Verified
PG&E
Experimenting- Utility
- ~29,000 staff
- 700+ AI-enabled cameras
- Agent Assist for 900 agents
- Neutron at Diablo Canyon
- No AI-only return
PG&E deploys AI mostly in grid safety: AI-enabled wildfire cameras, machine learning weather and outage models, and a Continuous Monitoring Center that scans smart meter and sensor data for faults. PG&E runs generative AI in 2 named places, Agent Assist for its 900 contact center agents, built with AWS, and Atomic Canyon's Neutron document search for 1,300 Diablo Canyon workers. PG&E has published no return it credits to AI alone, and CFO Carolyn Burke said in Jul 2026 that AI cost savings had barely begun.
pge.comTheir approach to AI
- The platform
- PG&E builds its own machine learning models for weather, outages and grid monitoring, and buys generative AI from vendors: Agent Assist built with AWS, Atomic Canyon's Neutron at Diablo Canyon and a PolyAI voice agent on its phone lines.
- The models
- PG&E has named no general-purpose model provider; Neutron runs Atomic Canyon's FERMI models on NVIDIA GPUs at Diablo Canyon, and the weather, outage and smart meter models are PG&E's own.
- The reach
- PG&E has given no company-wide AI user count; Agent Assist reached all 900 contact center agents by May 2026 and Neutron all 1,300 Diablo Canyon workers by fall 2025, out of about 29,000 employees.
- The catch
- PG&E's one AI-related dollar figure, $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026, is credited to continuous monitoring, which joins sensor networks with machine learning, so no return is credited to AI alone.
Snapshot
- Employees
- ~29,000
- Utility regular employees at Dec 31, 2025; PG&E Corporation itself had 10
- Technology budget
- Not disclosed
- PG&E reported a $73B capital plan through 2030 (Jul 2026) but no technology or AI budget
- AI/ML headcount
- Not disclosedest.
- A Mar 2026 posting put about 30 people under the Senior Director, Artificial Intelligence
- Who owns AI
- Ajay Waghray, EVP and CIO
- David Leach, SVP and Chief Data & Analytics Officer, runs the central AI and analytics group
- Platform
- Hybrid
- In-house grid models; Agent Assist built with AWS; Neutron from Atomic Canyon
- Models
- In-house ML, Atomic Canyon FERMI
- No general-purpose model vendor named
- Reach
- 900 agents; 1,300 nuclear workers
- Agent Assist (May 2026) and Neutron (fall 2025); no company-wide count
- Use cases
- Wildfire, grid, customer, nuclear
- Cameras, weather and outage models, continuous monitoring, Agent Assist, Neutron
Key findings
PG&E has published no return it credits to AI alone: the $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026 is credited to continuous monitoring, which combines tens of thousands of grid sensors and about 5.5 million smart meters with machine learning (Jul 2026).
PG&E rolled out Agent Assist, a generative AI tool built with AWS, to all 900 of its contact center agents by May 2026, and Chief Customer Officer Vincent Davis said it cut call time by 4% and lowered costs without giving a dollar figure.
PG&E made Atomic Canyon's Neutron, a generative AI document search tool running on NVIDIA GPUs at the plant, available to all 1,300 Diablo Canyon workers by fall 2025, the first on-site generative AI deployment at a U.S. nuclear plant.
Patti Poppe said in Jul 2026 that PG&E's biggest use of AI is in meteorology, and PG&E had more than 700 AI-enabled wildfire cameras by Oct 2026 and about 1,400 AI- or machine learning-enabled weather stations by Jun 2025.
CFO Carolyn Burke named AI as one of 2 main sources of further O&M savings on Jul 23, 2026, saying PG&E had only begun to apply AI to the way it works; PG&E's earnings decks group AI with planning, execution and automation in one O&M savings category.
