Last week, as Nvidia’s stock surged 12% on the announcement of a $40 billion capital expenditure plan, I was staring at a very different dashboard. Over the previous seven days, a DeFi protocol called "Argo" had lost 40% of its liquidity providers. The correlation between these two events is not statistical—it is structural. Both stories speak to the same hidden force: the creation of artificial demand through capital-intensive signalling, not genuine user need. And as someone who spent the DeFi Summer of 2020 auditing Uniswap’s governance mechanisms, I’ve seen this playbook before. We called it "yield farming." Nvidia calls it "AI infrastructure investment." The mechanics are eerily similar—and the risk of a hangover is just as real.
Nvidia’s strategy is, on its face, a rational response to a booming market. The company aims to secure long-term supply agreements with TSMC, build out its own DGX cloud capacity, and pre-order HBM memory years in advance. But the language used by analysts—phrases like "artificial demand inflation" and "speculative hoarding"—sounds eerily familiar to anyone who lived through the liquidity mining craze of 2020. Back then, protocols inflated their total value locked by offering token incentives to farmers who had no intention of using the platform. Today, cloud giants and AI startups are ordering H100 clusters not because they have immediate compute requirements, but because they fear being locked out of future capacity. This is not demand; it is a fear-driven arms race. And when the fear subsides, the oversupply will crash down.
## The Infrastructure Trap Let me be precise about what $40 billion buys—and what it does not. According to industry estimates, that sum could build roughly 15–20 new gigawatt-scale data centres, or secure ten years of advanced packaging capacity. Nvidia’s investment is not a bet on a single product; it is a bet on the entire geometry of AI computing. The problem is that this geometry is being drawn on a mat that is rapidly shrinking. The number of entities that can actually use an H100 cluster at full utilization is probably fewer than 200 worldwide. The rest are buying them to "rent out to others" or simply to sit idle as a hedge against future shortages. We saw this exact pattern in the crypto mining boom of 2021: rigs were ordered months in advance, creating a false supply-demand equilibrium. When the price of Bitcoin dropped, second-hand GPUs flooded the market and mining companies went bankrupt. The difference now is that Nvidia itself is creating the illusion of demand through its own investment arm.
The core insight here is that capital investment, when decoupled from genuine user adoption, becomes its own feedback loop. Nvidia’s $40 billion is not a response to today’s compute consumption—it is a response to forecasts that assume exponential growth forever. Those forecasts are built on the assumption that every AI startup will eventually scale to millions of users. History suggests otherwise. During the DeFi Summer, many protocols raised millions based on TVL projections that assumed constant growth. Most of them are now ghost chains. Governance isn't a smart contract, it's a social contract. And the social contract of AI chip allocation is being written by a single party: Nvidia. That concentration should give every decentralist pause.
## A Governance Analogy from DeFi's Past To understand why Nvidia’s strategy is fragile, look at Uniswap’s governance overhaul in late 2020. I led a volunteer team of 15 developers that produced a 50-page white paper titled "Democratizing Liquidity." Our core finding was simple: the mechanism design of automated market makers was sound, but the governance layer—the layer where humans make decisions about fees, treasury, and upgrades—was still a traditional hierarchy dressed in DAO clothing. The same is true for Nvidia’s supply chain. The hardware is elegant, but the allocation rules are opaque. Who decides which AI project gets priority access to H100s? Nvidia’s sales team. How much of their investment is actually passed through to end users as lower costs? Unknown. The lack of transparency creates an artificial scarcity that drives prices above equilibrium.
During the 2022 Bear Market, I initiated the "Resilience Hub," a free mentorship program that connected junior developers with senior veterans. One of the things we learned was that when resources are scarce, the most efficient distribution mechanism is not a central authority—it’s a transparent, permissionless market. Nvidia’s $40 billion investment is the opposite of that: it is a centralized bet that centralization will continue to be the most efficient way to allocate compute. But ask yourself this: if the demand is real, why do tier-two cloud providers offer H100 instances at half the price of Nvidia’s own DGX cloud? Because the market is pricing in a glut that Nvidia wants to deny. Code is law, but people are the protocol. And the protocol here—the social arrangement for AI compute—is broken.
## The Contrarian Angle: Artificial Demand as a Necessary Evil Before we condemn Nvidia entirely, consider a contrarian perspective. Perhaps artificial demand is not a bug but a feature. During DeFi Summer, the explosion of yield farming—much of it driven by liquidity mining that was effectively "free money"—catalysed the development of Layer-2 scaling solutions, audit standards, and insurance protocols. The temporary inflation of TVL created a sandbox where developers could experiment. Similarly, Nvidia’s oversupply of chips could drive down the cost of inference, making AI accessible to startups that otherwise couldn’t afford it. The danger is not the investment itself—it is the centralised control. If Nvidia funnels chips only to its own cloud service or to favoured partners, the ecosystem becomes a feudal system. But if the chips leak into the open market, the artificial demand could become real supply.
This is where my work on the 2026 AI+Crypto Convergence Ethics Framework comes in. I convened a working group of 30 ethicists and developers to draft the "Autonomous Agent Accountability Charter." One of our central tensions was whether to incentivise the creation of more AI agents on-chain, even if many of them were useless, in order to stress-test the infrastructure. We concluded that permissionless experimentation is virtuous as long as the collateral is real—i.e., agents must stake something of value. The same logic applies to Nvidia’s investment: as long as the capital is deployed into open, competitive markets, the temporary inflation is acceptable. But if it reinforces a walled garden, it becomes a threat to the decentralised AI future that many of us are building.
