Working paper · UCLA MQE capstone · 2026

Can Texas's renewables keep up?

AI data centers are pouring into Dallas. Texas's wind and solar are hundreds of miles west. I used three years of hourly grid prices to see whether one can actually cover the other.

Ketki Prasad SawantAdvisor: Prof. Randall R. RojasERCOT · 2023–2025 · 25,202 hours
The short answer

Texas has the megawatts. It can't get them to Dallas around the clock.

AI demand is already in the grid data. As it grows, wind and solar do less and less to relieve it, and by 2030 the margin disappears in the base case.

534 MW
of data center capacity already visible in North Texas load by 2025
≈0
solar's effect on grid stress when Dallas demand is high
+$5.43
per MWh: unexplained jump in the West–North price gap after 2024
5.4 TWh
2030 shortfall if data center growth runs hot
01 · Why I did this

Everyone's forecasting AI's power bill. Few are measuring it.

The big AI energy numbers are projections, built from announced projects and industry assumptions. I wanted to know how much of that demand has already shown up in real market data, and whether clean energy is keeping pace.

Texas is the best place to test it. It has more wind and solar than any other state, one of the fastest-growing data center markets, and a grid that publishes prices for every zone, every hour. If renewables can keep up anywhere, it should be here.

183 → 426

TWh of US data center use, 2024 to 2030 projection (IEA)

226 GW

of large loads asking to connect to ERCOT; about 73% are data centers

#3

Dallas–Fort Worth's rank among US data center markets (CBRE)

#1

Texas's rank in installed wind and solar

02 · How I measured it

The price gap is a stress gauge.

Power has a location

Electricity is priced separately in each zone, every hour. If the wires between two zones had unlimited room, prices would match.

West: $22/MWh
North: $22/MWh
Gap: $0 → power flows freely

Full wires split prices

When the lines fill up, each zone has to settle on its own, and the prices drift apart.

West: $30/MWh
North: $18/MWh
Gap: +$12 → the system is strained

Watch what moves it

That gap, West minus North, is the basis spread. Over 25,000 hours I tested what pushes it up and what pulls it down.

Avg gap: $8.83/MWh
Swings: ±$16.56
Range: −$69 to +$169
Lower gap · less strain Higher gap · more strain ◀ WIND · SOLAR pull it down DALLAS DEMAND · COLD SNAPS push it up ▶ If renewables are keeping up, their pull should hold, even as demand grows.
03 · What I found

Four findings

Finding 1

The AI boom is already in the grid data.

North zone load grew 23.5% in 2024 and 14.0% in 2025, five times faster than any other zone. Two tests point to data centers.

The 24/7 fingerprint. Data centers never sleep, so they lift night-time load more than evening peaks. North was the only zone where off-peak load grew faster than peak load (+2.71 pts in 2024).

The counterfactual. Had North grown like comparable zones, it would be 427 MW lighter today. That's 3.74 TWh a year, about 534 MW of installed capacity.

In plain English

You can measure AI's electricity footprint from public market data, and it lands within 13% of what the real estate industry reports.

My estimate vs. the industry's count Interactive

Implied data center capacity, North Texas. Primary baseline (West + NCENT growth, 1.9%/yr), 80% capacity factor. Range across baselines: 465–538 MW. CBRE, North America Data Center Trends H2 2025.
Only North has the 24/7 fingerprint
Off-peak minus on-peak load growth, percentage points. Peak = weekdays 4–9 PM. South, Houston and West were all negative in 2024.
Finding 2

The more Dallas needs, the less wind and solar help.

On average, 1,000 MW more wind cuts the gap by $1.13/MWh and solar by $0.45. Dallas load pushes it up by $3.26 per 1,000 MW.

But those renewable effects shrink as Dallas demand rises. At high load, wind is 51% less effective, and solar's effect drops to almost zero.

On top of that, the gap sits $5.43/MWh higher in 2024–25 than in 2023, after controlling for weather, fuel, the economy and renewable output. The timing lines up with the data center surge.

In plain English

Building more West Texas renewables gives Dallas less and less relief. The grid needs power that's there at night and can reach the city.

How much does new wind help Dallas? Drag

Price-gap reduction in $/MWh, from Model 1 with interactions (R² = 0.495). The effect is linear in Dallas load and scales linearly with MW added; observed load range 739–2,732 MW. Wind × North t = 4.41; Solar × North t = 7.11; both p < 0.001. Newey–West HAC SE, 24 lags; hour, month and day-of-week fixed effects.
A step up after 2024
Post-2024 coefficient +5.43 (t = 6.96, p < 0.001). North × Post-2024 also positive (p < 0.05); wind and solar × Post-2024 not significant, which points to a demand-side shift.
Finding 3

Data centers drive normal days. Cold snaps drive the worst ones.

I compared what drives an ordinary hour with what drives the most extreme 5% of hours. They're different.

Dallas load raises the gap through the middle of the distribution, then fades out at the top. Heating demand barely matters on a normal day but takes over in the extreme tail, about 96× stronger at the 95th percentile than at the 25th.

In plain English

Data centers raise everyday costs. Winter storms cause the blowouts. They're separate risks and need separate fixes.

