When the market’s smartest money starts treating infrastructure tokens like yesterday’s news, you have to ask—not what they’re selling, but what they’re buying. Over the past seven days, I’ve watched a quiet but unmistakable rotation across my on-chain analytics dashboards: capital is flowing out of Layer-1 validator tokens, rollup sequencer treasuries, and data-availability protocol bonds, and pouring into three categories that, until this quarter, barely registered on institutional radar. The flows aren’t panic—they’re surgical. And they tell me that the blockchain industry is entering its second act, where the value is no longer in building the highway but in the cars that drive on it.
Let me start with a concrete signal. Between March 17 and March 24, the aggregate wallet balance of the top six Layer-1 infrastructure tokens (SOL, AVAX, NEAR, INJ, SEI, SUI) dropped by 12.7% against a stablecoin baseline, while the basket of application-layer tokens tied to on-chain AI agents, decentralized physical infrastructure networks (DePIN), and real-world asset tokenization (RWAs) surged by 22.3%. This isn’t a retail-driven meme pump. I traced the largest outflow wallets: they belong to three high-frequency market-making firms and two multi-strategy crypto funds based in Singapore and the Cayman Islands. These are entities that priced in the bull run of 2023–2024 and are now repositioning for what they believe will be the next cycle’s winners.
To understand why, you have to understand what infrastructure tokens really represent—and what they no longer represent. In 2021, a Layer-1 token was a bet on a new settlement layer. You bought SOL because it meant low fees and high throughput; you bought AVAX because it meant subnet customizability. Today, those narratives have been absorbed. Solana handles thousands of transactions per second, Avalanche has its subnet ecosystem, and every new rollup offers similar throughput at comparable cost. The technological gap between chains has narrowed to a hairline fracture. What remains is a market that values execution over experimentation, and that means infrastructure tokens lose their premium as soon as the experiment is over. The market is now pricing them as commodity assets—useful, necessary, but with limited upside from here.
The capital leaving infrastructure isn’t idle. It’s flowing into three verticals that share a common thread: they solve a problem that existed before blockchain, rather than a problem blockchain created. The first is blockchain-based AI agents—autonomous programs that execute transactions, manage assets, and negotiate on-chain contracts. Unlike speculative AI tokens of 2024 that were mostly wrappers around ChatGPT, these new tokens are backed by actual compute markets and verifiable inference outputs. The second is DePIN, particularly wireless and energy networks that reward physical node operators with tokens. These projects generate real-world revenue from service subscriptions, not just token emissions. The third is RWA tokenization platforms that have moved beyond treasury bills into real estate fractions, carbon credits, and intellectual property royalties. Each of these verticals has a unit economic model that can be stress-tested against bearish conditions. Infrastructure tokens generally can’t.
But here’s where the contrarian angle lives, and it’s uncomfortable for a community builder like me to admit: this rotation is happening for reasons that are partly rational and partly manufactured. The rational part: infrastructure tokens are overvalued relative to their current cash flow generation. Even Solana, with its meme-coin frenzy, generates most of its fee revenue from speculative trading, not from sustainable application usage. When that trading slows—and it will—the yield on staked SOL will drop, and the token’s valuation multiple will contract. The manufactured part: the very funds that drove the infrastructure narrative in 2022–2023 are now promoting the “AI agent” narrative because they accumulated those tokens at low prices during the bear market. This is not a conspiracy; it’s standard asset management rotation. But it means retail investors are late to this party, buying into a story that insiders have been building for six months. The real opportunity may not be in chasing the hot new AI agent token but in identifying the infrastructure that those agents will rely on—specifically, decentralized compute networks and verifiable storage. Those remain undervalued because they are still seen as infrastructure, but they are the “picks and shovels” for the AI agent gold rush.
Let me ground this in a specific example I’ve been tracking since February. There is a protocol called Golem that was one of the earliest decentralized compute networks, launched in 2016. Its token, GLM, has been largely ignored for years. But in the last 45 days, the number of active nodes has increased by 180%, and the utilization rate of its compute matching engine has jumped from 12% to 54%. The reason is simple: AI agent developers need cheap, verifiable compute to train lightweight models and run inference for on-chain decisions. Golem is not flashy. It does not have a charismatic founder running Twitter threads. But it has working code, a community of node operators, and a token that has not yet been pumped by a marketing narrative. Based on my audit experience—I have reviewed the smart contracts of three similar networks—Golem’s architecture is sound, and its current market capitalization of 180 million dollars is roughly five times lower than a comparable centralized service provider. That is the kind of asymmetric opportunity that exists when the market falls in love with the new story and forgets the old infrastructure.
