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

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

by Bloomberg

46m•February 2, 2026
Listen to Episode

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

This Bloomberg/Odd Lots episode interviews Andy DeVries (head of investment grade credit & head of utilities and power at Credit Suisse) about whether the widely‑cited data center — and AI — power demand story is being overstated. Andy runs a simple supply‑vs‑demand math check and argues the utilities’ committed capacity to serve data centers already looks larger than most third‑party demand forecasts for 2030 — creating a real risk of overbuild, mispriced power curves, credit stress and regulatory/political questions.

Episode snapshot

  • Hosts: Traci Allaway and Joe Weisenthal (Bloomberg)
  • Guest: Andy DeVries, Credit Suisse (utilities & power analyst)
  • Core claim: Utilities are already “committed” to connect far more data center load than third‑party demand forecasts imply — timing is the key source of risk.
  • Format: interview + deep dive on numbers, market pricing, transmission/regulatory issues, and credit/financing.

Core arguments and key numbers

  • Current data center capacity: ~45 GW (compute capacity today).
  • Common 2030 demand forecasts: ~90–95 GW (so ~+50 GW demand vs today).
  • Utilities’ near‑term firm/committed connections: ~140 GW (their announced firm projects).
  • PUE (Power Usage Effectiveness) adjustment: when converting third‑party compute estimates to grid‑connected load (includes cooling and facilities), Andy reduces the 140 GW to ~110 GW apples‑to‑apples.
    • Net takeaway: utilities’ firm commitments (~110 GW adjusted) ≳ the add‑on demand implied to 2030 (~50 GW), suggesting potential oversupply once timing and double‑counting are considered.
  • Example prices:
    • Typical forward power in Texas: mid‑$50s to low‑$60s/MWh on peak.
    • Some big tech deals: ~$95/MWh for around‑the‑clock power (e.g., Vistra/Comanche deal).
  • Natural gas forward curve (example cited): slightly inverted (e.g., ~$3.70 → ~$3.60/MMBtu toward decade end), which market participants argue should be higher if long‑term gas/LNG demand is growing.

How Andy does the math (methodology)

  • Demand: aggregates public reports, sell‑side and consultant forecasts (BNEF among the referenced sources), and media/Gmail alerts for deal announcements.
  • Supply: compiles utilities’ investor‑calls and filings to capture "firm/committed" vs pipeline projects (avoids double counting).
  • Adjusts for PUE so compute‑only estimates become grid‑connected load.
  • Compares these apples‑to‑apples demand and supply numbers and examines forward power & gas curves for market expectations.

Main implications and risks

  • Oversupply risk: if utilities actually build/connect the committed pipeline and end demand is less than forecast, there will be excess generation/transmission capacity and weakening utility/project economics.
  • Timing mismatch: data centers can be operational in ~2–3 years; new generation often takes 6–7 years (plus long turbine lead times). Oversupply risk concentrates toward later years (around 2030), making timing the crucial variable.
  • Price signals vs forecasts: forward power and gas curves currently do not reflect the high growth implied by some hype forecasts — markets look more muted, implying skepticism about long‑term demand.
  • Ratepayer and political risk: if utilities (or their regulators) socialize costs of buildouts that don’t pay off, residential/retail ratepayers could face higher bills unless explicit protections exist.
    • Example: NIPSCO / Nysource structure where deal with Amazon returns ~$1B over 15 years to ratepayers — cited as a “gold standard” protection.
    • Conversely, some states/utilities lack such protections (Louisiana, Mississippi, Tennessee, Texas were mentioned).
  • Credit & financing risk:
    • Private credit appetite for data‑center financings is growing (example: PIMCO financing a Meta data center deal); spreads can be wide but tighten quickly, incentivizing more lending.
    • Off‑balance structures: many hyperscalers structure deals off‑balance (special vehicles, leases); rating agencies often impute lease obligations but documentation details/covenant protections vary — creating hidden risk if guarantees fall away on sale or termination.
    • Covenant erosion and weaker documentation as competition for deals rises is a material credit risk (lower protection for lenders/bondholders).

Regional and market structure points to know

  • Texas (ERCOT) is a special “walled” market (~87 GW peak referenced). Andy argues substantial incremental load could be absorbed by existing dispatchable capacity for limited hours rather than immediate new build.
  • MISO (Midcontinent ISO) and other regions that relied heavily on coal may be more stressed.
  • ISO‑NE (New England) is tight and expensive; local constraints matter greatly — so geography, transmission limits and regional market structure will determine where shortages or overbuilds manifest.
  • Transmission, not just generation, is often the binding constraint: connecting distant renewables (or new loads) requires costly transmission builds and lengthy permitting.

Notable insights & memorable lines

  • “Simple math”: today ~45 GW compute → 2030 ~95 GW (≈ +50 GW); utilities have ~140 GW firm, PUE‑adjusted → ~110 GW.
  • One gigawatt roughly ~1 million homes (region and usage dependent).
  • Data center build cost vs generation: new gas plant capital intensity cited at ~$3,000/kW vs data center build numbers cited much larger — incremental generation cost is small relative to hyperscaler overall project cost, which helps explain why big tech will pay premiums.
  • Timing matters more than ultimate magnitude: the oversupply issue is a 2030 event in Andy’s view — investors’ ability to time that is key.

What to watch (actionable checklist for investors)

  • Compare utilities’ stated “firm/committed” GW to third‑party demand forecasts (and apply PUE adjustments).
  • Monitor forward power curves by region (peak vs off‑peak slopes) — rising curves would support demand growth; flat/inverted curves signal skepticism.
  • Watch natural gas forward curve shapes and LNG export buildouts — gas is the main driver of marginal power prices in many areas.
  • Track state‑level regulatory treatments and ratepayer protections for data‑center interconnection costs (are costs socialized, returned to customers, or absorbed by developers?).
  • Scrutinize private credit deals’ covenant strength and whether financings are on‑ or off‑balance; check guarantees that could be voided on sale.
  • Observe transmission development timelines and permitting bottlenecks (where transmission delays persist, localized constraints or price spikes are likelier).
  • Keep an eye on chip & compute efficiency trends (Jevons Paradox may increase usage, but per‑compute energy cost improvements can blunt demand growth).

Bottom line / Takeaway

Andy DeVries’ takeaway is straightforward: using a transparent, conservative adjustment, the utilities’ own firm commitments to connect data centers approach or exceed commonly cited demand forecasts to 2030 — so the real risk is oversupply and mispriced risk rather than a hard shortage of electrons. The outcome depends heavily on timing (build rates, turbine lead times, transmission permitting), state regulatory frameworks (who ultimately pays), and how markets price forward power and gas. For investors, the priority is not ideological (AI hype vs doom) but pragmatic: follow the numbers, watch timing and contracts, and price credit and regulatory risk accordingly.