In the quiet hours before the market opens, I often reflect on the difference between noise and signal. Today’s noise is deafening: a single quote from venture capitalist Matt Seseri has been amplified as a ‘megaphone for the next bull run’ — AI infrastructure capital expenditures are entering a multi-year cycle, ‘driven by sustained demand for compute resources that will reshape how companies allocate capital.’ But as someone who watched the 2017 ICO craze collapse under the weight of its own marketing, I recognize the pattern. The market is FOMOing on centralized compute, forgetting that the original promise of blockchain was to decentralize trust, not to centralize compute under the umbrella of a few hyperscalers. The silence that precedes a crash is already settling in.
To understand why this cycle is a trap for the crypto ethos, we must first examine the narrative itself. Seseri’s statement, reported by Crypto Briefing, posits that big tech and infrastructure providers will pour billions into AI chips, data centers, and energy over several years. On the surface, this is bullish for NVIDIA, AMD, and the cloud triumvirate. But for a crypto education platform founder who has spent years teaching the value of decentralization, this narrative is deeply uncomfortable. It assumes a world where AI progress is wedded to centralized infrastructure — a world where permissionless innovation becomes a second-class citizen. The blockchain community’s counterpoint — decentralized compute networks like Akash, Render, and Bittensor — seems like a whisper against a hurricane. Yet history teaches us that the hurricane inevitably passes, and what remains is what was built on solid, distributed ground.
The Scaling Law is a Gamble
The entire capital expenditure thesis rests on a single, fragile assumption: that model scaling will continue to demand ever more compute. But this is not a law of nature; it’s a bet. If new architectures emerge — sparse models, mixture-of-experts, neuromorphic chips — the demand for traditional GPUs could plateau or even decline. The centralized players are all-in on this bet, but there’s no guarantee they’ll win. During my deep interviews with 12 core developers during the 2017 ICO era, many warned that over-reliance on a single scaling path creates systemic risk. Today, I see the same pattern: hyperscalers are building the ‘single chain’ of AI compute. If the scaling law breaks, the entire infrastructure cycle could unwind faster than it started. For crypto AI projects, this is a double-edged sword. On one hand, it means centralized compute may become cheaper and more abundant in the short term. On the other, it exposes them to the same fragility — if their business models depend on that same scaling law, they’re just as vulnerable.
The Power Wall
Data centers need immense power. Current grids in North America’s prime hubs — like Northern Virginia — already have waiting lists of several years. This is the hidden constraint that no analyst’s quote can wave away. The capital expenditure cycle assumes that power can be scaled as fast as GPU orders, but that’s simply not true. Utilities are not Silicon Valley startups; they take a decade to build new capacity. This bottleneck will force hyperscalers to build in regions with lax regulations and cheap energy, often in developing nations. The result is a form of ‘compute colonialism’ — the extraction of energy resources from vulnerable communities, controlled by a handful of corporations. Decentralized networks that leverage underutilized global compute, like idle gaming GPUs or solar-powered edge devices, bypass this entirely. They are more resilient because they don’t centralize power consumption. Noise fades. Value remains.
The ROI Trap
Seseri’s view ignores the monetization question. AI applications today — chatbots, copilots, image generators — are not yet generating enough revenue to justify the billions being spent. Microsoft’s Copilot, for instance, has yet to achieve broad profitability. If the killer app doesn’t materialize soon, these capital expenditures will become stranded assets. This mirrors the 2017 ICO boom, where projects spent exorbitantly on marketing and infrastructure without product-market fit. When the music stopped, only the projects with real utility survived. For crypto, this signals a flight to quality towards decentralized networks that have actual revenue streams — like Render’s GPU rental or Bittensor’s subnet rewards — rather than speculative token plays. Code executes. Ethics sustain.
Centralization vs. Decentralization
The biggest threat is that if AI development becomes dependent on a few centralized entities, the entire ethos of open, permissionless innovation is at risk. Crypto’s role is to provide an alternative — decentralized compute, decentralized AI models, and decentralized governance. But current crypto AI projects are tiny: the market cap of all decentralized compute tokens combined is less than a single quarter of NVIDIA’s revenue. The contrarian view, however, is that this massive investment will actually create the supercomputing resources that decentralized networks can later tap into via secondary markets. For example, overprovisioned GPUs from big tech could be rented out on Akash, turning a centralized asset into a decentralized resource. But that requires big tech to be willing to share economic value — a bet I wouldn’t take. The more likely outcome is that the infrastructure cycle reinforces a winner-take-all dynamic, squeezing out small players and making crypto AI a niche experiment.
The Counterpoint
Yet there is a contrarian angle that deserves attention. The capital expenditure cycle might actually be positive for decentralized crypto AI networks in the long run. By commoditizing compute through massive supply, it will eventually drive down costs, making affordable GPU time available for decentralized networks to scale. The same happened with cloud computing: early centralization of AWS led to commoditization, which eventually enabled startups to build on cheap infrastructure. If this cycle plays out, the cost of training and inference could drop by an order of magnitude, opening the door for permissionless innovation. The key is whether the crypto community can build the middleware to capture that value. Projects like Filecoin for storage, Akash for compute, and Bittensor for model training are well-positioned — if they survive the interim period of high costs and low demand.
Takeaway
The next wave of value creation in AI and crypto will not come from who owns the biggest cluster, but from who builds the most resilient, community-owned networks. The infrastructure cycle will pass. What remains will be the code that respects human autonomy and distributes power. That is the legacy we should build towards. Silence speaks louder than pumps.
