The market is obsessed with the wrong narrative. Every tweet, every analyst note, every CNBC segment fixates on the AI sell-off. They blame profit-taking, technical corrections, or the latest macro scare. But beneath the surface noise, a structural truth is solidifying: AI compute demand will outstrip supply for the foreseeable future. Morgan Stanley’s recent report—framing the sell-off as a buying opportunity—is not wrong, but it is dangerously incomplete. It speaks to Wall Street flows, not to the underlying architecture of scarcity. And that scarcity is where crypto’s next real battle will be fought.
Let me be clear: the institutional view is a lagging indicator. They see the demand-supply gap and recommend buying Nvidia. We see the same gap and ask: who will actually deliver the compute? The answer is not just hyperscalers. It is a fragmented, inefficient, but rapidly maturing ecosystem of decentralized physical infrastructure networks (DePIN). Render, Akash, io.net, Golem—these names are no longer speculative footnotes. They are becoming the marginal suppliers of compute for a market that is about to see its pricing power shift dramatically.
But before we get to the opportunity, we must dissect the Morgan Stanley thesis. Their core argument: AI compute demand will exceed supply for years, driven by scaling laws, inference growth, and enterprise adoption. That is likely true for the next 18-24 months. The supply side is constrained by chip fab capacity (TSMC’s CoWoS packaging is already bottlenecked), power grid limitations (a single 100MW facility takes 3-5 years to build), and cooling infrastructure (liquid cooling is still niche). The result: a structural bid for compute. For traditional asset managers, this means buying semiconductor ETFs. For us, it means understanding that the same scarcity will pull compute prices higher, making tokenized compute markets—where you pay per job with a native coin—wildly sensitive to token velocity and utilization rates.
This is where the crypto narrative needs to sharpen. Most DePIN projects pitch themselves as “the Airbnb of compute.” That metaphor is lazy. Airbnb thrived on excess capacity; compute is about to have a deficit, not a surplus. The real value proposition is arbitrage—not of idle hardware, but of geographic and temporal pricing mismatches. A GPU cluster in Iceland running on cheap geothermal power can undercut a cluster in California by 40%. A network that can route jobs to the lowest-cost, lowest-latency provider in real time creates genuine economic surplus. That is the promise of projects like Akash and Render’s recent shift to real-time bidding. But execution risk is high. The coordination problem—matching supply, demand, and trust across thousands of independent nodes—is non-trivial. I have audited enough smart contracts to know that slashing mechanisms are often an afterthought. They are the Achilles’ heel.
Let’s talk numbers. According to estimates from the Morgan Stanley report and corroborated by my own modeling, the global AI compute demand (measured in H100-equivalent GPU hours) is growing at 40-60% CAGR through 2028. Supply growth from traditional cloud providers is constrained to 25-35% CAGR due to the bottlenecks I mentioned. That gap—the “compute deficit”—is roughly 15-25% of total demand annually by 2026. That is a multi-billion-dollar void that must be filled. DePIN projects currently represent less than 1% of total compute capacity, but they are growing faster (100-200% YoY) because they are starting from a low base. If they can capture even 5% of the deficit, you are looking at a market worth $10-15 billion in annualized tokenized compute revenue. That is not a niche; it is a new asset class.
But the contrarian angle is critical here. The crypto community is already exuberant about DePIN. I see it in Telegram groups, in tweet threads about “the next 100x.” That exuberance is a red flag. The reality: 90% of these projects will fail. Not because the technology is bad, but because the tokenomics are flawed. They emit tokens like a faucet, but the demand for compute is denominated in dollars, not a speculative coin. If the token price drops 50%, the cost to run a job on the network becomes cheaper in dollar terms—but the node operators (who earn the token) will leave. The network becomes a death spiral. We have seen this play out in storage markets (Filecoin, Arweave) and compute markets (Golem). The solution is a dual-token model or a stablecoin-pegged payment system, but that kills the speculative appeal. Until we solve this structural tension, the sector remains a casino dressed as infrastructure.
My own experience tells me to focus on the verification layer. In 2020, I watched DeFi protocols collapse because they had no reliable oracle system for liquidation triggers. In AI compute, the equivalent is the “proof of computation” problem. How do you know a node actually ran your job, did not cheat, and produced correct results? Current solutions—trusted execution environments (TEEs), optimistic verification, zero-knowledge proofs—are either expensive, slow, or both. The projects that solve this will dominate. I am watching projects like Fluence and io.net’s recent integration of TEE attestation, and also the emergence of verifiable compute on EigenLayer’s AVS. But these are early. The next six months will separate the serious teams from the vaporware.
Let’s ground this in a specific scenario. Imagine a startup that needs 10,000 GPU hours to fine-tune a model. They go to AWS: $35/hour for an H100 instance. Total: $350,000. They go to a DePIN network: $15/hour, but the job takes twice as long because of latency and consensus overhead. Total: $300,000. The savings are real, but the time trade-off is painful. For a VC-backed startup burning cash, time is money. The DePIN network must offer both price and speed—or at least price that justifies the delay. That is a narrow window. The market will reward networks that minimize latency, not just those with the cheapest idle GPUs.
Now the macro lens. This whole dynamic is unfolding against a backdrop of tightening global liquidity. The Federal Reserve is still hawkish on QT, the Yen carry trade is unwinding, and risk assets are under pressure. The AI sell-off that Morgan Stanley calls “technical” may also reflect a broader re-rating of growth stocks as rates stay higher for longer. Crypto is not immune. In fact, DePIN tokens—being high-beta, low-liquidity assets—could get crushed first. But here is the contrarian truth: a bear market in speculative tokens is exactly when real infrastructure projects should be built. Cash flows become king. The networks that demonstrate genuine revenue from compute sales—not token emissions—will survive and thrive. The others will decay.
Take a lesson from the 2020 DeFi liquidity crisis. I wrote a report then identifying that over-leveraged positions in Compound would fail. The team hedged against it. We profited. Today, the same principle applies: be suspicious of any DePIN project that relies on token incentives to attract supply. The only sustainable model is one where the end user’s payment covers the true cost of compute plus a margin for the node operator. If the token must be inflated to make the math work, it is a Ponzi scheme dressed as a protocol.
Collateral is just debt wearing a mask of trust. That remains true. In DePIN, collateral is the node’s stake. If the stake is too low, we get bad actors. If it is too high, we exclude good operators. The optimal staking design is still an open problem. I have seen smart contracts that try to solve this by slashing nodes for latency violations—but latency can be manipulated. The code does not care about your feelings, but it can be exploited. Always audit the audit.
We do not ride the wave; we engineer the tide. The tide right now is compute scarcity. We cannot change the wave of FOMO, but we can position ourselves to benefit from the inevitable structural shift. The institutional money will eventually flow into DePIN, but only after the tokenomics are proven, the verification is robust, and the market makers are not just dumping on retail. That window may open in late 2025 or 2026—exactly when the compute deficit becomes acute.
Here is my takeaway. Forget the AI sell-off headlines. They are noise. The signal is this: the compute deficit is real, crypto is the only market that can provide flexible, borderless compute supply, and the projects that survive this cycle will be the ones that treat their tokens as utility—not as lottery tickets. I have already begun shifting my portfolio toward infrastructure tokens with verifiable revenue pathways, not just hype narratives. If you are long DePIN, you should ask one question: can this network pay its node operators without inflating the token? If the answer is no, you are not investing; you are gambling. And in a bear market, gamblers lose their shirts. The tide engineers the outcome, but only those who see the full flow can ride it.
