Hook
Tracing the code back to its chaotic genesis, I've spent years watching capital flows distort the very fabric of decentralization. But nothing prepared me for this: Canada Pension Plan (CPP) Investments—a $600 billion behemoth—just dropped $1.75 billion into EQT’s AI infrastructure strategy. That’s real money. Not a token sale or a VC-led seed round. A pension fund. Committed to building centralized data centers for AI training. The logic is impeccable on paper—AI demand is surging, compute is scarce, and long-term contracts offer stable returns. But as someone who has spent the last decade arguing that decentralization is a moral imperative, I can't help but feel a cold shiver. This isn't just an investment; it's a bet that the future of AI compute will remain locked inside privileged, permissioned walls.
Context
The AI compute landscape today is a paradox. On one hand, we have a Cambrian explosion of models—GPTs, LLaMAs, Mistrals—each demanding more GPU cycles than the last. On the other hand, the hardware to run them is bottlenecked by supply chains, geopolitics, and—most critically—capital. The market has responded with a frenzy of data center announcements: Blackstone, KKR, DigitalBridge, and now EQT. These are not your grandfather's real estate assets. These are hyper-specialized facilities designed to house tens of thousands of NVIDIA H100 or B200 GPUs, each requiring 700W of power, liquid cooling, and InfiniBand networking. The cost per megawatt (MW) of IT load has ballooned to $8-10 million. CPP’s $1.75 billion will likely build around 2 GW of new capacity—enough to host roughly 50,000 to 75,000 GPUs. But here's the kicker: that capacity is one hundred percent centralized. It will be owned by EQT, operated by EQT, and leased to a handful of hyperscalers or AI labs. No community governance, no token incentives, no permissionless access.
This is where the blockchain ethos collides with economic reality. We’ve been preaching that decentralized compute networks—like Akash, Render, or io.net—will democratize access to GPUs. But the capital being deployed here is orders of magnitude larger than the entire market cap of every decentralized compute token combined. As of today, Akash Network’s total stake is roughly $150 million. Render’s market cap is around $1.5 billion. Together, they are barely a rounding error to CPP’s single check. The message is clear: institutional capital wants centralized, not decentralized, infrastructure.
Core: Tech + Values Analysis
Where logic meets the absurdity of market hype, we have to dissect the technical assumptions behind this investment. EQT’s strategy is based on a simple thesis: AI training and inference will continue to require massive, contiguous clusters of high-bandwidth GPUs. This is true for today’s Transformer-based models. But the blockchain community knows a thing or two about scaling constraints. We’ve seen how sharding, rollups, and sidechains emerged to solve Ethereum’s congestion. AI compute will face a similar evolution. Already, researchers are exploring mixture-of-experts (MoE) architectures that activate only a fraction of parameters per token, reducing the need for huge unified clusters. Edge inference is becoming more viable. And then there is the possibility of algorithmic breakthroughs—state space models (SSMs) or even photonic computing—that could radically reduce the energy and hardware requirements.
But the core problem is more philosophical. By pouring $1.75 billion into fixed, centralized assets, CPP and EQT are locking themselves into a specific technological path. Data centers have a lifespan of 15-20 years. By year 10, the GPUs inside will be obsolete, but the building—the concrete, the power grid, the cooling towers—will remain. This creates a massive incentive to keep using that legacy hardware, slowing down innovation. Decentralized compute networks, on the other hand, are inherently more flexible. They can aggregate heterogeneous hardware from thousands of operators, adapt to new algorithms through software upgrades, and redistribute resources dynamically. Yet they lack the scale and reliability that institutional capital demands.
I recall auditing governance proposals on Uniswap and Aave in 2020, watching whales and VCs manipulate votes with 5% turnout. The same pattern repeats here: centralized decision-making by a handful of fund managers determines the future of compute for billions of users. The irony is thick. The blockchain ethos promised to replace trust in institutions with trust in code. But here, the code is still being written by NVIDIA and the infrastructure is being built by private equity.
Contrarian: Pragmatism Test
An evangelist who doubts his own gospel—let me play devil’s advocate. Maybe this investment is exactly what the crypto world needs. The current shortage of high-end GPUs is squeezing not just AI companies but also blockchain projects that rely on verifiable compute. zk-Rollups, for example, require significant proving power. If centralized data centers flood the market with new capacity, the price of GPU compute could drop, making it cheaper for decentralized networks to rent idle capacity. EQT’s data centers might even become customers of Akash or Render—if their own utilization falls below 80% and they need to resell spare cycles. But this is a big “if.” Centralized providers have no incentive to cede control to a permissionless protocol. They will likely build their own resale marketplaces, kept within the walled garden.
Furthermore, the sheer scale of EQT’s operation could attract regulatory scrutiny. Once a pension fund owns a critical piece of AI infrastructure, governments will demand compliance: data sovereignty, censorship of certain models, anti-money laundering checks. That’s a poison pill for permissionless innovation. We saw it happen with centralized exchanges; we’ll see it happen with centralized compute. The more institutional money pours into these walled gardens, the harder it becomes for a truly open alternative to emerge. So while the short-term effect might be lower GPU prices, the long-term effect is a harder regulatory lock-in. Is a few basis points in compute cost worth sacrificing the very principle of uncensorable computation?
Takeaway
In the silence between the block hashes, I wonder: will the future of AI be dictated by the collective will of a decentralized network—or by the quarterly returns of a few pension funds? The $1.75 billion bet says the latter. But blockchain’s entire history is a series of defied odds. The code is still being written. The game is not over. But the infrastructure race has begun, and we are losing ground. The question is: can we build a decentralized compute network that attracts not just retail hype but real institutional confidence—or will we always be playing catch-up to the centralized giants? I know my answer. I’m just not sure the market will agree.
