Where should hydrogen go when nobody knows how much wind will be built?

Why a hydrogen plan sized for one forecast can cost too much in the others, and how a plan that weighs every build-out differs

Case file CF3 of Energy Systems, Step by Step: why a hydrogen plan sized for one forecast can cost too much in the others, and how a plan that weighs every build-out differs.
  • Analyst · L2
  • Case file
  • 10 min
  • After lesson 12

1 · The puzzle

In ERCOT, wind and solar output is concentrated far from electricity demand in the north and southeast, and from hydrogen demand on the Gulf Coast.

How much planned renewable capacity is built by 2030, and how fast demand grows, are both uncertain.

2 · The question in one sentence

Where should electrolyzers, hydrogen storage and fuel cells be built in ERCOT, and how should hydrogen move between zones, to absorb surplus renewable generation and supply industrial hydrogen demand under uncertain 2030 renewable build-out?

Question type: prescriptive (what should be built, where, and how should it run?).

3 · Scale and boundary

  • Inside: ERCOT's eight weather zones, every hour of a year (8,760 hours), nine scenarios of renewable build-out and electricity demand, and industrial hydrogen demand in each zone.
  • Left out: uncertainty in hydrogen demand, which is held fixed; truck and rail transport; post-2023 federal hydrogen credit rules.

4 · Decisions, goal and rules, in words

  • Decisions: the sizes of electrolyzers, hydrogen tanks and fuel cells in each zone, chosen once; then, in each scenario, how batteries, electrolysis, fuel cells, gas-fired backup, reformer hydrogen and pipeline flows run every hour.
  • Goal: the lowest expected cost over all nine scenarios, net of hydrogen and oxygen sales, with the CO₂ from gas plants and reformers captured and stored at a cost.
  • Rules: electricity demand is met in every zone and hour; industrial hydrogen demand is met by green or reformer hydrogen; hydrogen moves only by pipeline between neighbouring zones; no unit runs above its size.

5 · Data and method

The method in general: choose equipment sizes now, then run them differently in each possible future, weighing each future's cost by its chance.

  • Inputs are built from ERCOT's 60-day market disclosures, the LBNL interconnection queue and EIA data.
  • The same hourly capacity factors of each zone scale existing and planned wind and solar, so both share one hourly basis.
  • A two-stage stochastic linear program sized electrolyzers, hydrogen tanks and fuel cells once in each of ERCOT's eight weather zones, with hydrogen pipelines between neighbouring zones, then scheduled the system hour by hour over a full year in nine scenarios:
    • three completion rates of planned renewable capacity by three electricity-demand levels.
  • Hydrogen equipment is sized once, then operated hour by hour in nine scenarios; the sizes minimize expected cost over all nine scenarios instead of fitting a single forecast.

6 · What came out, and where it holds

When gas-fired backup power became much more expensive, the model built fuel cells and turned stored green hydrogen back into electricity.

Industry then drew more hydrogen from gas reformers, and the round trip from power to hydrogen and back lost energy, so total emissions rose although gas-plant emissions fell.

Electrolyzers and hydrogen tanks, combined with hydrogen transport between zones, could meet most of the hydrogen demand in certain zones.

Across ERCOT, renewable (green) hydrogen covered only a few percent of industrial hydrogen demand; reformer hydrogen remained the main supply.

Inputs reflect 2022–2023 conditions; post-2023 federal hydrogen credit rules are not modeled.

Holds when: ERCOT's eight weather zones, inputs from 2022–2023, and hydrogen demand held fixed; the emissions result comes from runs in which power from gas-fired plants became more expensive.

Leaves out: post-2023 federal hydrogen credit rules; truck and rail transport.

The study in time

  1. March 2025 · Journal article (this case). Electrolyzers, hydrogen tanks and fuel cells are sized for ERCOT's eight weather zones under uncertain renewable build-out and electricity demand, with hydrogen pipelines between neighbouring zones; published online on 2 March 2025, in the March 2025 issue. Source: Li et al. (2025), International Journal of Hydrogen Energy 113, 77–89
  2. 30 October 2026 · Code released. A documented, runnable implementation of the planning framework, with synthetic examples (version 1.0.2). Source: txh2-ercot-hydrogen on GitHub, as of 5 November 2026

Every result about the study is quoted from approved excerpts of the published paper named in the sources. The toy below uses illustrative round numbers.

The idea behind it, then where the result holds

Size it before the wind is built (illustrative)

M8's toy: one windy zone, three build-outs with chances 25%, 50% and 25%, illustrative round numbers in 2025 dollars.

  • Plan for the average: the best size if the wind built were sure to equal its average, 5,000 MW.
  • Hedged plan: the size with the lowest expected cost, each build-out's cost weighted by its chance.
  • Worst-case plan: the size whose cost in the worst build-out is lowest; it uses no chances.
Annual cost of the hydrogen supply for three sizes, in each build-out and in expectation
Plan Size None built (25%) Half built (50%) All built (25%) Expected
Plan for the average 1,300 MW $767.89M $662.63M $601.72M $673.72M
Hedged plan 845 MW $731.49M $663.09M $626.74M $671.10M
Worst-case plan 380 MW $703.40M $688.66M $652.32M $683.26M

Expected cost of the hedged plan: 25% × $731.49M + 50% × $663.09M + 25% × $626.74M = $671.10M a year.

