Code over hype.
Seventy to eighty letters of intent. That’s the number allegedly floating around Anthropic’s procurement desk, according to a report from Crypto Briefing—a source whose credibility in AI infrastructure is as questionable as a Bitcoin whitepaper printed on a napkin. But let’s assume the number is real. What does it mean when a company that hasn’t turned a profit locks up potential gigawatts of compute capacity? It means either they are building the next frontier of intelligence, or they are laying the groundwork for one of the most spectacular capital burn rates in tech history.
I’ve spent years inside the blockchain infrastructure trenches—watching protocols scale from hobbyist nodes to institutional-grade validator networks. The pattern is eerily similar: when a project starts signing multiple LOIs for colocation space, it’s not about innovation. It’s about survival. And Anthropic is fighting for survival against the two-headed hydra of OpenAI and Google.
Context: The Infrastructure Arms Race
Anthropic, the safety-focused AI lab behind Claude, has long leased compute from AWS and GCP. But leasing is a variable cost. Signing 70-80 LOIs for data center capacity is a fixed-cost commitment. It signals a shift from “we’ll scale as we grow” to “we’re betting the farm on exponential demand.” The LOIs likely cover multiple geographies, multiple operators (Equinix, Digital Realty, CyrusOne), and multiple power levels—from 10 MW to 100 MW per site. If even half materialize, we’re talking about 350-800 MW of new capacity. That’s enough to power a small city. Or, in AI terms, enough to train a few more generations of models that may or may not align with human interests.
Core: The Centralization Paradox
Here’s where my blockchain reflexes kick in. Anthropic’s entire pitch is “constitutional AI”—alignment, safety, ethical governance. But building a centralized compute empire with 70-80 data centers is the opposite of decentralized resilience. It’s putting all the eggs in a few physical baskets. If one of those centers goes down due to a power outage, a natural disaster, or a cyberattack, Claude’s brain goes dark. The irony is thick: a company that lectures the world on AI safety is creating a single point of failure for its own intelligence.
Truth decays slowly. But the market doesn’t care about irony. It cares about moats. And these LOIs are a moat—a massive, concrete, energy-hungry moat. They signal to investors that Anthropic can handle enterprise SLAs, that it can offer private deployments for regulated industries, that it can compete on latency. But they also signal something else: a desperate need to lock in supply before the GPU shortage tightens again. Every LOI is a hedge against NVIDIA’s allocation whims.
Let’s look at the numbers. Assume each LOI averages 15 MW. That’s 1,050 MW total. At 6.5 MW per 1,000 NVIDIA H100 GPUs (a rough estimate), that’s about 161,538 GPUs. For context, Meta is buying 350,000 H100s this year alone. Anthropic’s bet is large, but not insane. However, the LOI phase is negotiation theater. Only 30-50% of LOIs typically convert to signed leases. If Anthropic signs only 30, they get 450 MW—still a lot, but less intimidating. The number 70-80 is likely a PR tool to inflate perceived demand, create FOMO among investors, and pressure data center operators to lower prices.
Contrarian: What If the LOIs Are a Smoke Screen?
Here’s the contrarian angle: these LOIs might be a sign of weakness, not strength. Anthropic is burning cash at an estimated $2 billion per year. Its revenue—while growing—is nowhere near covering that. Locking into long-term leases for capacity that may not be fully utilized is a recipe for a write-down. Remember the 2022 crypto winter? Cosmos, Solana, and others signed 100 MW leases during the bull run, only to sublease at a loss when the market collapsed. Anthropic could be repeating the same mistake.
Hold the line. But also evaluate the counter-argument: maybe the LOIs are structured as options, not firm commitments. LOIs are non-binding. They give Anthropic the right to lease capacity later, not an obligation. If the AI market slows, they can walk away with a small penalty. This is smart treasury management, not reckless spending. The real test will be when the first 10-20 LOIs convert to signed contracts. Watch for SEC filings or partnership announcements.
Takeaway: Compute Centralization vs. Decentralized Alternatives
What does this mean for the blockchain world? It means the war for AI compute is being fought with centralized weapons. Anthropic, OpenAI, and Google are building digital fortresses. But the decentralized compute movement—Render Network, Akash, Golem—is still alive, albeit small. If Anthropic’s 70-80 LOIs lead to a capacity glut in 2027, decentralized providers might pick up the leftover scraps. Or, if the centralized giants squeeze the supply, decentralized compute could become a premium, not a discount.
Build anyway. Whether these LOIs are a brilliant strategic move or a desperate gamble, one thing is clear: the infrastructure bottleneck is not going away. The next 36 months will determine whether AI remains a centralized oligopoly or finds a path to distributed sovereignty. I’m watching the LOI conversion rate, not the headline number. That’s where the truth lives.
