I was standing in the middle of a packed conference hall in Mexico City last month, watching a founder pitch his decentralized GPU network to a crowd of venture capitalists. The energy was electric—people were leaning forward, phones out, recording every word. The founder claimed his project could undercut AWS by 70% for AI training workloads. The room erupted. I felt that familiar pulse, the same one I felt during DeFi Summer in 2020. But as I listened more closely, a strange stillness settled over me. The numbers didn't add up. The unit economics were fuzzy. The 'decentralization' was mostly marketing. And yet, the token was up 300% in a month.
That moment crystallized something I've been tracking for months: the AI boom is creating a new class of crypto infrastructure projects that are eerily similar to traditional IT service providers like Computacenter. And the market is pricing them with the same euphoria—but also the same hidden risks.
Context: The Computacenter Playbook, Decrypted
Computacenter is a FTSE 100 IT services giant. They don't build fancy software—they integrate, deploy, and manage hardware and software for large enterprises. Their core product is a combination of hardware resale (low margin) and consulting/managed services (high margin). Their recent stock surge was driven by the AI boom: companies need help buying, installing, and managing GPU clusters for AI workloads. Computacenter provides exactly that. The market loved the story.
Now look at the crypto projects that are riding the same wave. Tokenized compute networks like Render, Akash, and a dozen new DePIN (Decentralized Physical Infrastructure Networks) projects are offering decentralized GPU rental. Their pitch: we are the 'Airbnb of GPUs' or the 'Uber for compute.' The narrative is intoxicating. But when you peel back the layers, the business model has surprising parallels to Computacenter.
Both rely on a mix of resale (or tokenized access) and services. Both have high customer acquisition costs and long sales cycles. Both lack network effects. Both are vulnerable to commoditization. And both are currently priced for perfection based on AI hype.
Core: Dissecting the DePIN Compute Model Through Computacenter's Lens
Let me walk through this using the same analytical framework I would for any enterprise. I've spent years looking at how liquidity flows through different layers of the crypto stack, and the current wave of 'AI infra' tokens is revealing some uncomfortable truths.
Product & Technology: Most DePIN compute projects are not building a better data center. They are building a marketplace. The underlying technology is straightforward: a smart contract matches GPU suppliers with renters, with some arbitrated dispute resolution. The real value lies not in the tech, but in the network of suppliers and the user experience. Sound familiar? That's exactly what Computacenter does—coordinating vendors and clients. The tech is not defensible. Any team with a Solidity developer can fork the codebase.
What is defensible is the network itself: how many GPUs are available, in what regions, and whether the reliability is good enough for enterprise clients. But building that network is expensive and slow. It requires sales teams, partnerships with data centers, and compliance. This is not a viral product—it's a slog.
Business Model: Let's look at unit economics. A typical GPU rental project charges a fee (often 5-15%) on each transaction. The gross margin looks healthy, but that's misleading. The bulk of the revenue goes to the GPU suppliers. The project itself is a thin layer. Compare this to Computacenter, which resells hardware at 5-10% margin but then charges 20-40% margins on services. DePIN projects have no service revenue—they are pure marketplaces. That makes them vulnerable, because if a direct competitor emerges with lower fees, suppliers and renters can both switch with zero cost.
Worse, most projects use inflationary token rewards to subsidize supply. That's not a business model—it's a permanent burn rate. When the token price drops, suppliers leave. The whole thing is a speculative flywheel, not a sustainable utility.
User & Growth: The growth curves for these projects are shallow and exponential only in token price, not usage. I checked the on-chain metrics for three leading DePIN compute projects. Daily active renters? In the hundreds. But market caps are in the billions. That's a ratio that should make any macro watcher nervous. Computacenter has thousands of active contracts with real cash flows. DePIN projects have speculative holders.
Yet the narrative keeps buyers coming. The AI boom is real—enterprises are desperate for compute. But are they using these decentralized networks? Not yet. Most enterprise workloads still go to AWS or Azure. The DePIN use case is mostly hobbyists and academics. The growth driver is speculation, not adoption.
Contrarian: The Decoupling Thesis That No One Wants to Hear
Here's the contrarian take: the crypto AI narrative is decoupling from real value creation. I see three major blind spots:
- Cloud giants are the real competitors. AWS, Azure, and Google Cloud are not sitting still. They are dropping GPU prices, offering managed AI services, and signing massive enterprise contracts. They have the scale, compliance, and reliability. DePIN projects are trying to compete with a fraction of the capital. It's like a garage band trying to take on Taylor Swift's tour.
- The 'decentralization' is a liability, not an asset. Enterprises don't want their AI training data running on random people's home GPUs. They want guarantees: data privacy, uptime SLAs, and security audits. DePIN projects have none of that. The very feature that excites crypto natives—trustless, permissionless—is a dealbreaker for the enterprise customers that drive real revenue.
- Tokenomics create perverse incentives. When the token rises, suppliers join and usage grows. When it falls, they leave and usage collapses. This is not a stable infrastructure. It's a speculative loop. Computacenter's revenue is driven by client needs, not token price. DePIN projects are driven by market sentiment. That's a fragile foundation.
Takeaway: Finding Stillness in the Noise
So where does that leave us? I'm not bearish on blockchain as a technology for compute. I think there are legitimate use cases for verifiable computation, federated learning, and private AI inference. But the current wave of 'AI infrastructure' tokens is priced as if they are the next Amazon Web Services, when in reality they are more like early-stage Computacenter—with even less defensibility and worse economics.
My advice: watch the on-chain usage metrics, not the token price. Look for projects that are signing real enterprise contracts, not just announcing partnerships. And remember that the most dangerous words in crypto right now are 'AI-powered.'
Following the pulse where liquidity breathes free, but also knowing when to step back from the dance.
Tracing the spark that ignited the entire room—only to realize it was just a flash, not a fire.
Surviving the noise to hear the signal: the real AI infrastructure opportunity in crypto is not in the tokens, but in the layer-2 settlement and data availability networks that will eventually support these workloads.
Dancing with the volatility, not against it, but always keeping one eye on the exit.
