Thesis. Headline EBITDA growth of 34% came almost entirely from acquired LS Power assets while per-share earnings fell and missed — but NRG did announce a $3.2B, 1.2 GW "bring your own power" plant for an investment-grade hyperscaler, which closes the gap to its three IPP peers and does not arrive until late 2029.
Reviewed 19 Sep 2026, after Q2 2026 results on 4 Aug. Next review after Q3 2026 earnings on 5 Nov 2026.
What it does
NRG is a retail electricity supplier with owned generation, weighted toward Texas. It sells power directly to residential and commercial customers and into wholesale markets. That combination is a natural hedge: retail margins widen when wholesale prices fall.
Its datacentre model is different from its peers'. TLN and CEG sell existing nuclear output under long PPAs. NRG's flagship deal is "bring your own power" — it builds a new gas plant dedicated to one customer and is paid capacity payments covering 95% of projected free cash flow, whether or not the datacentre runs. CEO Robert Gaudette: "We're paid for the megawatts we build and make available, not for how much the data center runs."
That structure removes utilisation risk and replaces it with construction risk. The money arrives in 2029 or it does not arrive.
Bull case
- A $3.2B, 1.2 GW combined-cycle gas plant in Texas for an unnamed global cloud and AI hyperscaler, confirmed investment-grade. Expandable to 2.4 GW, commercial operation late 2029, minimum 15-year term, ~$2,700/kW. Guided to $500M of annual EBITDA and $375M of annual free cash flow before growth
- Capacity payments cover 95% of projected free cash flow, insulating NRG from datacentre utilisation risk
- 5.4 GW of turbine and EPC capacity secured through 2032 via GE Vernova and Kiewit — in a market where turbine slots are the binding constraint, having them booked is itself an asset. Broader development pipeline above 10 GW
- A separate 295 MW supply agreement for two datacentres on NRG-owned Texas sites, with an option to expand to 1 GW across additional sites; initial powering from H2 2026
- Adjusted EBITDA $1.2B, +34%. Free cash flow before growth $1.025B, +$111M
- Large retail customer base provides recurring revenue that pure merchant generators lack, and it owns dispatchable generation in ERCOT and PJM — the two markets with the most datacentre load growth
- The LS Power acquisition added real scale, contributing $370M in the East segment in its first full quarter
Bear case
- Per-share earnings fell to $1.49 from $1.73 and missed. Headline EBITDA growth of 34% did not reach shareholders — the gap between the two is the problem
- The EBITDA growth came almost entirely from acquired assets, with $370M of East-segment contribution from the first full quarter of LS Power. Strip it out and organic performance is weak
- Houston power cleared at $33/MWh against a $52 plan, and Texas segment EBITDA fell $131M. That miss in a core market is the direct cause of the EPS shortfall
- The flagship deal does not reach commercial operation until late 2029. Between now and then it is construction risk, capital commitment and an unnamed counterparty
- The counterparty is not identified — creditworthiness is management's characterisation, not a disclosure
- Still the least contracted of the four IPPs here in near-term terms, against TLN's ~$18B AWS PPA, CEG's ~920 MW of 18.5-year investment-grade PPAs and VST's 2,609 MW of Meta agreements
- Retail electricity is a genuinely competitive market with limited pricing power, and the company carries full ERCOT weather and price volatility
Major customers
- An unnamed global cloud and AI hyperscaler — the $3.2B, 1.2 GW Texas plant, minimum 15 years, from late 2029. Investment-grade per management
- An unnamed datacentre developer — 295 MW across two datacentres on NRG-owned Texas sites, with an option to 1 GW
- Residential and small commercial retail electricity customers, primarily Texas and the Northeast — millions of accounts, no concentration, but also no contracted long-duration revenue
- Commercial and industrial customers buying power under supply agreements
- Wholesale markets (ERCOT, PJM) and capacity markets — merchant sales and availability payments
Read-through. NRG expresses the AI power theme differently from its peers: not by contracting existing nuclear, but by building new gas capacity to order. That makes it a direct demand signal for GEV — the 5.4 GW of turbine and EPC capacity booked through 2032 is exactly the kind of slot reservation showing up in GE Vernova's 116 GW figure, and it is one of the few places in this file where you can see both sides of the same transaction. It shares ERCOT exposure with VST, so both respond to Texas weather, the large-load interconnection audit and ERCOT price moves. The "bring your own power" structure is also worth watching as a template: if capacity payments covering 95% of free cash flow become standard, the economics of new-build generation for datacentres change across the whole group.
What would change the view
- Whether the 1.2 GW plant stays on schedule for late 2029, and whether the 2.4 GW expansion option is exercised
- Whether the hyperscaler counterparty is ever named, or a second "bring your own power" deal is signed from the 10 GW pipeline
- Houston and ERCOT power clearing prices against plan
- Organic EBITDA excluding the LS Power contribution — the real performance measure
- EPS trajectory versus EBITDA. The gap between them is the thesis problem, not the absolute level of either
- Retail customer count and margin per customer
Update log
19 Sep 2026 — Converted to the standard template and a material correction made. The page's central bear claim — that NRG is "the only one of four independent power producers here with no marquee hyperscaler PPA" — is no longer true and was already out of date when written. At the same Q2 print (4 Aug) NRG announced a $3.2B, 1.2 GW "bring your own power" gas plant for an unnamed investment-grade hyperscaler, minimum 15 years, $500M of annual EBITDA from late 2029, plus 5.4 GW of GE Vernova/Kiewit turbine and EPC capacity booked through 2032 and a separate 295 MW datacentre supply agreement. The Backlog property carried a "not verified this session" flag; it is now verified and rewritten. The remaining bear case — acquired rather than organic growth, the EPS miss, the ERCOT price shortfall and the 2029 timeline — stands. The Avoid verdict predates this and should be revisited.
4 Aug 2026 — Q2 2026 reported. Adjusted EBITDA $1.2B (+34%), free cash flow before growth $1.025B (+$111M), adjusted EPS $1.49 (−$0.24) and a miss. Texas segment EBITDA −$131M on ERCOT prices averaging $33/MWh against a $52 planning assumption. $3.2B / 1.2 GW hyperscaler project unveiled.
Aug 2025 — 295 MW supply agreement for two datacentres on NRG-owned Texas sites, option to 1 GW. Initial powering H2 2026, full operation targeted 2030.
May 2025 — LS Power acquisition (~$12B) added 13 GW of gas generation and a 6 GW virtual power plant platform.
Probabilities: strengthen ~38% / weaken ~45%. These predate the hyperscaler deal above and have not been revisited.
Research and education only — not investment advice.
Made with Bullet