Where PG&E sits
- Stage
- Experimenting
PG&E has reported no business return it credits to AI alone since May 2025. PG&E's largest figure, $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026 (Jul 2026 earnings release), is credited to continuous monitoring capabilities that combine tens of thousands of grid sensors with machine learning, with no share for AI; Agent Assist's 4% shorter calls (May 2026) is a task speedup with no cost figure; Neutron's safety-valve document search falling from about 180 days to 40 (Mar 2026) is one task timing; and the earnings decks credit O&M savings to planning, execution, automation and AI together. PG&E has not reported a company-wide rollout with a user count either: Agent Assist covers 900 contact center agents and Neutron 1,300 Diablo Canyon workers, out of about 29,000 employees. PG&E has recent AI work on the public record, so it sits at Experimenting; an AI-only return in one use case, or a company-wide AI tool with a stated user count under a named owner, would move it to Piloting.
How the stages are defined - Ownership
- EVP and CIO Ajay Waghray; SVP and Chief Data & Analytics Officer David Leach runs the central AI team
- Platform
- Hybrid: In-house grid models; vendor generative AI (AWS, Atomic Canyon, PolyAI)
- Models
- In-house machine learning models, Atomic Canyon FERMI models; no general model vendor named
- Hosting
- AWS for contact center AI; Neutron on on-site NVIDIA GPUs at Diablo Canyon; other hosting not disclosed
- Reach
- 900 contact center agents and 1,300 Diablo Canyon workers; no company-wide count
How they deploy
AI for wildfire safety first
PG&E's AI work started where its largest risk sits: wildfires caused by its equipment. PG&E's 2024 R&D strategy report, The AI-Enabled Utility (Nov 2024), says most early AI efforts focused on safety, citing computer vision for inspections, transformer failure prediction and machine learning models that forecast wildfire risk about 5.5 days ahead. Patti Poppe said on Jul 23, 2026 that PG&E's biggest use of AI is in meteorology and fire prediction.
Trusted data, then predictive operations
Ajay Waghray, PG&E's EVP and CIO, described a 3-pillar AI strategy of foundation, democratization and execution in a Sep 2026 interview with Metis Strategy, built on trusted asset, weather, sensor and smart meter data. Waghray said cybersecurity, responsible AI and standardized processes have to advance together as PG&E moves from reactive to predictive operations.
AI as an O&M cost lever
PG&E targets 2% to 4% annual cuts in non-fuel O&M costs to keep customer bill growth at 0% to 3% a year. CFO Carolyn Burke said on Jul 23, 2026 that strategic sourcing and AI are the 2 areas with the most room for further savings, and that PG&E had only begun to apply AI to how work gets done. PG&E's Q2 and Q3 2026 earnings decks list planning, execution, automation and AI as one category of O&M savings, without an AI share.
A central AI team in a hub and spoke model
PG&E's Artificial Intelligence and Analytics Insights group works as a central hub that partners with business teams, according to a Mar 2026 posting for its Senior Director, Artificial Intelligence. The posting gives that role PG&E's AI Transformation (AIX) strategy across advanced analytics, generative AI and agentic AI, the job of moving solutions from pilots to scaled adoption, and AI platforms in cloud environments.
Buy generative AI, keep sensitive data close
PG&E has bought its generative AI tools instead of building models: Agent Assist with Amazon Web Services, a PolyAI voice agent on its phone lines, and Atomic Canyon's Neutron at Diablo Canyon, where the tool runs on site on NVIDIA hardware and reaches plant document systems without the cloud, according to Business Insider (Mar 2026).
- Deployment model
- PG&E runs AI through a central data, analytics and AI organization under SVP and Chief Data & Analytics Officer David Leach, which a Mar 2026 posting describes as a hub and spoke center of excellence, while business units own their deployments: Wildfire, Emergency and Operations runs the Continuous Monitoring Center and the Hazard Awareness and Warning Center, Customer Experience runs Agent Assist and Call Tagging, and Diablo Canyon runs Neutron. EVP and CIO Ajay Waghray leads PG&E's technology and security organization and sets the AI strategy. New generative AI uses pass an intake and review process that aligns cybersecurity, legal, data and business teams (Oct 2026).
