Anthropic's 70 LOIs: Scaling Promise or PR Spectacle? A Forensic Teardown

Seventy to eighty letters of intent. That's the number floating through the rumor mill—Anthropic's alleged lock on future data center capacity. The report from Crypto Briefing, a source more accustomed to token launches than AI infrastructure, dropped this figure without context, without verification. In a market where whitepapers promise moonshots and GitHub repos deliver empty directories, I've learned to read between the lines. These LOIs are not a balance sheet. They are a bet. Context: Anthropic, the self-proclaimed "safety-first" AI lab, is racing to catch up with OpenAI and Google. Its Claude models have gained traction, but compute is the new oil, and Anthropic is drilling. The report claims the company signed 70-80 LOIs for data center capacity. What does that mean? Let's dissect the claim, the numbers, and the unspoken risks. Core: First, the LOI game. A letter of intent is a non-binding document. It signals interest but not commitment. Industry averages suggest a 30–50% conversion rate to actual leases. That means 70 LOIs might become 25–35 operational data centers. Still significant, but not the slam dunk implied. The architecture of trust, engineered for failure, begins with such ambiguity. Second, the scale. Assume each LOI covers 10–20 MW. Total capacity: 700–1,600 MW. That's 1–2 GW. Compare to OpenAI's reported plans for 10 GW. Anthropic's revenue is a fraction of OpenAI's. This is a massive bet, a doubling down on a future that may not materialize. The gap between intent and execution is measured in gigawatts. Third, the chip problem. Data centers are empty shells without GPUs. NVIDIA's lead times are long. AMD's supply is constrained. Anthropic may need to secure chips separately, possibly through long-term purchase agreements. Are these LOIs contingent on GPU availability? The report doesn't say. Based on my experience auditing supply chains in crypto, I've seen how a single missing component can collapse a project. The same applies here. Fourth, the power problem. 1 GW of new load requires enormous power infrastructure. Grid constraints, renewable credits, nuclear deals. Anthropic's ESG promises are likely strained. The company has positioned itself as responsible, but massive data center builds are carbon-intensive. The contradiction is real. Fifth, the financials. CapEx of this magnitude means billions of dollars. Anthropic's funding rounds are large but not infinite. Its latest raise was around $7.5 billion. Building out 1 GW could cost $5–10 billion over several years. Burn rate will skyrocket. If revenue doesn't scale, cash runway shrinks. The architecture of trust, engineered for failure, becomes a timeline to insolvency. Sixth, the PR angle. Why leak to Crypto Briefing? Probably to signal to investors and partners that Anthropic is serious. But also to create FOMO. Classic playbook: announce intent, drive valuation, then scramble to deliver. I've seen this in crypto projects that promised Layer 2 rollups but delivered nothing. The pattern is the same. Contrarian: But let's give credit where due. The bulls are right that Anthropic needs compute. The LOIs indicate a shift from renting to owning, which reduces long-term costs. If executed well, Anthropic could build a moat. The question is execution. The architecture of trust, engineered for failure—if they fail to deliver, the trust evaporates. The contrarian angle: perhaps the LOIs are real, and Anthropic is making a strategic bet that will pay off. But the evidence is thin. The sources are unnamed. The numbers are unverified. Takeaway: The market should treat these LOIs as a hypothesis, not a conclusion. Track the conversion rate. Watch for official announcements with specific partners. Until then, the architecture of trust is built on letters of intent—and letters are cheap. The gap between intent and execution is measured in gigawatts, and so far, Anthropic hasn't turned on the power.