Residential Ownership vs TPO in California 2020-2026¶
This notebook analyzes direct solar ownership versus third-party ownership for residential applications using public CPUC data downloadable from DGStats. At the time of download, the latest accessible data from DGStats up until the end of August, 2026.
This analysis is done by SolarWAVE Action, a California based nonprofit defending clean energy ownership. All our code is open source and can be found in our GitHub repository. For any questions or comments, please reach out to hello@solarwaveaction.org.
Below, Python packages are imported and the application data from Pacific Gas & Electric (PG&E), Southern California Edison (SCE) and San Diego Gas & Electric (SDG&E) are consolidated.
import datetime
from IPython.display import HTML
import os
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
import plotly.io as pio
import sys
module_path = os.path.abspath(os.path.join('..'))
if module_path not in sys.path:
sys.path.append(module_path)
import data_analysis as data_analysis
import data_visualization as data_vis
import make_posts as make_posts
import write_graphs as write_graphs
pio.templates.default = 'plotly_white'
pio.renderers.default = 'notebook'
data_dir = '/Users/jenny.folkesson/Data/solar/'
dgstats_dir = os.path.join(data_dir, 'Interconnection_Applications_Dataset_2026-08-31/')
logo_path = "https://solarwaveaction.org/assets/images/SolarWAVEActionLogoTransparent.png"
df_total = data_analysis.read_stored_data(dgstats_dir, "dgstats_preprocessed_data.csv")
Reading existing file: /Users/jenny.folkesson/Data/solar/Interconnection_Applications_Dataset_2026-08-31/dgstats_preprocessed_data.csv
For this analysis, we are looking at solar applications, so we'll filter for applications that have capacity greater than zero.
df_res = df_total[(df_total['Customer Sector'] == 'Residential')
& df_total['has_solar']
& (df_total['System Size DC'] > 0)].copy()
Residential Direct Owned and TPO Solar From 2020-August 2026¶
In order to evaluate the state of the residential rooftop solar market, we divide the customer segment (residential) by their 'Third Party Owned' (TPO) status in DGStats. We create two graphs; the top one displays residental solar applications by application date as a bar chart, where bars are split into TPO or Owned. The bottom graph shows a three month rolling average of the TPO share for the same time interval. Relevant policy changes are displayed as dashed vertical lines on the dates those changes happened:
- The Net Billing Tariff (NBT) decision of December 15, 2022
- The Net Energy Metering (NEM) 2.0 cutoff on April 15, 2023
- The signing of the One Big Beautiful Bill (OBBB) Act on July 4, 2025
- The end of the residential clean energy tax credit (25D) on January 1, 2026.
fig = data_vis.residential_tpo_barchart(
df_res,
date_type='App Received Date',
y_label='Count', year_min=2020,
freq='monthly',
incomplete_months=3,
)
write_path = os.path.join(data_dir, "residential_applications_tpo_share.html")
title_text = "The California Residential Solar Market: Two Shocks, One Graph"
caption_text = "Top: Monthly number of residential solar applications by application date, split into \
owned and Third-Party Owned (TPO) applications. Bottom: Three month rolling average of TPO <br>application share. \
DGStats only publish completed applications, and with a median application time of less than a month, \
the last three months are shaded as incomplete."
source_text = "Source: SolarWAVE Action analysis of CPUC California DG Stats data. 2026 includes completed applications through Aug 31."
write_graphs.write_web_chart(
fig,
write_path,
title=title_text,
subtitle=None,
caption=caption_text,
source=source_text,
logo_path=logo_path,
)
make_posts.residential_tpo_social(df_res, '/Users/jenny.folkesson/Data/Solar', 'https://solarwaveaction.org/blog/tpo_ownership_rate/')
make_posts.residential_tpo_newsletter(df_res, '/Users/jenny.folkesson/Data/Solar', 'https://solarwaveaction.org/blog/tpo_ownership_rate/')
html_str = fig.to_html(include_plotlyjs='cdn', full_html=False)
HTML(html_str)
The last three months of data are marked as incomplete due to the fact that DGStats only publishes completed applications. To validate that three months is a reasonable time interval to mark as incomplete, let's look at the median application time for owned and TPO applications for the year 2025.