- Model providers
- PG&E has not named a general-purpose large language model provider such as OpenAI, Anthropic or Google. Neutron at Diablo Canyon uses Atomic Canyon's FERMI family of nuclear-specific models, developed with Oak Ridge National Laboratory and described by World Nuclear News as open source, served with NVIDIA AI Enterprise and Triton Inference Server software (Nov 2024). Agent Assist was built with Amazon Web Services, and PG&E has not said which model it uses. PG&E's weather, outage, fire-risk, transformer and smart meter models are machine learning models it built itself, according to its 2024 R&D strategy report.
- Data and infrastructure
- PG&E's grid AI draws on about 5.5 million smart meters and tens of thousands of sensors, including early fault detection sensors on 900 circuit miles, distribution fault anticipation sensors on 8,900+ circuit miles and line sensors on 19,000+ circuit miles, joined in a Grid Data Analytics Platform (May 2026). PG&E has used Palantir Foundry as an operational data platform since 2021, and its 2024 R&D report describes joining asset data in Foundry with climate data in AWS. Neutron indexes Diablo Canyon records spread across 6 on-site systems, including about 9,000 operating and maintenance procedures, on NVIDIA Hopper GPUs at the plant. PG&E's 2025 Form 10-K warns that generative AI may be used to carry out cyberattacks on its systems.
Platforms and internal tools
Continuous Monitoring CenterMay 1, 2026
The Continuous Monitoring Center in San Ramon is PG&E's hub where machine learning models scan smart meter and grid sensor data for patterns that precede equipment failures, outages or fires, and a trained team reviews each flag before sending a crew. PG&E said its continuous monitoring capabilities intercepted 17 potential ignitions, avoided 12 million minutes of outages and saved about $6 million in operating costs in 2025, before the center opened.
SmartDetect
SmartDetect applies machine learning to PG&E SmartMeter data to monitor grid performance and flag anomalies that may signal emerging risks; it feeds the Continuous Monitoring Center (May 2026). Patti Poppe said in Jul 2026 that smart meter signals PG&E previously did not use now reveal faults on service lines to homes.
Hazard Awareness and Warning Center
The Hazard Awareness and Warning Center is PG&E's 24/7 operations center for wildfire and other hazards, with real-time visibility through more than 700 AI-enabled cameras and about 1,600 weather stations (Oct 2026).
Agent Assist and Call TaggingCall Tagging late 2025; Agent Assist Apr 2026
Agent Assist gives PG&E customer service representatives real-time guidance and suggested responses during calls, and Call Tagging labels and sorts calls by content for analysis. PG&E chose Amazon Web Services to build Agent Assist after evaluating several partners.
- Users
- All 900 call center agents (May 2026)
NeutronPilot in the first half of 2024; plant-wide by fall 2025
Neutron is Atomic Canyon's generative AI search and summary tool installed at Diablo Canyon, using optical character recognition and retrieval-augmented generation to find operating procedures, engineering design changes and regulatory requirements. PG&E announced the deployment in Nov 2024 as the first on-site generative AI deployment at a U.S. nuclear plant.
- Users
- 1,300 Diablo Canyon workers (fall 2025)
- Models
- Atomic Canyon FERMI models on NVIDIA Hopper GPUs
Sherlock & Waldo
Sherlock & Waldo are tools PG&E's system inspections team built so desktop inspectors can view and mark equipment problems on drone and helicopter images; the images also train machine learning models that identify grid components. PG&E said in Nov 2024 that the tools save 7,500 inspector hours a year.
Palantir FoundryJan 2021
Palantir Foundry is the operational data platform PG&E began deploying in 2021, first to support Public Safety Power Shutoff decisions and later asset risk and outage investigation; PG&E's 2024 R&D report still describes asset data held in Foundry.
Vendors and partners
Amazon Web ServicesBy 2024
Cloud and build partner for Agent Assist; climate data platform · PG&E said in May 2026 that it chose AWS for Agent Assist for its ability to help build, scale and support the tool, and its 2024 R&D report describes climate data held in AWS.