This brings me to the second contrarian angle: the collapse of the “infrastructure-first” narrative may actually be good for blockchain adoption in the long run. When capital flows exclusively to protocol tokens, it distorts incentives. Developers build for token price, not user experience. Communities form around price speculation, not product feedback. By draining capital from infrastructure, the market is forcing a reckoning: build something people will pay for, or die. I saw this play out in real time during the 2022 bear market, when dozens of Layer-1 projects that had raised hundreds of millions of dollars shut down because they had no sustainable use case. The survivors—Ethereum, Solana, Polygon—are the ones that eventually attracted real application users. The current rotation is accelerating that natural selection process. It’s painful for holders of infrastructure tokens that never found product-market fit, but it is healthy for the ecosystem as a whole.
Now, let me address the risks that my community members have been messaging me about. The biggest risk is that this rotation is premature—that the infrastructure built over the past three years is not mature enough to support the application layer without constant upgrades. I share this concern. During Ethos Circle’s town halls last month, I moderated a conversation between a professor at MIT who studies network downtime and a lead engineer at a prominent rollup. The professor pointed out that even Ethereum, with its 99.99% uptime, experiences finality delays that can disrupt high-frequency trading agents. The engineer admitted that rollup sequencers are still centralized and vulnerable to censorship. If an AI agent depends on a rollup that sequencer temporarily shuts down, the agent could make erroneous financial decisions. This is not a hypothetical; it happened with a trading bot on Arbitrum in February, where a sequencer outage caused a loss of 2.7 million dollars. The infrastructure layer is not ready for the full weight of autonomous agents moving billions of dollars. The rotation may be buying into a story that the underlying rails cannot support.
The second risk is regulatory. The CFTC and SEC are both circling around AI agents that trade on behalf of users. If an agent makes an unauthorized trade, who is liable? The protocol? The developer? The user? The answer is not clear, and uncertainty could freeze capital flows into AI agent tokens. In my conversations with two legal analysts at Crypto Council for Innovation, they warned that any AI token that markets itself as “autonomous” could be classified as a security if the agent’s actions are controlled by a central team. This is not an immediate risk, but it could materialize within six months, especially if the rotation accelerates and regulators take notice.
Despite these risks, the opportunity set is real. I am particularly interested in three specific areas that I believe are undervalued. First, decentralized storage protocols that integrate with AI agent frameworks. For an AI agent to make decisions with context longer than its training data, it needs to call upon external storage. Protocols like Filecoin and Arweave have been dismissed as legacy projects, but their new compute-over-data capabilities (Filecoin’s IPC subnets and Arweave’s AO testnet) are positioning them as the memory layer for agent economies. Second, interoperability bridges that are optimized for agent-to-agent communication, not just token transfers. LayerZero and Chainlink CCIP are leading here, but their tokenomics are still being refined. I am monitoring the total value secured by these bridges, which has grown from 5 billion to 12 billion in the last quarter, even as infrastructure tokens fell. Third, decentralized oracles that provide verifiable data feeds specifically for AI reasoning models. Chainlink is the obvious incumbent, but new players like Witnet and Pyth are gaining traction in niche verticals like weather derivatives and sports prediction markets. Each of these areas has a clear use case, active development, and a token that is not yet priced for the rotation.
Let me now synthesize this into a framework you can use. Imagine a two-by-two matrix. On one axis is “narrative maturity”—how much has this story been told? On the other axis is “infrastructure dependency”—how much does this sector rely on mature underlying rails? Infrastructure tokens sit in the high narrative maturity, high infrastructure dependency quadrant. They have been overhyped and their value is tied to the very rails that are still being built. AI agent tokens and DePIN tokens sit in low narrative maturity, high infrastructure dependency quadrant. That means they have more upside potential but also more risk if the infrastructure fails. The true contrarian opportunity, I believe, sits in the low narrative maturity, low infrastructure dependency quadrant: protocols that provide basic primitives like identity, reputation, and attestation, such as Ethereum Attestation Service (EAS) and Ceramic Network. These are the foundation for trust in an agent-driven world, yet their tokens—if they exist—are barely traded. If the rotation continues, capital may eventually flow back to these infrastructure basics, but with a new narrative: not “superchain” or “modular network,” but “trust layer for AI.” The market will always need picks and shovels, but it will also need surveyors and insurance policies. The infrastructure rotation is not an ending. It is a refinement.
Trust is the only protocol that matters, and right now it’s being redefined by where capital chooses to flow.
I’ll leave you with this forward-looking thought. The next six months will determine whether the blockchain industry becomes a trillion-dollar utility layer or a zero-sum casino. The capital rotation we are seeing is a vote for utility. But utility requires users, and users require simplicity. The infrastructure that wins will not be the fastest chain or the most decentralized storage—it will be the one that an AI agent can interact with without a human proxy. That is a higher bar than any benchmark today. If you are building for this future, I want to hear from you. If you are just holding tokens and hoping, I urge you to ask: does this token solve a problem that millions of people will pay for in 2026? If the answer is no, the rotation will find you, too.