70 letters of intent. That's the number circulating in infrastructure circles, attributed to a single AI company—Anthropic. The claim, sourced from Crypto Briefing, suggests the Claude creator has signed 70 to 80 non-binding letters of intent for data center capacity. On paper, it's a massive bet on scaling. In practice, it's a textbook case of infrastructure wishful thinking. Letters of intent are not contracts. They are exploratory signals, often with conversion rates below 40%. Yet the narrative is already being spun: Anthropic is building a global inference network, challenging OpenAI and Google. But the data tells a different story. The original report lacks technical specifics, financial commitments, and verified sources. It's a PR play dressed as a strategic move. And in a bear market where capital discipline is king, such announcements demand forensic scrutiny, not applause.
Context: Anthropic is a high-profile AI startup, backed by Google and Salesforce, with a valuation exceeding $18 billion. Its flagship model, Claude, competes directly with GPT-4. To support enterprise clients and reduce latency, Anthropic has historically rented cloud capacity from AWS and GCP. But renting is expensive and lacks control. The alleged 70-80 LOIs represent a pivot to building its own infrastructure footprint. The number itself is striking: most hyperscalers operate fewer than 50 data centers globally. To claim 70-80 LOIs implies a capacity target of 1-2 GW—a scale that rivals the expansion plans of Microsoft and Amazon. Yet the source is a crypto-focused outlet, not a tech journal or a regulatory filing. The credibility gap is wide. Based on my experience auditing infrastructure commitments during the 2022 crypto collapse, I've seen how easily these letters can become liabilities. When Terraform Labs promised 50,000 validators, they delivered fewer than 200. The pattern repeats: hype first, delivery second.
Core: Systematic teardown reveals three critical flaws. First, the technical vacuum. The LOIs contain no information about chip architecture, model roadmap, or power efficiency. Infrastructure without technical specifics is a blank check. Without knowing whether these data centers will host NVIDIA H100s, AMD MI300s, or Anthropic’s rumored custom ASICs, the capacity estimate is meaningless. The 2024 FTX debacle taught us that promises of scale without technical audits are a red flag. Second, the economic reality. A single 100 MW data center costs approximately $1 billion to build and operate over five years. Scaling to 1 GW implies a $10 billion capital commitment. For a company that earned less than $500 million in revenue in 2024, this is a staggering leverage ratio. Past performance predicts future panic. The 2022 LUNA collapse showed that over-leveraged infrastructure bets can vaporize billions overnight. Anthropic’s burn rate would accelerate to $2-3 billion annually, far exceeding its current funding. Third, the regulatory blind spot. Data centers are subject to local zoning, environmental impact assessments, and energy grid constraints. The LOIs likely span multiple jurisdictions—US, Europe, Asia—each with different compliance requirements. Regulations are lagging, not absent. Anthropic’s “safety-first” branding will be tested when it faces carbon emission caps in Germany or data sovereignty laws in Singapore. The LOIs do not mention any regulatory risk mitigation.
Contrarian: The bulls have a point. Anthropic’s current model, Claude, requires significant compute for both training and inference. As enterprise adoption grows, a dedicated infrastructure layer could reduce latency and improve service reliability. The LOIs may be a genuine attempt to lock in capacity before competitors like OpenAI (backed by Microsoft) and Google (with its own TPU clusters) corner the market. Check the source code, not the hype. But the bulls are ignoring the execution risk. The LOIs are non-binding. They can be terminated without penalty. If Anthropic fails to secure funding or its revenue growth stalls, these letters become worthless. The 2024 due diligence I conducted on Fireblocks’ custody solution revealed a similar pattern: enthusiastic LOIs that led to zero binding contracts. The crypto market is littered with “announced” infrastructure that never materialized. The contrarian angle is that the LOIs are a signal of intent, not a guarantee of delivery. The real test will be the conversion rate over the next 12 months. If Anthropic converts only 30% of these LOIs, it will still be a significant infrastructure build-out. But if the market perceives the announcement as a desperate attempt to boost valuation before a funding round, the stock (if it existed) would tank.
Takeaway: These 70 letters will either be the foundation of a trillion-dollar empire or the footnotes of a cautionary tale. The difference lies in execution, not ambition. Liquidity vanishes; insolvency remains. For investors, the signal is not the LOIs but the company’s cash position and revenue growth. For regulators, the absence of environmental and compliance disclosures is a warning. For the crypto-AI ecosystem, this story highlights the fragility of infrastructure narratives. The next time you see a headline about “massive” capacity commitments, ask for the conversion rate, the capital stack, and the chip allocation. Otherwise, you’re buying hype, not hardware. The code does not lie—but the press releases often do.