The Picks and Shovels Paradox: JPMorgan’s AI Warning Is a Mirror for Crypto’s Infrastructure Bubble

JPMorgan’s recent analysis of the AI semiconductor cycle reads less like a sector report and more like a prophecy—one that crypto’s infrastructure faithful would do well to heed. The bank’s core insight is simple yet devastating: the suppliers of AI compute (NVIDIA, SK Hynix) have been extracting disproportionate value from their downstream customers (cloud giants like Microsoft, Google, Amazon). That imbalance, they argue, is unsustainable. Capital expenditure growth is forecast to collapse from +100% in 2026 to a meager +7% by 2028. The inevitable result? Pricing power shifts, demand slows, and the “picks and shovels” narrative cracks. Code doesn’t care about narratives. It only reflects the underlying economics—and the economics of infrastructure-first markets have a brutal habit of resetting. As someone who spent 2017 auditing whitepapers during the ICO boom, I’ve seen this script before. Then, the “picks” were smart contract platforms promising to host every future application. Today, the picks are Layer 2 rollups, modular blockchains, and restaking protocols. The pattern is identical: a flood of capital into infrastructure, spurred by the belief that demand for blockspace is infinite. But infinite demand is a myth, and JPMorgan’s prediction of slowing cloud capex is a reminder that even the most hyped technology faces a return-on-investment reckoning. Consider the current state of Ethereum’s ecosystem. The total value locked in L2s like Arbitrum, Optimism, and Base has surged past $50 billion, while mainnet gas fees have fallen to multi-year lows. On the surface, this signals healthy scaling. Dig deeper, and you see the imbalance: the infrastructure providers (L2 sequencers, data availability layers) capture fees and token value, while the applications—the actual users of that infrastructure—struggle to monetize. DEX volumes on L2s are growing, but the fees collected by applications often lag behind the token inflation of the infrastructure tokens themselves. Soulless finance is just empty pixels if the underlying economic activity doesn’t generate sustainable yield. This mirrors the JPMorgan thesis precisely. Cloud providers (the applications layer) are seeing their AI service margins squeezed by the pricing power of NVIDIA (the infrastructure). The cloud giants are fighting back with custom ASICs—in crypto terms, application-specific rollups or sovereign chains that bypass general-purpose L2s. The message is clear: when infrastructure becomes too expensive, the downstream will invent alternatives. I’ve seen this firsthand in my own governance participation during DeFi Summer 2020, where Compound’s community debated whether to keep protocol fees low to incentivize usage—essentially, protecting the application layer at the expense of the underlying token price. Code doesn’t lie. The data on smart contract activity across L2s versus their token valuations tells a story of disconnect. Take the top six L2s by TVL: their combined daily active users rarely exceed 2 million, yet their fully diluted valuations sum to over $80 billion. That’s a price-to-active-user ratio that would make even NVIDIA’s multiple look conservative. The contrarian angle? Perhaps crypto is different because it enables programmable money and composable applications that accelerate network effects. Maybe AI agents will eventually drive real demand for blockspace. But JPMorgan’s report reminds us that such optimism must be tempered by the reality of capital cycles. Cloud providers are already slowing their AI chip orders; crypto will not be exempt from the same cycle. The real lesson is about narrative decay. In 2022, I saw it happen to Terra—a project that promised infinite demand for its synthetic assets. When the infrastructure stopped yielding returns, the narrative collapsed. Today’s infrastructure narrative—modular, L2, restaking—may face a similar test. The key signal to watch is not TVL or total gas consumed, but whether applications can generate real revenue that justifies the infrastructure spend. If they cannot, the pricing power will shift from infrastructure tokens to the user-facing applications, or worse, the capital will simply stop flowing. Based on my experience in the 2017 ICO cycle and the 2022 bear market, I’ve learned that the most dangerous phrase in crypto is “this time it’s different.” JPMorgan’s AI warning is a mirror—it reflects a structural truth that applies equally to our industry. The picks and shovels are valuable only as long as the miners keep digging. When the gold becomes scarce, the shovel prices adjust. Soulless finance is just empty pixels. The market will eventually demand that infrastructure proves its value through the utility of the applications it hosts. Code doesn’t create value—people and their willingness to pay for services do. As we enter a potential slowdown in capital expenditure across both AI and crypto, the wise investor will watch not the shovels, but the dirt.