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

Episode Summary

Executive Summary: The episode examines how the AI data-center buildout has transformed utilities from sleepy bond proxies into secular growth stories, but guest Andy DeVries argues markets are overestimating demand and underestimating how much capacity is already being built. He contends utilities are rushing into oversized commitments, while pricing, timing, and ratepayer protections remain the key risks.

Main Topics: Utilities have shifted from bond proxies to growth sectors (Priority: 5/5): The hosts frame utilities as formerly dull, yield-oriented stocks now in focus because AI/data centers are driving secular demand and faster earnings growth. Contrarian demand vs. supply math for data centers (Priority: 5/5): DeVries argues the market is overstating future power needs: third-party demand forecasts are rising, but utility firms are already lining up enough committed supply to potentially overshoot actual demand. Forward power curves and market pricing skepticism (Priority: 4/5): The discussion digs into how Texas and gas/power forward curves appear too flat to reflect the implied demand surge, suggesting traders are not pricing in the bullish narrative. Ratepayer protection and who pays for overbuilds (Priority: 5/5): A major theme is whether residential customers or utility shareholders bear the cost if data-center buildouts disappoint; some jurisdictions have explicit protections, others do not. Private credit and off-balance-sheet financing (Priority: 4/5): The episode covers how private credit is funding data centers, including the Meta/Pimco deal, and why some financing structures may reflect balance-sheet management as much as funding needs. Timing mismatch between data centers, generation, and policy (Priority: 4/5): Even if demand is real, the timing is uneven: data centers can be built in 2-3 years while power plants may take 5-7 years, and current renewable policy may create near-term oversupply before late-decade demand arrives. Nuclear and small modular reactors as a long-shot solution (Priority: 3/5): Conventional nuclear is portrayed as too costly and slow after Vogtle; the realistic path, if any, is big-tech-backed small modular reactors with equity and offtake support.

Key Arguments: Utilities are experiencing a secular re-rating because AI and data centers are creating growth that was not part of the old bond-proxy narrative. The demand forecasts for data centers are likely too aggressive: utility-connected supply is already approaching or exceeding estimated need by 2030/2035. Forward power prices in Texas and gas markets are not signaling the same level of demand acceleration that bulls describe. If utilities overbuild transmission, substations, and generation for data centers that never materialize, someone will absorb the costs—either ratepayers or shareholders. Strong ratepayer protections are crucial; where they exist, deals can include explicit payments back to customers, reducing political and financial risk. Private credit is moving into data-center financing because traditional lenders see opportunity, but spreads and structural protections may also signal caution. Timing matters more than the headline thesis: near-term renewable additions may pressure power prices before late-decade AI demand arrives. Large-scale nuclear remains unattractive under current economics; any new capacity likely requires tech-company equity participation and risk sharing.

Data Points: Current data-center power consumption: 45 gigawatts - Guest’s estimate of data centers’ present power demand 2030 third-party data-center demand estimate: 90-95 gigawatts - Range of outside forecasts for data-center power demand by 2030 2035 third-party data-center demand estimate: ~160 gigawatts - Rough outside forecast for demand by 2035 Utility near-term committed supply: 140 gigawatts - Guest says utilities have firm/committed/contracted pipeline at this level PUE-adjusted committed supply: ~110 gigawatts - Adjusted downward to apples-to-apples compare against compute-only demand estimates Utility growth rate historically: 4%-6% annually - Typical utilities earnings/growth rate before data-center surge Utility growth rate recently: 5%-7% annually - Recent range as data-center demand lifts outlooks Some utility growth forecasts: ~8% annually - Certain names now projected at this level due to AI/data-center demand Texas peak market size: 77-87 gigawatts - Two different Texas market references made in discussion Texas expected data-center add by 2030: ~30 gigawatts - CFO discussion cited by guest Existing grid headroom in Texas: Existing grid could handle +10 gigawatts and run most of the year - Guest relays Constellation CEO stress test estimate Existing grid utilization claim: All but 40-50 hours a year - Guest says current grid could support a large addition except for a few peak hours Cost to build combined-cycle gas plant 10 years ago: $1,100-$1,200 per kW - Historical reference for utility construction costs Cost to build combined-cycle gas plant later: ~$2,000 per kW - Intermediate cost level mentioned by guest Current cost to build combined-cycle gas plant: ~$3,000 per kW - Guest’s estimate of current construction cost inflation Data-center build cost: ~$40,000 per kW - Guest compares utility capex to the much larger capex of data centers themselves Texas round-the-clock power price: mid/high $50s per MWh - Guest’s estimate of market prices in Texas Big-tech contracted power price in Texas: $95 per MWh - Vistra/Comanche Peak contract cited as evidence of willingness to pay Meta data-center debt deal: $25 billion at +220 bps over Treasuries - Pimco loan to Meta data-center project in Louisiana Trading level of Meta debt after pricing: +140 bps over Treasuries - Guest says it began trading tighter, creating profit for Pimco Pimco day-one gain: $2 billion - Estimated mark-to-market gain from the Meta loan Northern Indiana ratepayer giveback: $1 billion over 15 years - Example of a deal that returns money to ratepayers Annual ratepayer giveback in NIPSCO example: $67 million per year - Derived from the 15-year total Vogtle nuclear plant cost overrun: $14 billion planned vs. $32 billion actual - Used as evidence why new nuclear is viewed skeptically Vogtle schedule delay: 10 years late - Guest cites delay as key deterrent to new nuclear builds

Pivotal Quotes: "“the old industries that were either stable or cyclical becoming secular in the way they grow.”" — Joe Weisenthal: Framing the central market shift from utilities as bond proxies to utilities as AI-growth beneficiaries "“there is a lot of supply of data centers coming, and it’s very unclear if there’s going to be demand for this.”" — Andy DeVries: Core contrarian thesis on the data-center buildout "“Someone, if it turns out that there’s an overbuild and there is not as much demand for it, someone’s paying for it.”" — Andy DeVries: Explains the ratepayer/shareholder risk if the buildout overshoots

Implications: Utilities, power traders, and credit investors should treat AI infrastructure as a timing and pricing problem, not a guaranteed demand boom. The winners may be those who price risk correctly, protect ratepayers, and avoid overbuilding before actual usage arrives.