Atomic CanyonNov 2024
Neutron generative AI document search at Diablo Canyon · Atomic Canyon's FERMI models were developed with the Department of Energy's Oak Ridge National Laboratory; financial terms were not disclosed.
NVIDIANov 2024
GPUs and inference software under Neutron · Neutron runs on NVIDIA AI Enterprise, Triton Inference Server and Hopper GPUs at Diablo Canyon.
ALERTWest
AI-enabled wildfire camera network · PG&E's 2024 R&D report says PG&E deployed more than 600 AI-enabled cameras through its ALERTWest partnership; the cameras send automated alerts when they detect smoke or fire.
PolyAI
Generative AI voice agent on PG&E's customer phone lines · PolyAI's undated case study says its agent handles authentication, routing, outage updates, billing and FAQs in English and Spanish; PG&E has not published figures for it.
PalantirJan 2021
Foundry operational data platform · Palantir described the 2021 contract as multi-year and worth millions of dollars; PG&E has not published its current scope.
XPRIZE Wildfire
PG&E sponsors the competition for AI and autonomous wildfire detection and suppression · PG&E showcased 2 XPRIZE Wildfire teams at its Jun 2025 wildfire readiness event.
Leadership
AI, data and platform leaders
Executives
AI org map
Central data and AI center of excellence under the CDAO; deployments owned by operating units
PG&E's AI strategy sits with EVP and CIO Ajay Waghray, who leads the technology and security organization, and its central Artificial Intelligence and Analytics Insights group reports through SVP and Chief Data & Analytics Officer David Leach, who is also Chief Security Officer. PG&E's operating units run the deployments: the Continuous Monitoring Center under Wildfire, Emergency and Operations, Agent Assist under Customer Experience, and Neutron at Diablo Canyon. PG&E reorganized its leadership on Jan 1, 2026, making Sumeet Singh CEO of the utility over a combined energy delivery organization.
PG&E has not published how many people work on AI. PG&E's central Artificial Intelligence and Analytics Insights group, a center of excellence for data scientists, machine learning engineers and AI product managers, had about 30 people under its Senior Director, Artificial Intelligence, including 2 directors, according to a Mar 2026 posting. Other AI work sits in operating teams, such as the Continuous Monitoring Center's engineers and PG&E's meteorologists and data scientists who run the fire-weather models.
Reporting lines
CEO, PG&E Corporation
CEO, Pacific Gas and Electric Company, and Energy Delivery
Sumeet Singh
Operations, engineering and shared services since Jan 1, 2026; reporting line inferred, not published
Wildfire, Emergency and Operations
Mark Quinlan, SVP
Reporting line inferred, not published
Continuous Monitoring Center
Craig Kurtz, Senior Director
Reporting line inferred, not published
Technology and Security
Ajay Waghray, EVP and CIO
Officer of both PG&E Corporation and the utility; reporting line not published
Data and Analytics, and Security
David Leach, SVP, CSO and CDAO
Inferred from Waghray leading the technology and security organization; not published
Artificial Intelligence and Analytics Insights
Senior Director, Artificial Intelligence
Hub and spoke AI center of excellence; about 30 people (Mar 2026 posting)
Customer Experience
Vincent Davis, SVP and Chief Customer Officer
Reporting line not published
Strategy and Growth
Jason Glickman, EVP
Long-term planning since Jan 2026; reporting line not published
PG&E's 2025 Form 10-K and 2025 annual report list Ajay Waghray as EVP and CIO and David Leach as SVP, Chief Security Officer and Chief Data and Analytics Officer, but PG&E has not published who either reports to. The Senior Director, Artificial Intelligence reports to the SVP and Chief Data and Analytics Officer according to a Mar 2026 job posting; PG&E has not named who holds the role.
