Resolved. I Standardized Every Reference to the Same Benchmark: The U.S. 10-year Treasury Yield at Approximately 5.2%.

“The AI race has moved from securing demand to delivering capacity on time.”

AI demand is no longer scarce. Deliverable capacity is. The AI race has moved from securing demand to delivering capacity on time, and with the U.S. 10-year Treasury yield around 5.2%, delay directly destroys returns.

The pattern is clear across the market. Anthropic has committed $11.6 billion to Akamai cloud capacity. Oracle invoked force majeure at Project Jupiter because power may arrive late. BlackRock and IFM are reportedly considering up to $25 billion for Stack Infrastructure's Asia-Pacific data centers. Vertiv is expanding further into liquid cooling.

The fundamental question now: who can convert contracted AI demand into productive capacity before power, permitting and financing consume the economics?

1. 01 Anthropic — Compute Becomes a Strategic Input

   Anthropic signed a seven-year, $11.6 billion Akamai agreement, potentially expandable by another $9 billion. The point is not simply more compute — it is diversification across infrastructure providers. Compute is becoming a strategic input secured years in advance.

   Anthropic gains optionality, but also long-duration commitments. The test is whether enterprise AI revenue grows faster than those obligations. signal: Compute is now a strategic input secured years ahead — if revenue outpaces the commitment.

2. 02 Oracle — Contracts Without Power Are Not Capacity

   Oracle invoked force majeure after power delays threatened Project Jupiter. Customers exist. Financing exists. Construction capital exists. Power does not arrive on schedule. The equation is: Contracted Capacity × Delivery Probability = Economic Capacity.

   With the U.S. 10-year Treasury yield around 5.2%, a one-year delay means more financing cost, later revenue and faster hardware depreciation. Time-to-power is time-to-cash. signal: Contracted capacity is worthless without delivery probability.

3. 03 Akamai — Distributed Inference Becomes Infrastructure

   Akamai receives today's largest positive GVI revision. Its Anthropic contract could transform a traditional internet-infrastructure company into an AI compute platform. The architecture is shifting from centralized training to centralized training + distributed inference.

   Akamai already owns global distribution. AI can turn that footprint into a strategic compute asset. signal: Distribution footprint becomes an AI compute asset as inference moves to the edge.

Five Investor Takeaways

deliverability as the moat

• Buy powered capacity. A gigawatt without electricity and commissioning is not an economic asset. Favor Schneider Electric · Siemens Energy · Vertiv · Quanta.
• Time-to-power is a financial KPI. Every delay raises financing cost and reduces hardware's useful economic life — execution speed determines ROIC.
• Contract quality beats capex scale. Prepaid, long-duration demand deserves a premium over speculative capacity. Watch Microsoft · Anthropic · Akamai.
• Distributed inference is the next layer. Inference moves closer to users, enterprises and agents; Akamai is today's clearest signal.
• At a ~5.2% U.S. 10-year Treasury yield, backlog must become cash faster. High rates punish idle assets and permitting delays — capital productivity is becoming the real AI moat.