df_res['App Days'] = (df_res['App Approved Date'] - df_res['App Received Date']).dt.days
df_res['Count'] = 1
df_2025 = df_res[df_res['App Received Date'].dt.year == 2025]
df_owned = df_2025[df_2025['Third Party Owned'] == 'No']
df_tpo = df_2025[df_2025['Third Party Owned'] == 'Yes']
print('Median TPO application time: {:.0f} days'.format(df_tpo['App Days'].median()))
print('Median direct owned application time: {:.0f} days'.format(df_owned['App Days'].median()))
print('80th percentile for TPO application time: {:.0f} days'.format(df_tpo['App Days'].quantile(.80)))
print('80th percentile for owned application time: {:.0f} days'.format(df_owned['App Days'].quantile(.80)))
Median TPO application time: 15 days Median direct owned application time: 27 days 80th percentile for TPO application time: 73 days 80th percentile for owned application time: 87 days
TPO applications are nearly twice as fast as direct owned ones, but both are less than a month. By marking the last three months as incomplete, the remaining data will likely capture over 80% of applications. What would be even better was if the CPUC would publish pending application in an easily accessible form so that the current state of the market would be analyzable.
Below, we calculate the average number of residential applications per month in 2020, 2022 and 2024 to compare the market before and after the NEM 2.0 cutoff.
df_2020 = df_res[df_res['App Received Date'].dt.year == 2020]
df_2020 = df_2020[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2020 = df_2020.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
print('2020 monthly mean application count: {:.0f}'.format(df_2020['Count'].mean()))
2020 monthly mean application count: 12156
df_2022 = df_res[df_res['App Received Date'].dt.year == 2022]
df_2022 = df_2022[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2022 = df_2022.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
print('2022 monthly mean application count: {:.0f}'.format(df_2022['Count'].mean()))
2022 monthly mean application count: 21678
df_2024 = df_res[df_res['App Received Date'].dt.year == 2024]
df_2024 = df_2024[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2024 = df_2024.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
print('2024 monthly mean application count: {:.0f}'.format(df_2024['Count'].mean()))
2024 monthly mean application count: 10547
Comparison of customer owned applications in the first and second half of 2025.
df_owned = df_res[df_res['Third Party Owned'] == 'No']
df_2025 = df_owned[(df_owned['App Received Date'].dt.year == 2025) & (df_owned['App Received Date'].dt.month.isin(range(1, 7)))]
df_2025 = df_2025[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2025 = df_2025.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
print('First half of 2025 monthly mean application count for owned applications: {:.0f}'.format(df_2025['Count'].mean()))
First half of 2025 monthly mean application count for owned applications: 5453
df_2025 = df_owned[(df_owned['App Received Date'].dt.year == 2025) & (df_owned['App Received Date'].dt.month.isin(range(7, 13)))]
df_2025 = df_2025[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2025 = df_2025.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
print('Last half of 2025 monthly mean application count for owned applications: {:.0f}'.format(df_2025['Count'].mean()))
Last half of 2025 monthly mean application count for owned applications: 7236
Below is the year-on-year difference in spring (March to May) of 2026
df_2026 = df_res[df_res['App Received Date'].dt.year == 2026]
df_2026 = df_2026[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2026 = df_2026.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
df_2026_spring = df_2026[(df_2026.index.month > 2) & (df_2026.index.month < 6)].sum()
print("Applications in the spring of 2026: {}".format(int(df_2026_spring.Count)))
df_2025 = df_res[df_res['App Received Date'].dt.year == 2025]
df_2025 = df_2025[['App Received Date', 'System Size DC', 'Count']].set_index('App Received Date').rename_axis(None)
df_2025 = df_2025.resample('ME').agg({'System Size DC': 'sum', 'Count': 'sum'})
df_2025_spring = df_2025[(df_2025.index.month > 2) & (df_2025.index.month < 6)].sum()
print("Applications in the spring of 2025: {}".format(int(df_2025_spring.Count)))
print("Year-on-year applications in percent: {:.01f}%".format((df_2026_spring.Count - df_2025_spring.Count)/df_2025_spring.Count * 100))
Applications in the spring of 2026: 26561 Applications in the spring of 2025: 34282 Year-on-year applications in percent: -22.5%
TPO residential market share in 2026¶
With TPO on the rise, is the TPO market concentrating among a handful of large firms that have the tax appetite to use the credit, rather than local installers? To answer that, we'll look at third party owned applications in 2026 so far. There seems to be no standardized way that the Third Party Name column is entered into the DGStats database, so we'll have to do some cleaning in order to not create multiple groups for the same installer. Below, 2026 TPO applications are grouped by installer name, and the installers with the top ten share of the market are printed.
