Everyone is watching the AI model benchmarks. No one is watching the infrastructure plumbing. Last week, Gartner dropped a neat little bomb: ‘neocloud’ providers—those specialized AI cloud shops like CoreWeave, Lambda Labs, Vast.ai—will capture 20% of the AI cloud market by 2030, worth $267 billion. The headline is seductive. It whispers a new era of decentralized compute, of agility, of data sovereignty. But I’ve been here before. In 2017, I spent four months tracing the liquidity ghosts of the ICO fog. The patterns are eerily similar: a surge of capital chasing a scarce resource, a narrative of disruption, and a structural flaw hiding just below the surface.
Tracing the liquidity ghosts through the ICO fog.
The context is simple. Traditional cloud giants—AWS, Azure, GCP—built their empires on multi-tenant virtualization, high margins, and broad service ecosystems. AI workloads, especially GPU-intensive training and inference, break that mold. They demand low latency, huge memory bandwidth, and absolute hardware isolation. Neocloud providers step into the gap: they buy up NVIDIA H100s by the thousand, rent space in power-cheap data centers, and offer bare-metal access with InfiniBand networking. No virtual machine overhead. No GPU sharing. Just raw compute, billed by the second. Gartner says that this ‘superior performance and flexible deployment’ will drive market share from near-zero to 20% in six years.
But here’s the core insight that the mainstream coverage misses: the neocloud model is structurally identical to the crypto mining boom of 2021. Both are asset-heavy, high-leverage, commodity-like businesses dependent on a single hardware supply chain—GPUs in one case, ASICs in the other. Both attract capital through narrative-driven fundraising, not organic cash flow. And both are exposed to a liquidity cycle that can vanish overnight.
Let me unpack this with my own scars. In 2020, I built an arbitrage model on Uniswap V2, trying to exploit impermanent loss against fiat volatility. I was buying the narrative of ‘DeFi as a parallel financial system.’ But the real alpha was in the plumbing—the liquidity recycling. I found that 60% of initial ICO liquidity was recycled within four hours, creating a false sense of demand. Neocloud providers are recycling the same narrative: ‘AI compute is the new oil.’ They raise billions in debt (CoreWealone got $2.3 billion in credit lines in 2023), buy GPUs, and rent them out at thin margins. The real test isn’t market share. It’s utilization rate. If their GPU clusters idle at less than 60%, the debt service becomes impossible.
Watch the GPU utilization, not the VC narrative.
And here’s the structural flaw: the neocloud’s entire value proposition rests on NVIDIA’s continued GPU dominance and the relentless scaling of AI model sizes. But NVIDIA’s roadmap is accelerating. H100 is already being replaced by B200. A neocloud that borrowed at 12% interest to buy H100s in 2023 will face asset depreciation when B200s hit the market—especially if traditional clouds (which can write off hardware faster) undercut pricing. The same thing happened to Bitcoin miners in 2022 when the S19 Pro became obsolete overnight. The liquidity cycle turns, and the leveraged players get liquidated.
The AI cloud is a permissioned ledger – check the block producers.
But the contrarian truth is that neocloud providers are not the disruptors. They are the canaries in the coal mine. They exist because traditional clouds were slow to optimize for AI workloads. But once AWS rolls out its own Trainium2 chip at scale, or Google’s TPU v5 becomes open, the cost advantage of neoclouds will narrow. The real disruption is not neocloud vs. traditional cloud—it’s the decoupling of compute from platform lock-in. And that’s exactly where crypto comes in.
From my experience modeling the Terra collapse three days before it happened, I learned that algorithmic stablecoins die when the market stops believing in the mechanism. Neocloud providers face an analogous risk: their market exists only as long as AI companies need to train and inference at scale. If the AI hype cycle cools—as it did in 2022 after DALL-E 2’s launch—the demand for GPU compute will contract. The neoclouds will be left holding billions in depreciating hardware, with no liquidity to service their debt. The parallel to crypto is unmistakable. DeFi summer ended when yield farmers realized the token rewards were printed out of thin air. AI compute’s value is no different. It’s only as real as the next funding round for open-source model makers.
Yet there is a bullish case for the strategic positioning. The neocloud model is a perfect infrastructure layer for the emerging AI-agent economy—machines paying machines for micro-transactions in near real-time. In 2026, I prototyped a payment layer for AI agents using low-latency Layer 2 settlement. The key requirement was atomic, sub-second finality. Neoclouds, with their bare-metal GPUs and low-latency networking, are the natural hosting environment for these agents. They can provide the deterministic performance that blockchains need for consensus and execution. If AI agents become the primary consumers of compute, neoclouds could become the ‘validators’ of the machine economy.
But that’s a long bet. Short-term, the risk is clear: the 20% market share prediction assumes no supply shocks, no regulatory crackdowns on data sovereignty (which increases compliance costs), and no competitive response from traditional clouds. Each of those assumptions is fragile. Data sovereignty, for instance, is expensive. Neoclouds must build data centers in every jurisdiction where they claim sovereignty. That capital expenditure eats into the pricing advantage.
In my 2022 post-mortem on Terra, I wrote: ‘The bubble breathes. Don’t confuse volume with viability.’ The same applies here. Neoclouds are not a new technology. They are a new packaging of GPU compute, riding a liquidity wave fueled by AI venture capital. The question is not whether they will capture 20% of the market. The question is whether they will survive the first bear cycle in AI demand—and whether crypto will be there to provide a more resilient, decentralized alternative.