Effect on the price gap, normal hours to extreme hours Switch

Model 2: quantile regression at the 25th–95th percentiles of the gap. Bootstrapped SE, 500 reps. Hollow dots are not statistically significant. Coefficients are $/MWh per MW (load, wind, solar) or per degree-day (cold).
Finding 4

Demand raises the floor. Weather and renewables cause the swings.

On daily data, yesterday's wind, solar and cold help predict tomorrow's gap. Yesterday's Dallas load doesn't (p = 0.76).

A random forest, which assumes nothing about the shape of the relationships, ranks the same way: solar and wind on top, Dallas load fifth.

In plain English

Data centers push the price gap up slowly and steadily. The day-to-day volatility comes from sun, wind and weather, the things nobody can schedule.

What predicts tomorrow's gap?
VariableF-statp-valuePredicts?
Solar7.625<0.001Yes
Cold (HDD)3.2020.002Yes
Wind2.8820.005Yes
Dallas (North) load0.5980.758No
Gas price0.5810.772No
Model 3: VAR with Granger causality tests, daily data, N = 1,096, 7 lags (AIC).
What matters most to the model
Model 4: random forest, 100 trees, 80/20 split. Out-of-sample R² = 0.536, MAE = $5.82/MWh. Top 8 of 13 features; seasonality is encoded as sine and cosine of month (17.4% combined).
04 · Looking ahead

By 2030, the buffer is gone in the base case.

Raw capacity numbers overstate what helps Dallas. So I only counted renewables close enough to matter (20% of ERCOT's wind and solar) and only the share that lines up with the hours data centers run (25%). The scenarios pair fast data center growth with slower renewable build-out, and the reverse. Pick a preset or drag the sliders to test your own assumptions.

Run your own 2030 scenario Interactive

2030 demand
Net 2030 gap
Break-even growth
the fastest data centers can grow before renewables fall behind
new data center demanddeliverable renewablesshortfall
2030 net gap: your scenario vs. the paper's, TWh
Model: data center demand starts at 3.74 TWh (2025) and compounds; capacity assumes an 80% load factor. Renewables start from 2025 ERCOT wind + solar (184.1 TWh), of which 20% is treated as local; only the deliverable share of new local output offsets new demand. Presets reproduce the paper's Table 7: data center growth 10 / 18 / 28%/yr paired with renewable growth 12 / 10 / 7%/yr. Negative = covered; positive = shortfall.
05 · The conclusion

It's not a volume problem. It's a timing and location problem.

Texas will keep building record amounts of wind and solar. But data centers need power that's on at 3 AM in January and can actually reach Dallas. Wind and solar from West Texas can't guarantee that on their own, and the market data already shows the strain.

Demand is realAbout 534 MW of AI load is already measurable in Dallas.
Relief is fadingRenewables cut grid stress less as that load grows.
The floor movedThe price gap stepped up $5.43/MWh after 2024.
The margin closesNear zero by 2030 in the base case; 5.4 TWh short if growth runs hot.
06 · How to use this

Who should care, and what to do with it

ERCOT · utilities · planners

Grid planning

  • Model data center load as flat and 24/7, not weather-shaped.
  • Pair renewables with firm capacity: batteries and gas backup.
  • Prioritize West–North transmission, where the price gap shows the bottleneck.
Hyperscalers · developers

Data center strategy

  • A West Texas solar contract won't cover Dallas load at night.
  • Treat on-site storage or firm supply as part of the site plan.
  • Shift flexible workloads, like training jobs, into midday solar hours.
Traders · PPA buyers · investors

Pricing risk

  • West–North basis risk has stepped up, and it's structural.
  • Hedge tail risk around winter weather, not data center load.
  • For day-ahead calls, weight renewables and weather over load.
Policymakers · researchers

Tracking AI demand

  • Use this method to track AI demand from public market data, without waiting for disclosures.
  • Apply it to other clusters: Northern Virginia, Phoenix, Atlanta.
  • Fold data center baseload into long-term adequacy plans.
07 · Caveats

What this can't tell you

A proxy, not a meter

Data center demand is inferred from North zone load against a counterfactual. The CBRE check and placebo tests support it, but it isn't facility-level data.

Three years

2023–2025 covers the start of the boom. A longer panel would firm up the post-2024 shift.

Assumption-driven 2030

The scenarios depend on the growth, local-share and deliverability assumptions. Treat them as a range, not a forecast.

Prediction vs. cause

The random forest ranks predictors; it doesn't prove causation. Causal claims rest on the regressions and placebo tests.

08 · Data and methods

How it was built

Prices & load
ERCOT settlement point prices and native load by zone, hourly, Jan 2023 – Dec 2025
Renewables
ERCOT system-wide wind and solar output
Weather
Open-Meteo (ERA5) hourly temperature for Dallas–Fort Worth and Midland; heating and cooling degree days at 18°C
Economy & fuel
Texas Coincident Index (FRED, TXPHCI), first-differenced; daily natural gas price
Models
(1) OLS with HAC errors, interactions and a post-2024 break · (2) quantile regression · (3) VAR with Granger causality · (4) random forest
Robustness
Day-clustered errors; arcsinh-transformed gap; NCENT placebo (Dallas effect flips sign when swapped for Austin–San Antonio load); West–Houston placebo