A chance-weighted average: not a guaranteed cost, and not what any one build-out costs.

With gas reformers alone, every build-out costs $709.56M a year.

The toy's assumptions and its wind blocks
  • Illustrative: the zone's lines and local demand take up to 1,000 MW of wind in any hour; wind above that is surplus, curtailed unless an electrolyzer uses it.
  • Illustrative: 3,000 MW of wind built, 4,000 MW more planned; how much of the planned wind gets built is the uncertainty.
  • Illustrative: $80,000 per MW a year, the annualized capital and fixed upkeep; 18 kg of hydrogen per MWh (about 55 kWh per kg). It runs only on surplus wind.
  • Illustrative: industry takes 54,000 kg of hydrogen every hour; what the electrolyzer does not make comes from gas reformers at $1.50 per kg.
  • Surplus wind is free: the electrolyzer pays only its own yearly cost, and wind it does not use is curtailed.
  • The year is five blocks of hours sorted by wind output (a duration curve), not 8,760 hours in time order.
  • Industry's hydrogen demand is steady: the same in every hour and every build-out.
  • One zone: no hydrogen storage, no fuel cells and no pipelines; reformer hydrogen has one fixed price.
The five wind blocks: SPP's 2025 hourly wind shape scaled to 33.8%, sorted and cut
Wind block Hours a year Capacity factor
Windiest 500 hours 500 57.1%
Next 1,000 hours 1,000 51.7%
Next 1,500 hours 1,500 46%
Next 2,000 hours 2,000 36.9%
Calmest 3,760 hours 3,760 19.4%

Slide the price of gas-fired power

  1. When gas-fired backup power becomes more expensive, the model starts to use stored green hydrogen in fuel cells to make electricity in hours of low wind and sun.
  2. That hydrogen no longer supplies industry, so gas reformers fill the gap with fossil hydrogen.
  3. Converting electricity to hydrogen and back loses energy, so in the paper's high-price cases total emissions rose even though emissions from gas plants fell.

Illustrative model with stylized numbers: it shows the mechanism described in the paper, not the study's data or results.

Where the result holds

It holds when
  • ERCOT's eight weather zones, every hour of one year, in nine scenarios of renewable build-out and electricity demand;
  • inputs reflect 2022–2023 conditions;
  • hydrogen demand is held fixed across scenarios;
  • the CO₂ from gas plants and reformers is captured and stored at a cost;
  • the emissions result comes from sensitivity runs in which power from gas-fired plants became more expensive.
It leaves out
  • post-2023 federal hydrogen credit rules;
  • uncertainty in hydrogen demand;
  • truck and rail transport of hydrogen.

From the approved excerpt of Li et al. (2025), International Journal of Hydrogen Energy.

Go deeper · ModelerSizes now, operation later, in one program
  • Sizes chosen before the uncertainty resolves are here-and-now decisions; how the equipment runs in each scenario is recourse.
  • One linear program holds both: one size, one copy of the operation for each scenario, each copy’s cost weighted by its chance.
  • M8 builds a small one in the browser, prices the plan for the average future against it, and compares robust plans that use no chances.

Check yourself, then try it

Myth or fact? “The plan for the most likely future is the best plan to build now.”

  • Myth
  • Fact
Answer

Myth. In the toy, the plan for the middle build-out (1,300 MW) is expected to cost $2.62M a year more than the hedged plan (845 MW), which weighs all three.

Myth or fact? “A hedged plan beats building nothing in every build-out.”

  • Myth
  • Fact
Answer

Myth. In the toy, if none of the planned wind is built, the hedged plan costs $731.49M a year against $709.56M with gas reformers alone. It wins on average, not in every branch.

With open code

Optional, about 30 minutes; needs Python 3.10 or later (lesson B2).

Try it yourself: the framework solves a small invented region (four zones, three days) in a few seconds.

Halve the capital cost of the hydrogen equipment, or allow ±5% load flexibility, and compare the sizes and the emissions.

Its numbers say nothing about ERCOT or Texas.

txh2-ercot-hydrogen: public example repository · MIT licence · synthetic data.

Sources and what’s next

Keep going

Back to the course map

Where this came up before

M8 builds this case's toy with a solver: the plan for the average, the hedged plan and two robust plans, and what a perfect forecast would be worth.

A16 follows hydrogen from power to fuel and back, with the losses at each step.

Sources and further reading

Spotted an error, or stuck? Write to hlli@ksu.edu. Corrections are made in place and dated here.

Text licensed CC BY 4.0, code MIT: teachers and students may reuse and adapt it, with credit to Honglin Li.