df_res['Third Party Name'] = df_res['Third Party Name'].str.strip()
df_res.loc[df_res['Third Party Name'].str.contains('sunrun', case=False, na=False), 'Third Party Name'] = 'Sunrun'
df_res.loc[df_res['Third Party Name'].str.contains('tesla', case=False, na=False), 'Third Party Name'] = 'Tesla'
df_res.loc[df_res['Third Party Name'].str.contains('goodleap', case=False, na=False), 'Third Party Name'] = 'GoodLeap'
df_res.loc[df_res['Third Party Name'].str.contains('enfin', case=False, na=False), 'Third Party Name'] = 'EnFin'
df_res.loc[df_res['Third Party Name'].str.contains('lightreach', case=False, na=False), 'Third Party Name'] = 'LightReach'
df_res.loc[df_res['Third Party Name'].str.contains('sunpower', case=False, na=False), 'Third Party Name'] = 'SunPower'
df_res.loc[df_res['Third Party Name'].str.contains('everbright', case=False, na=False), 'Third Party Name'] = 'Everbright'
df_res.loc[df_res['Third Party Name'].str.contains('lifestyle', case=False, na=False), 'Third Party Name'] = 'Lifestyle Solar'
df_res.loc[df_res['Third Party Name'].str.contains('axia', case=False, na=False), 'Third Party Name'] = 'Axia Solar'
df_res.loc[df_res['Third Party Name'].str.contains('participate', case=False, na=False), 'Third Party Name'] = 'Participate Energy'
df_res.loc[df_res['Third Party Name'].str.contains('palmetto', case=False, na=False), 'Third Party Name'] = 'Palmetto Solar'
df_2026 = df_res[df_res['App Received Date'].dt.year == 2026]
df_tpo_2026 = df_2026[df_2026['Third Party Owned'] == 'Yes']
market_share = (
df_tpo_2026.groupby('Third Party Name')
.size()
.sort_values(ascending=False)
.rename('count')
.to_frame()
)
market_share['share_%'] = (market_share['count'] / market_share['count'].sum() * 100).round(1)
print(market_share.head(10))
count share_% Third Party Name Sunrun 20039 60.9 GoodLeap 2832 8.6 GAF Energy 1530 4.6 EnFin 1196 3.6 LightReach 1134 3.4 Tesla 1064 3.2 Lifestyle Solar 563 1.7 GAF Energy LLC 457 1.4 Axia Solar 441 1.3 Participate Energy 266 0.8
It isn't even close. Sunrun alone has over 60% of the TPO market in California for applications received in 2026 (through August). Let's look at directly owned applications for that same time interval to see if there's a pattern of national firms vs local installers.