Named people
Use cases
PG&E runs AI in production in wildfire detection (AI-enabled cameras), meteorology and outage prediction, continuous grid monitoring with machine learning on smart meter and sensor data, asset inspection, contact center Agent Assist and Call Tagging, a PolyAI voice agent, and Neutron document search at Diablo Canyon. PG&E has published no return it credits to AI alone for any of them: its largest figure, $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026, belongs to continuous monitoring as a whole, and PG&E reports no firm-level AI figure.
| Use case | Area | Status | ROI |
|---|---|---|---|
| AI-enabled wildfire cameras PG&E's high-definition wildfire cameras, deployed with ALERTWest, use AI to detect smoke and send automated wildfire alerts within seconds. PG&E had more than 650 cameras in Jun 2025 and more than 700 AI-enabled cameras by Oct 2026, after testing AI on 46 cameras in 2021. | Wildfire safety | In production | |
| Machine learning weather, fire-risk and outage prediction PG&E meteorologists and data scientists use machine learning models on data from about 1,600 weather stations, about 1,400 of them AI- or machine learning-enabled (Jun 2025), to forecast fire risk about 5.5 days ahead and target Public Safety Power Shutoffs more narrowly. Patti Poppe called meteorology PG&E's biggest use of AI in Jul 2026. | Wildfire safety | In production | |
| Continuous monitoring and SmartDetect PG&E's Continuous Monitoring Center runs machine learning models over about 5.5 million smart meters and tens of thousands of grid sensors to catch failing equipment before it sparks a fire or an outage. PG&E credits its continuous monitoring capabilities, sensors and models together, with 28 avoided ignitions, 19.6 million avoided outage minutes and $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026, and gives no share to machine learning. | Electric operations | In production | |
| Inspection imagery and transformer failure prediction PG&E's Sherlock & Waldo tools let desktop inspectors mark problems on drone and helicopter images that also train machine learning models to classify grid components, which PG&E said save 7,500 inspector hours a year (Nov 2024). PG&E also runs machine learning models that estimate whether a transformer will fail within 30 days, and its 2026 to 2028 wildfire plan adds machine learning models of asset health. | Electric operations | In production | |
| Agent Assist and Call Tagging Agent Assist, built with AWS and rolled out in Apr 2026, gives PG&E customer service representatives real-time guidance and suggested answers, so they no longer search hundreds of procedures by hand; Call Tagging, launched in late 2025, labels calls by content. Vincent Davis said in May 2026 that Agent Assist cut call time by 4% and lowered costs, without a cost figure. | Customer service | In production | |
| PolyAI voice agent PolyAI's voice agent, called Peggy, answers PG&E customer calls on outages, billing and FAQs in English and Spanish. PolyAI's undated case study claims 35,000 labor hours saved and 67% call containment; PG&E has not confirmed those figures. | Customer service | In production | |
| Neutron document search at Diablo Canyon Neutron lets Diablo Canyon staff search and summarize operating procedures, design changes and regulatory requirements spread across 6 on-site systems. Maureen Zawalick told Business Insider in Mar 2026 that a 12-person team gathered the documents for a safety-valve investigation in about 40 days, against about 180 days before. | Nuclear generation | In production | |
| Underground pipe detection and ticket duration PG&E described machine learning tools that help locate underground gas pipes and estimate how long underground service tickets will take, among AI pilots in its Nov 2024 R&D report. | Gas operations | Pilot |
Adoption
PG&E has not reported how many of its about 29,000 employees use AI tools, and it has announced no company-wide generative AI assistant. PG&E's named generative AI deployments reach 2 groups: all 900 contact center agents use Agent Assist (May 2026) and all 1,300 Diablo Canyon workers have Neutron (fall 2025). PG&E's machine learning runs mostly inside operations, where meteorologists, wildfire analysts and Continuous Monitoring Center engineers use model output, and a governance framework for new generative AI uses was in place by Oct 2026.
By the numbers
Rollout, first pilot to today
Jul 2025 to fall 2025
PG&E rolled Neutron out in phases with training sessions until all 1,300 Diablo Canyon workers had access.
