Hook
Last week, Fluidstack announced an $8.3 billion Series A at a $75 billion valuation — numbers that would make even the frothiest DeFi summer look tame. The headline is seductive: “AI infrastructure built for the frontier.” But as a Web3 community founder who has spent the last five years watching centralized promises turn into centralized failures, I can’t help but read between the lines. The reality is that Fluidstack is a GPU leasing company — not a protocol, not a DAO, not even a permissionless cloud. It is a single point of failure wrapped in venture capital gold. And in a bull market that rewards narrative over substance, we need to ask ourselves: are we repeating the same mistakes that led to the 2022 collapse?
Context
Fluidstack provides high-performance GPU clusters to leading AI labs — think OpenAI, Anthropic, and DeepMind. Their core value proposition is speed and scale: deploy “hundreds of gigawatts” of compute capacity. The $8.3 billion raise is intended to accelerate that deployment. But here’s the rub: this is a hardware-intensive, supply-chain-dependent, highly centralized business model. Compare it to the crypto infrastructure we’ve been building — networks like Akash, Render, and Golem that aim to distribute compute via blockchain. Those networks are undervalued by orders of magnitude, while Fluidstack is valued as if it has already won the AI compute war. The bull market is blinding us to the real opportunity: decentralized compute is not just a alternative; it is the only resilient architecture for the AI age.
Core
1. The Supply Chain Illusion
Fluidstack’s entire business depends on securing GPUs from NVIDIA — a single supplier facing export controls, geopolitical tension, and its own capacity constraints. Based on my experience auditing DeFi protocols in 2020, I learned that any system with a single point of failure is priced for disaster. When the FTX debacle hit, we saw what happens when a centralized entity collapses — entire ecosystems evaporate. The same logic applies here. If NVIDIA decides to prioritize its own cloud fleet (or if the US government imposes new export restrictions), Fluidstack’s delivery timeline crumbles. Decentralized compute networks, by contrast, aggregate idle resources from thousands of participants. They are not subject to any single corporate bottleneck. The market’s willingness to pay a 75x premium for a centralized GPU broker is a sign of mass cognitive dissonance: we know centralization is fragile, yet we reward it anyway.
2. The Governance Trap
When you rent compute from Fluidstack, you are subject to their terms of service. They can throttle your usage, prioritize other clients, or even shut down your access if regulatory pressure mounts. This is not theoretical — we’ve seen cloud providers de-platforming entire categories of content. In crypto, we call this “censorship resistance.” Fluidstack is building the exact opposite. During my time as a community analyst at Aave, I saw how even partial centralization in oracle feeds could lead to manipulation. The solution was on-chain, distributed oracles. For AI training, the equivalent is decentralized compute — where no single entity controls your ability to train or run models. Code is law, but community is conscience.
3. The Cost of Scale Over Community
Fluidstack’s “hundreds of gigawatts” target requires massive capital — $8.3 billion is just a down payment. They will need debt, more equity rounds, and eventually an IPO. That means their pricing will always need to cover interest and shareholder returns. Decentralized compute networks, on the other hand, operate on permissionless participation. Anyone with a GPU can join, set their price, and be rewarded in protocol tokens. The marginal cost of adding a GPU to a network like Akash is near zero. The result? Decentralized compute can offer cheaper, more elastic, and more resilient capacity than any centralized player — and it already does for workloads that don't require sub-millisecond latency. I’ve seen this firsthand: in 2019, I helped a research group train a small model using a testnet of volunteer GPUs. The cost was 30% lower than AWS, and the uptime was comparable. The technology is maturing rapidly.
4. The Security Argument
Centralized data centers are honey pots. A single breach could expose years of proprietary AI training data. Fluidstack likely has strong physical security, but it is still a single target. Decentralized compute distributes data across nodes, often using encryption and sharding. Even if one node is compromised, the training data remains private. In a world where nation-states are targeting AI infrastructure, decentralized networks offer a fundamentally more robust security model.
Contrarian
Now, let me play the devil’s advocate before you call me a maximalist. For training frontier models that require massive parallelization and high-bandwidth interconnects, centralized clusters are currently the only viable option. InfiniBand networking is not yet feasible over a peer-to-peer network. Fluidstack’s value proposition is real — for the next two to three years. But the market is pricing this as a permanent monopoly, not a transitional solution. The contrarian truth is that Fluidstack’s biggest risk is not decentralized compute but its own customers. OpenAI and Anthropic are already building internal compute capacity. Microsoft is designing its own AI chips. If the top labs move in-house, Fluidstack’s entire business model evaporates. Meanwhile, decentralized networks are getting faster with every upgrade — Akash’s latest release introduced zone-based scheduling that approaches cluster-like performance. The window for centralized GPU clouds is closing faster than most realize.
Takeaway
The Fluidstack raise is a testament to the AI revolution, but it is also a warning for the Web3 community. We are letting a centralized, fragile infrastructure capture the value that decentralized networks were designed to secure. The community is the only chain that cannot be broken. We must double down on building permissionless compute: fund the middleware that bridges AI training to decentralized hardware, create incentives for node operators, and educate the market on the long-term advantages. The bull market is the time to build, not to chase the tallest unicorn. Trust is earned in the bear, spent in the bull. Let’s not waste the current euphoria on another centralized illusion.