Generation

Capacity buildout

Description

Capacity additions are determined by our proprietary Capacity Expansion Model based on techno-economic criteria (see more here). In the front end of our forecast, we apply a minimum build limit based on the quantity of generation with high likelihood to come online based on interconnection progress. For gas generation, we set a maximum build due to supply chain constraints around the construction of new gas turbines.

After the buildout values are decided for each of ERCOT’s load zones, projects from the interconnection queue are systematically selected based on a likelihood order that’s determined by factors such as their progress in securing interconnection agreements and financing.

We exclude Texas Energy Fund (TEF) projects from the Capacity Expansion Model, as their build decisions do not follow a standard financial framework. Instead, they are incorporated separately based on our proprietary assessment of build decisions and expected commercial operation timelines. Given the policy-driven nature of these assets, alternative commissioning timelines for TEF-backed projects are explored through dedicated sensitivity analysis cases.

Plant retirements are incorporated based on publicly announced decomissioning (from EIA) where available. In the absence of such announcements, retirements are projected using technology-specific lifespan assumptions, drawing from the EIA Generator Inventory dataset. For coal-fired generation in particular, alternative decommissioning schedules and retirement volumes are examined through sensitivity analysis cases to reflect uncertainty around policy, market conditions, and plant economics.

Data sources

Unit Commitment

Unit commitment refers to the modeling of generator on/off decisions through binary variables that determine whether a generating unit is online or offline at any given hour.

In the model, a binary commitment variable is assigned to each thermal generating unit. Operational limits — specifically minimum and maximum generation levels — are conditional on this commitment status. When a unit is online, it must operate within its defined minimum stable output and maximum output levels. When offline, a unit’s output is constrained to zero. This structure allows the model to capture key operational behaviors such as startup costs, minimum load requirements, and restricted flexibility, reflecting real-world generator limitations.

Short-run marginal costs

Description

Short-run marginal costs (SRMCs) are estimated for all generation technologies in the system, including thermal and renewable assets. These values are derived primarily from fuel prices and heat rate assumptions. Fuel prices are sourced from the EIA’s Annual Energy Outlook commodity input data for coal and natural gas.

Where available, historical operating data is used to capture time-varying heat rates, allowing for seasonal differences and more realistic dispatch behavior. Specifically, we use ERCOT’s 60-day SCED disclosure data to estimate historical generator performance and derive implied unit-level heat rates. For new or prospective resources without operational data, technology-average heat rates are used (again derived from SCED data). We use multiple bid steps for thermal generation, including new generation, which again are informed by historical SCED data. For renewable technologies such as wind and solar, SRMCs are adjusted to reflect tax credits, which can result in negative marginal costs.

Commodity prices are derived from ICE natural gas futures in the short term and blended with EIA’s Annual Energy Outlook (AEO) projections over the long term to create a consistent fuel price trajectory across the forecast horizon. Recognizing the inherent uncertainty in fuel markets, alternative commodity price paths are evaluated through sensitivity analysis cases to assess their impact on dispatch, revenues, and capacity outcomes.

Policy drivers

• Production tax credits (PTC) and investment tax credits (ITC) are incorporated into SRMC and CAPEX assumptions for each eligible technology. Under the OBBB Act, these credits phase out for renewable projects commissioned after 2028, and phase down in a step-wise schedule for storage projects commissioned after 2034.

Assumptions and caveats

• Technology-average heat rates are used for future or unbuilt units.
• We use a two-step bid step curve, while plants may bid with dozens of steps in SCED if desired.
• For generators existing in our weather year (2024) we take their historical bids in that weather year, adjusted for inflation and commodity price changes.

Data sources

Start costs

Description

Startup costs are modeled for thermal technologies that require discrete commitment decisions. Nuclear, solar, wind, and storage are assumed to be continuously available and do not incur startup costs in the model.

Start cost values are derived from historical market data, imputed using forecast commodity prices to estimate time-varying startup costs. These values are expressed as a fixed cost in $/MW per start and are applied only in the unit commitment step of the model to influence plant scheduling decisions.

Assumptions and caveats

• Startup costs are only applied to technologies that require explicit unit commitment.
• No modeling of hot vs. cold start cost differences.
• Startup costs are excluded from economic dispatch; they influence commitment only.

Data sources

Renewable load factors and outages

Description

Both load factors and outages profiles are modeled similarly and use historical available capacity data from ERCOT’s SCED disclosures; specifically, the High Sustained Limit (HSL) of individual units. Each generator’s HSL is expressed as a percentage of its nameplate capacity (sourced from the MORA dataset) to calculate an empirical load factor.

For existing wind and solar generators, this observed load factor is held constant throughout the forecast horizon. New build generators are assigned load factors based on the regional (zonal) average for the same technology class, reflecting geographic diversity and typical resource performance in that area.

Outages are modeled using the same HSL-derived method; reductions in available capacity reflecting planned or unplanned outages—are captured in the empirical load factor calculation.

Assumptions and caveats

• Load factors and outages rates are based on a “weather year”, which is 2024, and do not evolve over time.
• New builds inherit regional average load factors for their technology class.
• Outages are captured via reduced HSL values, without explicit modeling of whether an outage is planned or unplanned.
• No adjustment is made to load factors or outages over time to reflect technology improvements or generator wear and tear.

Data sources