Over the past week, a single data point has dominated the AI infrastructure narrative: Anthropic has signed 70-80 letters of intent for data center capacity. Most coverage treats this as a bullish signal for the entire AI stack—a validation of demand, a precursor to massive GPU purchases, and a green light for crypto AI tokens like Render and Akash. I don’t think the market is pricing this correctly.
Let’s start with context. The AI infrastructure cycle has followed a predictable pattern: hype around model capabilities → realization of compute bottlenecks → capital deployment into data centers. In 2023, the narrative was “training compute is the moat.” In 2024, it shifted to “inference is the new battlefield.” Now, in 2025, the narrative is “scale or die,” and Anthropic is the latest player to signal intent. But intent is not the same as execution.
A letter of intent is a non-binding expression of interest. It’s a negotiation tool, not a purchase order. In my consulting work with AI infrastructure projects, I’ve seen LOIs used primarily to signal market demand to investors and to secure better terms from operators. The conversion rate from LOI to signed lease is typically 30-50% for established players, and lower for startups. Anthropic, despite its $20B+ valuation, has yet to turn a profit. Its revenue model is still dependent on API subscriptions from a niche enterprise base. The 70-80 LOIs represent a bet on future revenue, not a reflection of current demand.
The market is treating this as a signal of technical readiness. It’s not. From a technical standpoint, deploying 70-80 data centers—each potentially 10-20MW—implies a total capacity of 700-1600MW. That’s the equivalent of 1-2 massive hyperscale facilities or 10-20 mid-sized ones. For reference, OpenAI’s existing infrastructure (including Azure leases) is estimated at ~300MW. Anthropic is essentially planning to leapfrog its current capacity by 3-5x. That’s ambitious, but it’s also a huge risk. The timeline for building out that capacity is 18-24 months, during which chip supply, regulatory approvals, and power grid constraints could all cause delays.
The hidden narrative here is about capital allocation, not compute. Anthropic is burning cash at an estimated $2-3B per year. To fund this buildout, they will likely need debt financing or a massive equity round. The LOIs serve as collateral for debt—a way to convince lenders that there is committed demand. But if the market overheats and interest rates remain high, the cost of servicing that debt could crush margins. I’ve seen this pattern before in 2021 with DeFi protocols that over-leveraged on liquidity mining incentives. The structure was different, but the outcome was the same: when the yield faded, the capital fled.
Now, let’s connect this to crypto AI narratives. The crypto AI sector—tokens like Render (RNDR), Akash (AKT), io.net, and others—has been riding the coattails of AI infrastructure hype. The thesis is that decentralized GPU networks will capture overflow demand from centralized providers like Anthropic, OpenAI, and Google. But the LOI data suggests the opposite: Anthropic is building its own capacity, not renting from third parties. That’s a bearish signal for decentralized compute providers. The narrative that “decentralized networks will benefit from the AI boom” assumes that centralized players will be capacity-constrained. If they build their own, the overflow narrative weakens.
The contrarian angle: The LOI might be a PR move, not a supply signal. Anthropic is reportedly preparing for a Series E or F round at a $30-40B valuation. Leaking the LOI count to a crypto-focused outlet like Crypto Briefing is a classic narrative engineering tactic. It creates FOMO among institutional investors—they see a company that is “locked in” for growth, so they rush to invest. The same playbook was used by Nvidia in 2023 when it leaked its H100 allocations. The difference is that Nvidia had actual backlogged orders; Anthropic has intent papers. The market is conflating the two.
What does this mean for the crypto AI narrative? I believe the market is overpricing the near-term demand for decentralized GPU compute. The actual bottleneck is not total compute capacity, but the ability to deploy it efficiently. Anthropic’s LOIs don’t guarantee that their data centers will be filled with GPUs—they could be underutilized if model demand doesn’t materialize. For crypto AI tokens, the real opportunity lies in serving the “long tail” of AI startups that cannot afford Anthropic’s enterprise pricing, not in competing for hyperscale contracts.
Based on my experience auditing tokenomics for AI infrastructure projects, I’ve seen a recurring pattern: projects overestimate total addressable market by assuming centralized players will outsource. They won’t. The history of cloud computing shows that the biggest players (AWS, Azure, GCP) built their own data centers and only used third-party providers for peak demand. The same will happen with AI compute. Anthropic’s LOIs are a signal that they are following the playbook, not breaking it.
The takeaway: Watch for the conversion, not the intent. Over the next 6 months, the key signal will be whether Anthropic converts these LOIs into binding lease agreements. If they do, it validates the demand narrative. If they don’t, it reveals a strategic bluff. For crypto AI investors, the smarter play is to focus on projects that provide complementary services—like GPU software orchestration or data center cooling tokenization—rather than betting on direct competition with hyperscalers.
Narrative liquidity is more important than technical liquidity right now. The market is hungry for a story that justifies the next leg up. Anthropic’s LOIs provide that story, but the underlying substance is thinner than it appears. I don’t think the market is pricing the execution risk correctly. When the next quarterly report shows more cash burn than revenue, the narrative will shift from “scaling” to “sustainability.” And the crypto AI tokens that hitched their wagon to the wrong narrative will be left holding the bag.