Citigroup Raises Price Targets for Coreweave and Nebius — But the Real Story Is the Fragility of Centralized AI Compute

The market is buzzing with the news that Citigroup has raised price targets for Coreweave and Nebius, two of the most prominent AI cloud infrastructure providers. Coreweave goes from $142 to $159, a 12% bump. Nebius jumps from $278 to $324, a more aggressive 16.5% increase. On the surface, this is a straightforward bullish signal from a major Wall Street bank. But if you’ve spent any time in the trenches of decentralized infrastructure, you know that a price target is just a number on a spreadsheet. The real story is what these numbers conceal: the fragility of centralized AI compute, and the quiet rise of a more resilient alternative. Let me set the context. Coreweave and Nebius are not your typical cloud providers. They are specialized GPU-as-a-service platforms, designed specifically for the voracious needs of AI training and inference. Unlike AWS or Azure, which bundle compute with a thousand other services, these companies are pure plays on the Nvidia H100 and B200 chips. They are the modern-day equivalent of the gold rush shovel sellers. Citigroup’s upgrade suggests that the bank expects these companies to monetize their GPU fleets more effectively going forward. But the analysis behind the upgrade is opaque. No new contracts were announced. No earnings revisions were disclosed. The price targets float in the air, unsupported by the kind of granular data that a protocol PM like me relies on when auditing a smart contract. From hype cycles to hydraulic stability. That phrase has never been more relevant. The AI compute market is currently a hydraulic system: pressure builds as demand surges, but the release valves are expensive and fragile. Coreweave and Nebius are betting on a continuous flow of enterprise AI workloads. Yet the reality is that most AI workloads are still experimental. A single large customer like OpenAI or Microsoft can shift its strategy, build its own chips, or negotiate a better deal with a competitor. The price target assumes a stable demand curve, but the curve is actually a series of spikes and troughs. Based on my experience auditing the governance of lending protocols during the 2022 bear market, I’ve learned that when a single asset class is touted as "infrastructure" by analysts, the risk of overconcentration is high. The same principle applies here. These companies are not protocols; they are centralized companies with all the attendant risks: management turnover, geopolitical exposure to GPU supply chains, and the ever-present threat of a price war with hyperscalers. The 12% and 16.5% target increases might look like endorsement, but they are more likely a mechanical reaction to a rising tide. The tide is rising, but the boats are not all equal. Now, let me turn to the contrarian angle. The most bullish assumption behind Citigroup’s targets is that the demand for specialized AI compute will continue to outstrip supply. But what if the market overestimates the stickiness of centralized GPU clouds? The code is cold, but the community is warm. I’ve seen this dynamic play out in DeFi: centralized exchanges had a massive lead in 2020, but the ethos of self-custody and the technical elegance of decentralized protocols eventually eroded that advantage. The same could happen in AI compute. Projects like Akash Network, Render Network, and Golem are already offering decentralized compute marketplaces. They are not yet competitive on raw performance, but they are philosophically aligned with the values of the crypto-native community. And they are improving rapidly. Moreover, the price target report omits a critical factor: the depreciation cycle of GPU hardware. Nvidia is expected to release a new generation of chips every 18 months. Coreweave and Nebius will need to continuously reinvest capital to stay competitive. A decentralized network, by contrast, can leverage idle consumer-grade GPUs, reducing the cost of capital and passing savings to users. The Citigroup analysts may have modeled a 5-year depreciation schedule, but the real-world obsolescence of AI hardware is far faster. This is a structural risk that cannot be hedged with a higher price target. We are not just users; we are the protocol. That is the mindset shift that the market is ignoring. The AI infrastructure narrative is currently dominated by centralized companies, but the underlying technology stack is becoming more modular. The rise of verifiable inference, zero-knowledge proofs for model integrity, and on-chain compute attestation will eventually decouple AI compute from the physical hardware. When that happens, the value will accrue to the protocol layer, not the hardware layer. Citigroup is raising targets on the wrong level of the stack. Let me give you a concrete example from my own work. Earlier this year, I was leading a project to build a decentralized AI training dataset registry. We needed massive compute for preprocessing, but we refused to use a centralized GPU cloud. Instead, we used a combination of Filecoin for storage and a small Akash cluster for compute. The setup was slower and required more engineering effort, but the end result was a dataset that could be verified by anyone. That trust is something that Coreweave and Nebius cannot sell, no matter how many H100s they deploy. So what is the takeaway? Citigroup’s price targets are a signal, but they are a signal of momentum, not of fundamental strength. The real opportunity lies in the protocols that are building the decentralized alternative. If you are a developer or a project lead, do not be seduced by the ease of centralized GPU cloud. Build with the future in mind. The next bull run will not be about who has the most chips; it will be about who has the most trust. And trust, as we know, cannot be priced by a Wall Street analyst. Chaos is just order waiting to be optimized. The current chaos of AI compute is precisely the opportunity for decentralized networks to emerge. The price targets will fade; the protocol will remain.