Odd Lots

The Utilities Analyst Who Says the Data Center Demand Story Doesn't Add Up

February 2, 2026

Key Takeaways Copied to clipboard!

  • The utilities analyst, Andy DeVries, argues that utilities are already committed to connecting nearly twice the amount of power capacity (110 GW) needed to meet third-party forecasts for data center demand (50 GW needed by 2030). 
  • The current exuberance in the utilities sector, driven by AI data center demand, is creating a potential overbuild scenario where the supply of committed capacity significantly outpaces projected demand by the end of the decade. 
  • Risk is emerging in the credit markets as private credit lenders rush to finance data centers, mirroring past cycles where increased competition leads to diminishing documentation and protections for lenders, especially for second-tier data center operators. 

Segments

Utilities Analyst Career Shift
Copied to clipboard!
(00:01:00)
  • Key Takeaway: Utilities analysis has shifted from focusing on bond-like yields to being central in secular growth discussions due to AI-driven energy demand.
  • Summary: For years, utilities analysts focused primarily on yield relative to Treasuries, viewing the sector as bond-like. The rise of AI buildout has made utilities analysts highly sought after due to energy constraint concerns. This shift represents old, stable industries becoming secular growth drivers.
Guest Introduction and History
Copied to clipboard!
(00:03:32)
  • Key Takeaway: Andy DeVries has 25 years of experience as a utilities analyst, having covered major industry events like the Enron bankruptcy and the TXU LBO.
  • Summary: Andy DeVries is the co-head of investment grade credit and head of utilities and power at CreditSights. He has covered the sector for 25 years, witnessing major events including the Enron bankruptcy and the Calpine private equity return. Pre-data center analysis involved tracking rate cases, local legislation, and natural gas prices.
Data Center Capacity Math
Copied to clipboard!
(00:08:50)
  • Key Takeaway: Utilities are tracking 110 GW of committed data center capacity, which is more than double the 50 GW increase forecast by third parties by 2030.
  • Summary: Current data center consumption is 45 GW, with third-party estimates projecting demand to reach 95 GW by 2030 (a 50 GW increase). Utilities are tracking 140 GW of near-term supply, which, after Power Usage Effectiveness (PUE) adjustment, equates to 110 GW of connected capacity. This committed supply is nearly double the expected demand increase by 2030.
Texas Market Dynamics
Copied to clipboard!
(00:13:37)
  • Key Takeaway: The Texas power market forward curve does not reflect the high demand growth implied by utility CFOs, suggesting a pricing disconnect.
  • Summary: Texas is a walled-off peak market of 87 GW, with demand estimates suggesting an addition of 15-30 GW by 2030. Despite this, the forward power curve is only slightly upward sloping, implying traders do not expect significant long-term price increases. Big tech is paying high prices, such as $95/MWh, for power contracts in Texas.
Oversupply and Ratepayer Risk
Copied to clipboard!
(00:20:07)
  • Key Takeaway: If data center demand fails to materialize as expected, ratepayers face the risk of absorbing the costs of utility overbuilding, though some utilities are implementing ratepayer protection mechanisms.
  • Summary: The core argument is that utilities are committed to building out capacity that is twice what is forecast to be needed by 2030, making demand forecasts the key variable. Utilities like NYSOR’s NIPSCO are setting a ‘gold standard’ by having Amazon’s Genco kick back $1 billion to ratepayers to avoid rate debates. If demand doesn’t materialize, someone must pay for the excess infrastructure, potentially falling on customers.
Construction Lags and Inflation
Copied to clipboard!
(00:25:19)
  • Key Takeaway: New power generation projects face significantly longer construction timelines (5+ years) and higher costs (up to $3,000/KW for gas plants) compared to data centers (2-3 years).
  • Summary: Data center construction timelines are shorter (2-3 years) than new power plant construction (5+ years), meaning generation buildout may lag behind data center connections. The cost to build a combined cycle gas plant has risen from $1,200/KW to $3,000/KW due to inflation. Data center costs ($40,000/KW) are so high that the cost of securing fuel sources is negligible to big tech.
Private Credit and Circular Financing
Copied to clipboard!
(00:32:16)
  • Key Takeaway: The success of PIMCO’s highly profitable Meta data center loan has spurred increased, potentially risky, lending activity in private credit for the sector.
  • Summary: PIMCO reportedly made $2 billion on day one from a $25 billion loan to a Meta data center in Louisiana, trading at a much tighter spread than initially priced. This success is driving other private credit lenders into the space, risking covenant erosion as competition increases. Circular financing, where tech firms buy equity in their suppliers (e.g., NVIDIA buying CoreWeave equity), raises skepticism similar to past vendor financing issues.
Nuclear Power Outlook
Copied to clipboard!
(00:37:36)
  • Key Takeaway: New large-scale nuclear projects like Vogtle are unlikely to be repeated without big tech equity investment, suggesting Small Modular Reactors (SMRs) are the only viable path forward.
  • Summary: The Vogtle plant’s massive cost overrun ($14B to $32B) makes utilities hesitant to undertake similar large nuclear projects. SMR deployment is contingent on big tech companies agreeing to purchase SMRs and invest equity in the manufacturers. The US Navy’s long history of building nuclear submarines suggests SMR technology is feasible if backed by tech capital.