Apr 2026 to May 2026
PG&E rolled out Agent Assist, built with AWS, and extended it to all 900 contact center agents.
Jan 2021
PG&E began deploying Palantir Foundry as an operational data platform, first for Public Safety Power Shutoff decisions.
Nov 2021
PG&E began testing AI on 46 of its wildfire cameras to tell smoke from fog and other false signals.
First half of 2024
PG&E piloted Neutron at Diablo Canyon with a small group from engineering, maintenance and operations.
Nov 13, 2024
PG&E published its 2024 R&D strategy report, The AI-Enabled Utility, and announced Neutron at Diablo Canyon with Atomic Canyon and NVIDIA.
Late 2025
PG&E launched Call Tagging, which labels contact center calls by content.
Jun 6, 2025
PG&E said about 1,400 of its nearly 1,600 weather stations were AI- or machine learning-enabled and that its 650+ AI-enabled cameras send automated wildfire alerts.
Oct 23, 2025
Patti Poppe said a newly deployed AI-enabled machine learning model, with low-cost sensors and smart meters, enables system-wide monitoring of PG&E's secondary system.
May 1, 2026
PG&E opened the Continuous Monitoring Center in San Ramon, bringing its machine learning and sensor monitoring under one roof.
Economics & controls
Spend and ROI
Technology budget
PG&E does not report a technology or AI budget. PG&E's capital plan stood at $73 billion through 2030 in Jul 2026, mostly grid investment, and the company targets 2% to 4% annual cuts in non-fuel O&M costs; PG&E's Q3 2026 deck shows non-fuel O&M savings of $220 million, about 2.5%, in 2025.
What AI itself costs
PG&E has not disclosed what it spends on AI models, licenses or compute, or the value of its contracts with AWS, Atomic Canyon, PolyAI, ALERTWest or Palantir.
Reported ROI
PG&E has reported no return it credits to AI alone. PG&E's continuous monitoring capabilities, which pair sensor networks with machine learning, saved about $6 million in operating costs in 2025 (May 2026), an estimated $8 million of capital and more than $1 million of expense over the 5 quarters to Mar 2026 (Apr 2026), and $11.2 million through lower-cost repairs from Jan 2025 to Jun 2026 (Jul 2026); Patti Poppe described continuous monitoring in Apr 2026 as sensors, smart meters, analytics and machine learning together, and Ajay Waghray said in Sep 2026 that PG&E's data and AI capabilities had helped avoid 19 million outage minutes, prevent 28 potential ignitions and remove more than 5,000 emergency response hours. PG&E's earnings decks credit part of its O&M savings to planning, execution, automation and AI together, with no AI share, and Carolyn Burke said in Apr 2026 that satellite and LiDAR inspections are expected to save $24 million of O&M in 2026, a forecast credited to technology, not AI.
Productivity markers
PG&E reports task-level gains only: Agent Assist cut call time by 4% (May 2026); Neutron cut document gathering for one safety-valve investigation from about 180 days to 40 (Mar 2026); continuous monitoring cut emergency outage response time by 2,620 hours in 2025 (May 2026); and Sherlock & Waldo save 7,500 inspector hours a year (Nov 2024). CFO Carolyn Burke said on Jul 23, 2026 that PG&E had only begun to apply AI to its work.
Headcount
PG&E had about 29,000 regular utility employees at Dec 31, 2025, about 60% of them covered by union agreements, and a 3.8% turnover rate in 2025. PG&E has made no statement linking AI to headcount.
Governance and risk
Generative AI intake and review
PG&E built a governance framework for generative AI that defines lifecycle stages and standardizes intake and review, aligning cybersecurity, legal, data and business teams; PG&E presented it at Utility Analytics Week on Oct 22, 2026. A Mar 2026 posting refers to PG&E's Generative AI Acceptable Use Policy, maintained with Legal, Risk, Privacy and Cybersecurity.
People review machine flags
PG&E's Continuous Monitoring Center keeps a trained team between each machine learning flag and a crew dispatch: the team reviews what the models find and decides whether to send someone (May 2026).
