The announcement hit my feed like a block confirmation: CPP Investments, managing over $600 billion in assets, committed $1.75 billion to EQT’s AI infrastructure strategy. On paper, it's another data center play. But for anyone who has spent years in the blockchain trenches, this is more than a financial transaction—it's a philosophical referendum on how compute power should be governed. The market is euphoric, FOMO is rising, but I see the ghost of a centralized future hiding beneath the headlines.
Context: AI Compute and the Crypto Crossroads
We are living through a compute arms race. Every major tech company—Microsoft, Google, Amazon—is pouring billions into data centers designed to train and deploy the next generation of large language models. The thesis is simple: AI will demand exponentially more GPU cycles, and those cycles need physical homes. EQT’s $1.75 billion will likely fund roughly 2,000 megawatts of new capacity, enough to house perhaps 50,000 to 75,000 H100 GPUs. That’s a lot of computing power.
But here’s the disconnect that keeps me up at night: the blockchain industry is simultaneously building its own compute layer, aiming to democratize access through tokenized marketplaces like Akash, Render, and Bittensor. These networks promise permissionless, verifiable, and decentralized compute. Yet the capital markets are overwhelmingly backing the centralized model. CPP’s move signals that long-term, risk-averse capital sees the highest confidence in traditional, walled-off data centers.
From my perspective, this is a direct challenge to the core belief that the future of computation should be sovereign, not corporately controlled. I remember sitting in a small New York apartment in 2020, writing the third essay of my “Soul of Code” series, arguing that smart contracts could democratize lending. Now I feel a similar urgency around compute itself.
Core: The Architecture of Trust
Let’s dive into the technical and economic implications. The $1.75 billion will likely be deployed into build-to-suit data centers, designed specifically for high-power-density AI workloads. In practice, that means liquid cooling, InfiniBand or 800G Ethernet networking, and massive electrical infrastructure. The construction timeline is 18 to 24 months, meaning the new capacity won’t come online until late 2026 or early 2027.
But here is the critical point for blockchain: these data centers are not neutral. They are designed to be leased by a single large tenant—likely a hyperscaler cloud provider or a specialist AI cloud like CoreWeave. That tenant controls access, pricing, and governance. If you are a decentralized AI project needing compute, you don’t get to plug into that capacity. You must go through the same centralized gatekeepers.
Based on my audit experience, I see a parallel to the ICO era. Back in 2017, I found a reentrancy bug in EtherTrust that could have drained $4.2 million. I chose to publish the vulnerability rather than profit from a bug bounty. That taught me that infrastructure ethics matter. Centralized compute infrastructure, like centralized smart contract control, introduces a single point of failure—not just technical, but political and economic.
Let’s quantify the concentration risk. If EQT’s new facilities house 70,000 H100 GPUs, and that capacity is leased to a single entity, then that entity effectively controls a significant fraction of total available AI training capacity. In the crypto world, we talk about validator centralization; this is compute centralization at a far larger scale.
The contrarian might say: “But the blockchain industry can rent from these data centers too!” And indeed, some decentralized compute networks do lease hardware from centralized providers. But that creates a dependency. The provider can change terms, throttle access, or comply with government demands. True decentralization requires control over the entire stack, from chip to cooling.
Contrarian: The Pragmatic Reality
Now, let me challenge my own narrative. Is $1.75 billion really a bet against decentralization? Perhaps it’s the exact opposite. The sheer volume of investment validates the immense demand for compute. If demand continues to grow, decentralized compute networks will have a larger addressable market. Moreover, these new data centers are built with scalability in mind. They could, in theory, host blockchain nodes or mine cryptocurrencies if the economic incentives align.
In fact, some of the largest Bitcoin mining operations already use custom-designed data centers with co-location models. The difference is that mining is a single-purpose workload (hashing), while AI training requires low-latency, high-bandwidth interconnects. But the infrastructure overhead is similar. A well-designed data center can host both AI and blockchain workloads, as long as the cooling and power are adequate.
I recall a conversation with a fellow educator during DeFi Summer in 2020. We debated whether Compound’s governance was truly decentralized. The conclusion was that true decentralization requires self-sovereignty at every layer. The same logic applies to compute. You cannot claim to be building a decentralized future if your compute layer is leased from a single landlord.
The blind spot in my contrarian view: capital efficiency. Building your own decentralized compute network is capital-intensive and slow. Renting from existing centralized infrastructure is faster and cheaper in the short term. Many blockchain projects will choose pragmatism over purity. That is human nature. But it’s also how we get the “soul in the machine” diluted.
Takeaway: A Vision Forward
So where does this leave us? The crypto community must decide: do we integrate with the centralized AI infrastructure boom, or do we compete? My experience in the Compound governance working group taught me that code is law only if the community enforces it. We need to build the financial and technical incentives to make decentralized compute not just feasible, but preferable.
Conscience over consensus. We cannot simply follow the market’s euphoria. We must audit every layer for single points of failure. Trust is earned, not mined—and the trust in centralized data centers is an unearned legacy from the internet era. DeFi must mature into DeCompute.
Here is my forward-looking judgment: within five years, the most valuable AI networks will be those running on decentralized infrastructure. The $1.75 billion from CPP is a wake-up call, not a death knell. It reveals the scale of opportunity. If blockchain can offer verifiable, neutral, permissionless compute at competitive prices, the capital will follow.
But we must act now while the concrete is still wet. Build the protocols, secure the energy partnerships, and remember that technology without ethics is just machinery. The soul in the machine is what separates a system from a prison.
