The Power Allocation

U.S. bets $9B on SMRs for data centers, but $182/MWh cost estimates are 3x solar and 4x natural gas. Can scale economics change the math?

Show Notes

In this episode of The Power Allocation, we examine the economic reality of small modular reactors for data centers: current estimates put SMR electricity at around $182 per megawatt hour—roughly three times utility-scale solar and four times combined cycle natural gas.

The SMR value proposition depends on costs falling dramatically through factory fabrication and standardized designs. But data centers may tolerate premium pricing when the alternative is no power at all.

Key topics covered:

  • Why first-of-a-kind reactors carry full design, regulatory, and supply chain development costs
  • The nuclear industry's credibility problem after Vogtle cost overruns
  • How hyperscalers' long-term contracts can average costs over decades
  • The coordination problem: everyone waiting for someone else to prove costs come down
  • Why $182/MWh looks reasonable when the alternative is 5-year grid interconnection waits

About The Power Allocation: Brought to you by Spring Street Management Group, translating data center and energy hype into real infrastructure and assets on the daily.

Keywords: SMR costs, small modular reactor economics, nuclear power costs, LCOE, data center nuclear, Vogtle, factory fabrication, nuclear economics, energy costs, power purchase agreement, hyperscaler nuclear

]]>

What is The Power Allocation?

The AI boom isn't constrained by chips, algorithms, or talent. It's constrained by electricity.

The Power Allocation is a daily briefing on AI infrastructure — where capital is actually being deployed. Each episode cuts through the hype to examine the physical realities shaping the AI buildout: power constraints, grid interconnection, land acquisition, data center financing, cooling infrastructure, and utility relationships.

This isn't a software podcast. This is an infrastructure podcast.

Who it's for: Institutional investors, infrastructure allocators, data center developers, utilities, family offices, and anyone positioning capital for the physical layer of artificial intelligence.

What you'll learn:

Why power availability — not GPU supply — is the binding constraint on AI compute
How hyperscalers are locking in multi-decade power purchase agreements
Where data centers are relocating and why grid geography is reshaping the industry
The financing structures turning compute facilities into bond-like assets
What execution timelines, permitting delays, and interconnection queues mean for capital deployment
Format: 3-6 minute episodes. Dense. Clear. No hype.

New episodes daily. Subscribe wherever you listen.