df_res['Installer Name'] = df_res['Installer Name'].str.strip()
df_res.loc[df_res['Installer Name'].str.contains('tesla', case=False, na=False), 'Installer Name'] = 'Tesla'
df_res.loc[df_res['Installer Name'].str.contains('sunrun', case=False, na=False), 'Installer Name'] = 'Sunrun'
df_res.loc[df_res['Installer Name'].str.contains('citadel', case=False, na=False), 'Installer Name'] = 'Citadel Roofing & Solar'
df_res.loc[df_res['Installer Name'].str.contains('g a f', case=False, na=False), 'Installer Name'] = 'GAF Energy'
df_res.loc[df_res['Installer Name'].str.contains('solara', case=False, na=False), 'Installer Name'] = 'Solara Home Energy'
df_res.loc[df_res['Installer Name'].str.contains('sunpower', case=False, na=False), 'Installer Name'] = 'SunPower Corporation'
df_2026 = df_res[df_res['App Received Date'].dt.year == 2026]
df_tpo_2026 = df_2026[df_2026['Third Party Owned'] == 'No']
market_share = (
df_tpo_2026.groupby('Installer Name')
.size()
.sort_values(ascending=False)
.rename('count')
.to_frame()
)
market_share['share_%'] = (market_share['count'] / market_share['count'].sum() * 100).round(1)
print(market_share.head(15))
count share_% Installer Name Citadel Roofing & Solar 3633 11.4 GAF Energy 3463 10.9 Sunrun 2773 8.7 Tesla 2075 6.5 SunPower Corporation 668 2.1 EMPOWER HOME SERVICES 601 1.9 MARS BUILDER SOLUTIONS 555 1.7 Solara Home Energy 489 1.5 LIFESTYLE SOLAR ELECTRIC 419 1.3 SOLARMAX RENEWABLE ENERGY PROVIDER INC 407 1.3 SOLAR NEGOTIATORS 308 1.0 DICK DONALD P AIR CONDITIONING INC 278 0.9 SOLAR TECH ENERGY SYSTEMS INC 269 0.8 FREEDOM FOREVER 245 0.8 LIFESTYLE SOLAR INC 243 0.8
The market of owned system is less concentrated. Of note is that Sunrun shows up even here, with an 8.7% share of the market. Sunrun is mostly known for TPO systems but also has a new-homes channel, selling to certain commercial developers through our multi-family and new homes offerings. In those deals, the builder buys the system and the homebuyer owns it with the house. Both would appear in DGStats as owned systems with Sunrun as the installer. So some of the 8.7% may be Title 24 new construction.
Notes on Methodology¶
Data. CPUC California DG Stats Interconnected Applications data set for PG&E, SCE and SDG&E, downloaded from californiadgstats.ca.gov, with completed applications received through August 31, 2026. We use the Applications data set, not the Interconnected Project Sites data set, because Project Sites keeps one record per site and assigns it the most recent application's details, which moves modified systems away from their original date.
Filters. Customer Sector == 'Residential' and System Size DC > 0. Technology Type contains “Photovoltaic” or “Solar PV”. All application statuses are kept, including decommissioned and superseded applications, because a system
that was later removed or modified was still an application when it was filed.
Ownership. An application is third-party owned (TPO) if Third Party Owned == 'Yes'
(lease or power purchase agreement) and customer-owned if 'No'.
Timing. Applications are counted by month of App Received Date. The TPO share is TPO
applications divided by all applications in the month; the smoothed line is a centered
three-month average of the monthly shares. DG Stats publishes an application only once it
is approved, and owned systems take longer to approve than TPO systems (2025 medians of 27
and 15 days), so recent months are incomplete and overstate the TPO share. The last three
months are shaded for that reason.
What is being measured. These are shares of applications, not of customer sites or new adopters. The three utilities record changes to an existing system differently:
- PG&E files a new application linked to the original through
Preceding Id. - SCE closes the original application (status
Expansion-…) and files a new, unlinked application that restates the existing system. - SDG&E files a separate application for each technology and each upgrade.
Follow-on applications therefore appear alongside first-time installations and cannot be removed consistently across utilities.
Robustness.
- Removing applications with a
Preceding Id(about 7% of applications, mostly PG&E) lowers the statewide TPO share by about 1 percentage point before April 2023 and by half a point or less in 2024 and 2025. Monthly application counts are 5–8% lower on this basis, with the same trends. - Including all application statuses, compared with only active ones, changes the share by at most 1.2 points in any month.
- For SCE, about one in six applications received since 2024 appears to restate an existing system, based on the number of originals closed in that period. Application counts after April 2023 therefore overstate new adoption, and the decline in new adopters is larger than the chart shows.