounds too good to be true. It always is.
