Hook
When NVIDIA’s Q3 2023 revenue surged 206% year-over-year, driven entirely by data center demand, the signal was unmistakable: the AI infrastructure buildout had entered a new phase. But the real undercurrent surfaced last week when venture capitalist Rudina Seseri told Crypto Briefing, “We are entering a multi-year capital expenditure cycle in AI infrastructure that will reshape how technology giants allocate resources.” That line, delivered in the context of a crypto-native publication, wasn’t just about NVIDIA or Microsoft—it was a warning shot to every decentralized network that relies on the same GPU supply chain, the same energy grids, and the same capital flows. Over the past seven days, I’ve seen three separate DePIN protocols on my radar lose 40% of their liquidity as institutional allocators rotated into AI-centric plays. This isn’t a side story—it’s the macroeconomic tectonic shift that will define the next decade of blockchain infrastructure.
Context
Seseri is a partner at Glasswing Ventures, a firm that has backed AI-native companies since 2017. Her assertion—that the current wave of AI capital expenditure is not a one-off spike but a sustained, multi-year cycle—aligns with recent guidance from hyperscalers: Microsoft committed to a 40% increase in capex for fiscal 2025, Amazon pledged $150 billion over the next three years, and Meta doubled its 2024 infrastructure budget. These aren’t small numbers. For context, the entire global GPU market in 2020 was roughly $20 billion; by 2026, just the data center segment is projected to exceed $200 billion. The logic is straightforward: training cutting-edge models (GPT-5, Gemini Ultra 2, Llama-4) requires ever-larger clusters, while inference at scale demands geographically distributed compute. The result is an unprecedented investment in chips, networking, power, and cooling—what I call the “Seseri Floor.”
But why does this matter for blockchain? Because the same physical resources—GPUs, ASICs, data center space, and energy—are the bedrock of decentralized infrastructure. Proof-of-work mining, zero-knowledge proof generation, decentralized GPU networks (Render, Akash, io.net), and even Layer-2 sequencers all compete for the same silicon. When hyperscalers commit to multi-year capex, they lock up supply lines, bid up energy contracts, and raise the barrier to entry for any community-based project. “Code is law, but people are the protocol,” I wrote during the 2022 Bear Market, watching miners capitulate. Now, the law is being rewritten by capital allocation decisions made in Seattle and Shenzhen.
Core Insight
Let me be direct: the AI capex cycle is both a threat and an opportunity for decentralized networks, but the threat is more immediate. Based on my audit experience with TrustChain in 2017 and DeFi Summer governance work in 2020, I’ve learned to read infrastructure signals as leading indicators of network health. What I see today is a classic resource squeeze.
1. GPU scarcity will persist through 2025–2026. NVIDIA’s B200 GPU, expected to ship in late 2024, is already oversubscribed by hyperscalers. Small-scale miners and DePIN node operators will face extended lead times and inflated prices. During the 2022 bear market, I ran the Resilience Hub mentoring program, where we documented how 60% of small GPU mining operations folded because they couldn’t compete with institutional buyers. The same dynamic is replaying—except now the buyers are AI labs, not just crypto miners. For projects like Akash, which relies on consumer-grade GPUs, the bottleneck is less acute, but competitive pricing for high-end cards (H100, B200) will remain unfavorable.
2. Energy costs will become a geopolitical chessboard. Data center power consumption is projected to double by 2026, driven by AI training. Regions with cheap, stable energy (Iceland, Quebec, Texas) will see rent surges, squeezing both Bitcoin miners and PoS validators that rely on uptime guarantees. In my 2024 ETF transparency advocacy campaign, we mapped how institutional crypto adoption in Asia was already backtracking because energy allocation was being prioritized for AI cold storage. The invisible subsidy that blockchain infrastructure enjoyed—uncontested access to industrial power—is evaporating.
3. Capital allocation reshapes competitive moats. Seseri’s point about “reshaping resource allocation” is critical. When tech giants pour billions into AI infrastructure, they aren’t just building data centers—they are buying options on future compute. For decentralized networks, this creates an asymmetry: centralized cloud providers can offer AI-inference-as-a-service at near-zero margin while cross-subsidizing their crypto offerings (e.g., AWS’s Managed Blockchain). Meanwhile, DAO treasuries face a harder choice: buy GPUs for their own rollups or rent from hyperscalers at 20% premiums? Governance isn't a feature; it's a social contract. And that contract is being torn when token holders must vote between fiscal prudence and technical sovereignty.
But there is a contrarian angle I must explore.
Contrarian Angle
The conventional narrative—that AI capex will crowd out crypto—misses a second-order effect: the AI infrastructure buildout may actually accelerate demand for decentralized verification, data privacy, and agent coordination. Let me explain.
First, AI training is increasingly subject to regulatory scrutiny. The EU AI Act, China’s generative AI regulations, and emerging US state-level frameworks all require audit trails for model training data. Public blockchains offer immutable provenance. In 2026, I facilitated the Autonomous Agent Accountability Charter, where we concluded that on-chain attestation of training datasets is the only scalable way to prove compliance. This is a use case that only exists because of the AI capex surge—and it directly boosts demand for L1s with cheap storage (like Celestia or EigenDA) and zk-proof markets.
Second, AI agents are entering on-chain economies. As models like GPT-5 gain ability to execute smart contract transactions, the need for decentralized identity, reputation systems, and dispute resolution grows. The “multi-year capex cycle” ensures that AI agents will have access to abundant, cheap inference compute—making them economically viable to interact with DeFi protocols or DAOs. I’ve seen this firsthand: during the 2026 workshops, we simulated a DAO where AI agents managed treasury allocation. The results were terrifying but inevitable. Decentralized governance becomes the only firewall against agentic misalignment.
Third, the GPU arms race incentivizes alternative chip architectures. When NVIDIA’s monopoly forces prices up, capital flows to competitors: AMD, Intel, and—crucially—open-source chip designs like RISC-V. Crypto-native ASIC manufacturers (e.g., MinerVa) are already exploring AI acceleration. If the capex cycle runs long enough, it could fund a decentralized hardware ecosystem that breaks free from hyperscaler control. “Code doesn’t govern people; people govern people,” I often say. In this case, the people who design the silicon are the ultimate governors.
But the vulnerability I must own is that these second-order effects take time—3 to 5 years. In the immediate term, 2024–2025, the resource squeeze will dominate. I’ve personally spoken to three L2 teams that delayed their zk-rollup deployments because they couldn’t secure GPU capacity for proof generation. The bear market taught me that survival matters more than gains. Right now, the AI capex tsunami is a survival test for any blockchain project that touches real-world hardware.
Takeaway
The Seseri thesis is not wrong—it is incomplete. Yes, we are entering a multi-year capital expenditure cycle that will reshape tech giants’ allocations. But the blockchain industry’s response should not be to retreat or compete head-on. Instead, we must double down on the areas where decentralization provides unique value: audit trails, agent governance, and alternative hardware. The 2022 bear market was a filter; this AI capex cycle is a crucible. The projects that survive will be those that embed themselves as the ethical compute layer for AI—not as a rival to it.
I see a future, perhaps by 2027, where every major hyperscaler runs a public blockchain-based attestation service for their AI models, and where decentralized GPU networks like Akash become the default overflow capacity for inference spikes. But that future requires us to stop framing AI and crypto as competitors. “Code is law, but people are the protocol.” The people building the AI infrastructure are also the people who can embed blockchain values into it—if we show them why it matters.
Now, go read your node’s power purchase agreement. Check if your staking provider has diversified hardware supply. And ask your DAO treasury committee: are we prepared for the Seseri Floor?