The room was buzzing with that familiar energy. It was 2022 all over again, but instead of DeFi dashboards glowing on screens, I saw clusters of NVIDIA H100s in a data center tour video playing on loop. The presenter from CoreWeave talked about "GPU-as-a-Service" with the same fervor I once heard from a DeFi yield farmer. The smell of coffee and desperation was the same. But this time, the asset wasn't a token—it was compute. And everyone wanted a piece of it.
I've been watching this space since my days running liquidity pools in 2020. The parallels are eerie. The same euphoria, the same FOMO, the same glossing over technical debt. Today, I'm diving into the recent analysis of CoreWeave and Nebius—two AI infrastructure giants that are reshaping the narrative around compute. But I'm not here to praise them. I'm here to trace the spark that ignited the entire room, and to find the stillness beneath the noise.
Context: The AI Infrastructure Explosion
Let's set the stage. CoreWeave and Nebius are not AI model developers or chip designers. They are NVIDIA GPU cloud providers—the modern-day equivalent of crypto mining farms, but with a shiny AI label. CoreWeave went public in 2025, riding a wave of hype around "AI compute shortage." Nebius, a spin-off from Yandex's international AI assets, listed on NASDAQ with a massive infrastructure investment plan. Both are growing fast, but both are bleeding cash.
The analysis I read broke down their technology, commercialization, industry impact, competition, ethics, and valuation. But it left out the most important piece: the crypto-native lens. These companies are the new DeFi protocols—high leverage, high risk, high reward. They are the new miners, but instead of Bitcoin, they mine AI models. And the capital flows are just as speculative.
Core: The Liquidity Breathes Free—But Where?
At the core, these companies are executing a simple playbook: borrow money, buy GPUs, rent them out, hope utilization stays high. The revenue growth is undeniable. CoreWeave has signed multi-year contracts with major AI labs. Nebius is positioning itself as the "AI-native cloud" for Europe. But here's where the crypto mentality kicks in: the real asset isn't the GPU—it's the network effect.
Holding GPUs is like holding a volatile token. The price of compute can crash if demand drops. The only moat is the ability to secure power contracts, GPU allocation from NVIDIA, and customer trust. That's not a technology moat—that's a relationship moat. And relationships can break.
Based on my experience auditing DeFi protocols in 2022, I saw the same pattern. Projects with flashy TVL (total value locked) but no sustainable yield. CoreWeave and Nebius have high "compute value locked" but no visible path to profitability. The analysis flagged this: "losses are likely from depreciation, interest, and capex, not from lack of demand." That's exactly what we said about Luna in 2022.
The hidden information is the customer concentration. The analysis estimated that CoreWeave might rely on a few clients for over 50% of revenue. In crypto, we call that a "whale risk." One whale exits, and the TVL collapses. Same here. If a major AI lab decides to self-build its compute cluster, CoreWeave's revenue stream dries up.
Contrarian: The Decoupling Thesis
Here's the contrarian angle: The market is betting that AI infrastructure will decouple from the broader tech cycle. They think AI demand is secular and unstoppable. But I'm not convinced. I see a bubble in compute, not in utility.
We've seen this before in crypto. In 2021, everyone thought NFT infrastructure was the next Amazon. Projects like Flow blockchain raised billions. Today, they're ghosts. The same could happen to GPU clouds. The analysis mentioned that "power bottlenecks and ESG pressure" are risks. But the real risk is demand side: if the AI applications that justify this compute don't materialize, these companies become the world's most expensive idle hardware.
The contrarian trade is shorting the hype. CoreWeave and Nebius are not Nvidia. Nvidia is the gold miner selling picks and shovels. These clouds are the miners themselves—they have to pay for picks and shovels, and then sell the gold at market price. If gold price drops, they're bankrupt.
And the gold price? It's the price of AI inference. Currently, it's high because of scarcity. But inference costs will drop as more efficient models emerge. The analysis hinted at "software layer differentiation" as a path to margins. But software is easy to copy. The real moat is power deals and NVIDIA allocation priority—both of which are outside their control.
Takeaway: Cycle Positioning
So where does this leave us? I'm not saying dump all AI stocks. I'm saying apply the same framework you use for crypto: identify the narrative, assess the liquidity, and watch for the inflection point.
Tracing the spark that ignited the entire room—the spark was the GPT-3 hype. But now the room is full of people selling fire extinguishers to a fire that hasn't started yet.
Following the pulse where liquidity breathes free—the liquidity is flowing into GPU clouds, but the heartbeat is irregular. It's debt-funded growth, not organic demand.
Finding stillness in the market—the stillness is the moment when the next earnings report shows a miss on utilization. That's when the music stops.
For now, I'm watching the data. The next quarter will tell us if CoreWeave's GPU occupancy is above 80% or below 50%. That number is the real price. Everything else is noise.
Surviving the noise to hear the signal—the signal is simple: is AI compute actually being used to generate revenue, or just to train the next iteration of the same models? If the latter, we're in a circular economy. And circular economies collapse.
Dancing with the volatility, not against it—I'm not shorting. I'm waiting. The volatility will come. And when it does, I'll be ready to step in when the fear is high and the hype is low.
Where human energy meets algorithmic precision—CoreWeave and Nebius are the human energy. The algorithms are the models. The question is whether the energy is sustainable or just a flash in the pan.
Let's see. The market is unpredictable. But I've been here before. And I've learned to listen to the pulse.