## Where the Disconnect Hurts Most: The Developer Level Based on my audit experience during the DeFi Summer, the real cost of centralised resource allocation is not felt by the giants—it is felt by the small developer trying to prototype a new model on a shoestring budget. I spoke to a builder last week who had applied for an Nvidia GPU grant and was placed on a six-month waitlist. He ended up renting time on a decentralized GPU marketplace—one of the few that connects idle gaming cards to inference jobs. The irony is that Nvidia’s own investment may be crowding out the very innovation that justifies its existence. Artificial demand inflates prices, which excludes the very developers who create the next killer app. This is exactly what happened during the high gas fee era of Ethereum: small dApp developers were priced out, and the ecosystem shifted to Layer-2s. The response was not to control fees centrally—it was to build more open infrastructure. The parallel with Nvidia is obvious: the solution is not more centrally planned capacity, but more open, programmable compute markets.
During the 2022 Bear Market, we started the "Resilience Hub" because we saw talented developers leaving the space. Their reason was not lack of capital—it was lack of access. Access to compute, access to data, access to community. Nvidia’s $40 billion is a massive injection of capital, but if it doesn’t translate into improved access for the many, it will only benefit the few. Governance isn't a smart contract, it's a social contract. And the social contract of AI compute is currently being written by one company. That is not a sustainable foundation for an industry that claims to be about decentralised intelligence.
## A Personal Reflection: The 2022 Bear Market and the Vulnerability of Scale I have to be honest: part of my skepticism comes from personal experience. In 2022, I lost a significant portion of my savings because I trusted a protocol whose TVL had been inflated by a few whales. The numbers looked good—$2 billion locked—but the underlying usage was concentrated in five wallets. When one whale withdrew, the whole edifice collapsed. I learned that total quantity is not the same as distributed quality. Nvidia’s $40 billion is a quantity. The quality question is: how many independent, sustainable AI projects will be built with this capital? If the answer is "a few hundred hyperscalars," the investment is a liability. If the answer is "thousands of small teams," it is an asset. Based on the current allocation mechanisms, I fear the former.
The 2022 Bear Market also taught me that the most valuable resource in a downturn is not money—it is trust. We built the Resilience Hub without any token incentives; we just showed up every week for ten months. That trust is what kept 85% of our participants in the industry. Nvidia’s investment, if deployed without community alignment, will not earn trust. It will earn dependency. And dependency is the opposite of resilience.
## The Unspoken Parallel: AI and Layer-2 DA Oversaturation Let me draw a technical parallel that might seem strange but is deeply relevant. In the Layer-2 scaling space, there has been an explosion of rollups that use dedicated data availability layers—Celestia, EigenDA, etc. My assessment is that 99% of these rollups do not generate enough transaction data to warrant a dedicated DA layer. The infrastructure is being built for a future that may never come, funded by venture capital that expects exponential growth. This is exactly what Nvidia is doing: building computational infrastructure for a future of ubiquitous AI agents, while 80% of current users are still asking simple queries. The danger is that the capital allocated today creates a "placeholder" demand that masks the true need for lightweight, accessible solutions. The same way an L2 rollup with 10 transactions per second doesn’t need a custom DA layer, a startup with a prototype language model doesn’t need an H100 cluster. The market is being shaped by the tool, not the use case.
Uniswap V4’s hooks turned the DEX into programmable Legos—but the complexity spike scared off 90% of developers. Nvidia’s CUDA ecosystem is the same: incredibly powerful, but the barrier to entry is so high that only a fraction of AI developers can truly leverage it. The rest are forced into simplified abstractions that run on Nvidia’s cloud, thus locking them into the ecosystem. This is not a free market; it is a platform monopoly wearing a hardware disguise.
## Takeaway: The Real Failsafe Is Human Protocol So what is the takeaway for blockchain natives and AI builders alike? The same pattern repeats because the same incentives exist. Centralised capital, when invested without transparent governance and broad access, creates artificial demand that eventually collapses into a hangover. The countermeasure is not to stop investing—it is to embed accountability mechanisms into the allocation process. For Nvidia, that might mean publishing a public roadmap of chip allocation, or opening a portion of its capacity to a decentralised auction. For the blockchain ecosystem, it means building compute marketplaces that are permissionless and verifiable. We didn't build DeFi to have a few whales control the liquidity; we built it to distribute access. The same ethic must apply to AI compute.
I see three signals to watch over the next six months. First, cloud GPU utilisation rates: if they drop below 60%, the demand is fake. Second, the number of independent AI foundation models being trained on Nvidia hardware—if it declines, centralisation is winning. Third, the emergence of decentralised GPU networks (like io.net or Render Network) and their ability to attract serious workloads. If those networks thrive, Nvidia’s strategy is a hedge, not a threat. If they stagnate, we have a problem.
— Root: The 2022 Bear Market — Root: DeFi Summer
Code is law, but people are the protocol. And the protocol for AI compute is still being written. Let’s make sure it’s open.