I remember sitting in a conference room in 2017, watching a Polymath presentation on tokenized equity. A founder claimed blockchain would “automate trust.” I was 33, idealistic, and desperate to believe. The slides were elegant. The code was open. But the soul—the actual human adoption—never arrived. Today, I feel the same shiver when I read about the AI infrastructure boom. The temples are being built faster than gods can inhabit them.
Steve Eisman, the investor who shorted the 2008 housing market and later the ICO frenzy, recently placed a new bet: he sold his AI application stocks and kept his infrastructure positions. “The infrastructure is real,” he told an interviewer. “The applications are promises.” For a man who made a career out of smelling financial rot, this distinction is not a casual opinion. It is a diagnosis of a systemic misallocation of capital that mirrors exactly what happened to blockchain between 2017 and 2022.
Eisman’s logic is surgical. He sees NVIDIA and the cloud providers as “picks and shovels” sellers—companies that collect royalty regardless of whether the gold is found. Meanwhile, the startups building AI companions, code assistants, and marketing tools are burning cash faster than users are generating revenue. The numbers back him: as of late 2025, the top 20 AI-native applications have average user growth of 12% per quarter, but revenue growth trails at 4%. Unit economics are negative for 60% of them. This is not a healthy ecosystem. It is a stack of kindling waiting for a single interest rate spike.
But here is where the story becomes intimate for those of us who live in the blockchain world. The infrastructure-versus-application debate is not new to us. We lived through Ethereum’s ICO boom, where L1 tokens soared while dApps had zero active users. We saw Solana’s compute clusters rise in price while its DeFi protocols lost 90% of their TVL. We watched Bitcoin L2s explode in hype while actual transactions on them remain below 5,000 per day. The pattern is not a coincidence. It is a psychological trap that affects every general-purpose technology: the market rewards the enabler before the enabler has proven anything is enabled.
The infrastructure narrative is seductive because it feels tangible. You can touch a GPU. You can benchmark hashrate. But economic value does not flow from computation alone. It flows from the exchange of meaningful goods and services that computation unlocks. In crypto, we learned this the hard way. After the 2018 crash, projects like MakerDAO survived not because they had the best blockchain, but because they had a product that people used to manage risk. DAI had utility. Uniswap had real volume. The infrastructure narrative was a ghost.
Based on my experience auditing DAO governance proposals during DeFi Summer, I remember a proposal from a prominent L1 to allocate 20% of its treasury to “AI-enhanced smart contract auditing.” The community voted yes without any evidence that auditors needed AI. The team just wrapped the word “AI” around a old service. The token price jumped 8% the next day. That is the same psychological tic that Eisman is warning about: we are paying for the label, not the labor.
Now, the crypto market is repeating the same mistake with AI tokens. Over the past seven days, the combined market cap of the top 50 AI-focused crypto projects has dropped 40%, according to CoinGecko. Tokens like FET, AGIX, and RNDR—which were marketed as “decentralized AI infrastructure”—have lost their entire 2024 gains. Meanwhile, the narrative has flipped: every new L1 now calls itself an “AI blockchain,” regardless of whether it has any active AI workloads. The parallels to 2017’s “blockchain for X” and 2021’s “metaverse” are impossible to ignore.
The contrarian angle is that maybe Eisman is half-right but half-blind. Infrastructure can indeed be a bubble when its capacity outstrips demand. But in the crypto realm, decentralized infrastructure—like Bittensor’s subnet architecture or Render’s GPU network—has an additional dimension: it creates a market for idle compute. This is not just speculation. It is an economic thermostat. If demand drops, price drops, and compute returns to smaller-scale users. The infrastructure is not a single monolithic bet; it is a permissionless market that self-corrects. Eisman’s framework, rooted in centralized finance, may underestimate this resilience. However, the critical nuance is that even these decentralized markets depend on an application layer that is currently invisible. Without a killer app that commands premium compute, the infrastructure market becomes a race to the bottom. Render’s token price reflects this: down 60% from its high, despite the network having more GPUs than ever.
What does this mean for the crypto builder reading this? If you are building an AI infrastructure protocol, your survival depends on proving that someone is paying for compute with real value, not speculative token rewards. I have been in governance meetings where teams point to “total value locked” as a proxy for adoption. It is a lie we tell ourselves. The only metric that matters in a bear market is revenue—and not token-incentivized revenue, but revenue from users who would not use your service if it cost them fiat money.
Let me be vulnerable here. I have my own losses from this cycle. I bought into a decentralized compute protocol in early 2024 because I believed the narrative. The founder had a beautiful GitHub. The tokenomics were elegant. But when I interviewed 50 users for my last manifesto, only 3 had actually used it for inference. The rest were speculators waiting to sell to a greater fool. I held too long, hoping the application would materialize. It did not. Eisman’s sell signal resonated with me because it validated the quiet guilt I felt about my own portfolio.
Curating the soul in a world of derivative clones. This signature captures the challenge: we must distinguish between projects that build genuine utility from those that merely clone the infrastructure narrative. The application layer is where soul lives—where code meets human need. Ethereum’s survival after 2018 came from DeFi and NFTs, not from its consensus mechanism. Bitcoin’s resilience came from its store-of-value narrative, which is an application of trust, not a technical spec. Similarly, the AI winners in this cycle will be the ones that solve real problems for real people: a doctor using an AI to detect cancer, a small business using a chatbot to handle customer service, a farmer using satellite data to optimize irrigation. Those applications are not GPU farms. They are narrow, scrappy, and deeply human.
The future I see is one where the current infrastructure mania collapses, and a smaller, more authentic set of AI applications rises from the wreckage. These will not be built on hyped L1s or overfunded compute networks. They will be built on the simplest stack that works—maybe even centralized cloud—and they will succeed by being useful, not by being decentralized. As a DAO Governance Architect, I advise my clients to ignore the infrastructure noise and focus on protocols that have credible paths to application adoption. The metaphor of “picks and shovels” only works if there is gold. Right now, the gold is buried under a pile of speculative deck slides.
Eisman’s next move will be watched closely. If he is right, expect a bloodbath in AI infrastructure tokens in Q2 2026. But even if the infrastructure survives, the true wealth will go to those who built the applications that put the infrastructure to work. The soul of this cycle will not be found in the hashrate. It will be found in the quiet, unsexy code that helps a person do something they could not do before.
