The Velocity Edge
Morning Brief · 08:00 · Morning View

The Cost of Acceleration

AI demand is accelerating, but the cost of financing that acceleration is accelerating with it — and the next Great Split is between systems that convert scale into cash flow and those that consume capital faster than they monetize intelligence.

Francesco De Leo Kaufmann September 24, 2026 4 min read Capital
5.11%U.S. 10-year Treasury yield
9.75%Top coupon SoftBank is paying on AI-linked debt
$11.1BSize of SoftBank's bond issue
$10B+Microsoft's planned Gulf AI infrastructure commitment

“The first phase rewarded speed. The next will reward profitable speed.”

This morning opens with a five-number collision: 5.11%. 9.75%. $11.1 billion. $10 billion-plus. 2.8 million downloads in two weeks.

Those five numbers frame the market. Put differently: AI demand is accelerating, but the cost of financing that acceleration is accelerating with it. The fundamental question is now unavoidable — can AI returns rise faster than the cost of the capital, energy and infrastructure required to produce them?

The next Great Split is increasingly between AI systems that convert scale into cash flow and those that consume capital faster than they monetize intelligence.

01

Five Signals

the forces reshaping the cycle

  1. The U.S. 10-year Treasury is above 5%, while Japanese yields have reached multi-decade highs. Expensive money is becoming a structural constraint, and capital productivity is becoming as important as technological leadership. signal: Expensive money is now a structural constraint.

  2. SoftBank raised $11.1 billion in bonds, with dollar coupons as high as 9.75%, largely to finance its OpenAI exposure. Frontier AI has moved decisively into public credit markets — and the technology must generate returns fast enough to outrun the coupon. signal: Frontier AI now lives in public credit markets.

  3. Meta's Muse reached 2.8 million downloads in two weeks, while Charm pushes AI toward an always-on interface. The interaction model is shifting from chat to delegation. The next demand metric is not users — it is tasks × duration × agents. signal: The model shifts from chat to delegation.

  4. Microsoft plans to invest more than $10 billion in Gulf cloud and AI infrastructure through 2030, alongside major connectivity investment. AI infrastructure is clustering where energy, capital, compute, connectivity and resilience can be assembled together. The winning geography may be the one capable of supplying the complete system. signal: Winners will supply the complete system.

  5. The Trump–Xi meeting takes place today after Washington and Beijing extended their trade truce and explored an AI incident-notification mechanism. The emerging framework is neither full decoupling nor détente — compete, contain, communicate, continue trading. Predictable boundaries around competition may become more valuable than temporary tariff concessions. signal: Predictable boundaries beat temporary concessions.

02

The Collision Of Two Curves

return on intelligence vs. cost of acceleration

Five Investor Takeaways

positioning for a 5% world

  • Favor scarcity over generic AI exposure. Accelerated computing, memory, networking, optics, power and grid infrastructure retain structural pricing power.
  • Watch credit as closely as equities. Financing costs are becoming a leading indicator of how far the Supercycle can scale.
  • Separate demand growth from return on capital. Explosive usage does not guarantee attractive economics if debt, power and infrastructure costs rise faster than monetization.
  • Increase attention to agentic infrastructure. Persistent agents could multiply inference demand far faster than user growth alone suggests.
  • Own the businesses that can compound through a 5% world. The strongest positions combine scarcity assets, contracted demand, pricing power, high utilization and low relative funding costs.

Francesco De Leo Kaufmann · The Velocity Edge

More from The Velocity Edge

Jump straight into another note — no need to head back.

Keep reading

More field notes, from dawn to close.

Day signals and night briefings, published like a journal for board-level AI decisions.

Back to The Velocity Edge