The request
Every site search comes with the same question: how many months does a new grid connection in Texas take, and how far can that estimate be trusted? This example is the test Voltane ran on its own answer. The full write-up is How Voltane estimates time to power.
The estimate is a range, not a date: a low, a middle and a high value, taken from the lower quartile, the median and the upper quartile of past projects in the same group.
What the engine checked
The method is a hold-out by request year. Groups were fitted on ERCOT generation requests filed from 2015-01-01 to 2019-12-31, then compared with requests filed from 2020-01-01 to 2021-12-31 that the method had not seen. The queue history is as of 2026-08-31.
- For each test project that was energized, the error is its actual months from request to energization minus its group's predicted middle value.
- Coverage is the share of those projects that landed inside the group's predicted low-to-high range. 50% would be ideal.
- A group needs 30 energized training projects. Without them, the zone or ERCOT-wide numbers are used.
The data is the Berkeley Lab "Queued Up" interconnection queue data file, 2026 edition, compiled from ERCOT's monthly Generator Interconnection Status reports, plus the ERCOT GIS report of August 2026.
What came back
The test set held 135 projects. The median error was 11.0 months, and 57% of actual results fell inside the predicted range.
| Group | Test projects | Predicted low | Predicted middle | Predicted high | Actual middle | Actual inside the range |
|---|---|---|---|---|---|---|
| North zone | 49 | 42.3 | 54.5 | 65.7 | 45.6 | 61.2% |
| South zone | 29 | 45.5 | 60.1 | 71.1 | 47.2 | 55.2% |
| ERCOT-wide | 17 | 39.0 | 52.2 | 66.4 | 49.0 | 70.6% |
| Coastal zone | 10 | 46.9 | 53.6 | 67.4 | 45.4 | 40% |
| West zone | 18 | 33.6 | 46.7 | 59.2 | 35.1 | 50% |
| West zone, 345 kV | 12 | 34.0 | 48.9 | 60.5 | 40.4 | 50% |
All values are months from an interconnection request to energization. In every group the actual middle came in under the predicted middle. That is not good news about speed. The test projects that could be counted are the ones that finished within 48 to 72 months, so the actual medians run short and will rise as slower projects finish.
The same method on PJM's public history, for comparison (Time to power in PJM):
| Market | Requests in the fitted years | Reached operation, of those resolved | Months to a signed agreement, middle | Months to power: low, middle, high | Backtest median error |
|---|---|---|---|---|---|
| ERCOT (Texas) | 907 | 45.1% | 22.4 | 38.8, 50.6, 62.3 | 11.0 months |
| PJM | 3,074 | 15.9% | 37.8 | 26.9, 53.4, 70.6 | 14.0 months |
The two clocks do not stop at the same point. ERCOT's figures use approval to energize or synchronize where the report lists one, else operation. PJM's use operation. Voltane's parcel screening is Texas today. PJM is planned, not live.
What stayed unknown
- Large loads. This is generation queue history, not large-load history. A data center goes through a different process (and Texas SB 6 rules), and no public site-level large-load history exists yet.
- Projects still in the queue. 282 of the 564 test requests were still in the queue at the as-of date. They are neither failures nor successes and are left out of the error and coverage numbers.
- The estimate ran optimistic. Checked without that gap, the model gave the 564 test requests a 16.7% average chance of energization within 48 months, and 14.4% of them were. In PJM the gap was wider: 9.1% predicted against 0.9% actual, with only 27 of 1,794 test requests in operation and counted.
- The site itself. A group is not a site. The estimate does not study a substation, a line or a request. It is not available capacity, a queue position or a date a utility has agreed to.
- Small samples. A group's numbers rest on a few dozen projects, with solar, wind, storage and gas pooled.
- Whether history still applies. The training outcomes are as known today, so this is a hold-out by request year, not a forecast that was made in the past. PJM also changed its process: its transition date was July 10, 2023.
What a client would do next
Treat the range as a planning input, never as a date. In a Voltane report the estimate sits apart from sourced facts, is labelled "Estimate", and names its group, its range and the number of past projects behind it.
- Read the estimate against your target year, using the high value as well as the middle.
- Ask the utility or ERCOT for the study that actually sets a date. For a large load, that is the large-load process, which this history does not cover.
- Use the time to power definition when comparing estimates from different sources: where the clock starts and stops changes the number.
If you want this estimate attached to candidate sites for your own project, start at Find me a site. For a site you already hold, Analyze my site is available by request.
Questions people ask
How accurate was the ERCOT time to power backtest?
On 135 test projects the median error was 11.0 months, and 57% of actual results fell inside the predicted low-to-high range. The estimate also ran optimistic on the chance of energization within 48 months: 16.7% predicted against 14.4% actual.
Does the estimate apply to data centers?
Not directly. It is fitted on generation interconnection history: power plants and storage asking to connect. Large loads follow a different process, and no public site-level large-load history exists yet.
Is the estimate a date a utility has agreed to?
No. It says what happened to past projects in the same group. It is not available capacity, a queue position or an agreed date.
Sources
- ERCOT Generation Interconnection Status report
- Berkeley Lab "Queued Up" interconnection queue data file, 2026 edition (CC BY 4.0)
Related
- Find me a site
- Analyze my site
- Sample Opportunity Brief
- How Voltane estimates time to power
- Time to power in PJM
- Glossary: time to power
More worked examples
- Worked example: a Texas solar shortlist, from request to brief
- Worked example: checking the screen against real projects
- Worked example: a 300 MW data center request in ERCOT
Need a site, or have one?
Tell us what you need and Voltane finds, reports on and prices candidate sites in ERCOT. Or bring a site you already have for a path-to-power report (by request for now).