Generative AI at a nuclear plant
PG&E limits Neutron at Diablo Canyon to finding and summarizing documents, runs it on site and connects it to plant systems without the cloud, according to Business Insider (Mar 2026). Critics quoted by Business Insider said generative AI in a nuclear setting deserves more scrutiny and questioned whether its use would stay limited.
Cybersecurity oversight
PG&E's boards oversee cybersecurity through their Safety and Nuclear Oversight Committees, and the EVP and CIO and the SVP, Chief Security Officer and Chief Data and Analytics Officer manage the risk with enterprise risk management. PG&E's 2025 Form 10-K warns that AI, including generative AI, may be used to facilitate attacks on its systems.
Regulators
California's Office of Energy Infrastructure Safety approved PG&E's 2026 to 2028 wildfire mitigation base plan on Feb 5, 2026, noting PG&E's plan to use machine learning models to judge asset health. PG&E's 2024 R&D report says AI is a target of new legislation and that PG&E is setting up compliance and legal processes for it.
Recent signals
Oct 23, 2026
Patti Poppe told Fortune that AI is the best thing that could have happened to the grid because data center demand brings investment, a day after PG&E's third-quarter results missed forecasts.
Oct 22, 2026
Patti Poppe said on the Q3 2026 call that PG&E's Hazard Awareness and Warning Center sees its system through more than 700 AI-enabled cameras and 1,600 weather stations; the Q3 deck credits continuous monitoring with 35 avoided ignitions.
Oct 22, 2026
PG&E presented its generative AI governance framework, from intake to responsible adoption, at Utility Analytics Week.
Sep 10, 2026
Ajay Waghray described PG&E's 3-pillar AI strategy of foundation, democratization and execution, and said its data and AI capabilities had helped avoid 19 million outage minutes and 28 potential ignitions.
Jul 23, 2026
PG&E reported that continuous monitoring saved $11.2 million through lower-cost repairs from Jan 2025 to Jun 2026, and CFO Carolyn Burke named AI as one of 2 main sources of further O&M savings.
May 22, 2026
Vincent Davis wrote that Agent Assist, built with AWS, had reached all 900 PG&E contact center agents and cut call time by 4%.
May 1, 2026
PG&E opened its Continuous Monitoring Center in San Ramon, where machine learning models scan smart meter and sensor data for early signs of faults.
Apr 23, 2026
Patti Poppe said on the Q1 2026 call that continuous monitoring, which joins sensors, smart meters, analytics and machine learning, saved an estimated $8 million of capital and more than $1 million of expense over 5 quarters.
Mar 2026
Business Insider reported that Neutron is available to all 1,300 Diablo Canyon workers and cut document gathering for one safety-valve investigation from about 180 days to 40.
Mar 2026
PG&E advertised for a Senior Director, Artificial Intelligence to lead its AI Transformation (AIX) strategy across analytics, generative AI and agentic AI, reporting to the Chief Data and Analytics Officer.
Feb 5, 2026
California's Office of Energy Infrastructure Safety approved PG&E's 2026 to 2028 wildfire mitigation base plan, which adds machine learning models of asset health.
Dec 17, 2025
PG&E announced a leadership reorganization effective Jan 1, 2026, naming Sumeet Singh CEO of the utility and Vincent Davis chief customer officer.
Nov 7, 2025
Jason Glickman told GridFWD that more than 630 PG&E cameras monitor 90% of high fire-risk areas with AI smoke detection and that machine learning weather models predict outages and fire risk.
Quick answers
- How does PG&E deploy AI?
- PG&E deploys AI mostly in grid safety: AI-enabled wildfire cameras, machine learning weather and outage models, and a Continuous Monitoring Center that scans smart meter and sensor data for faults. PG&E runs generative AI in 2 named places, Agent Assist for its 900 contact center agents, built with AWS, and Atomic Canyon's Neutron document search for 1,300 Diablo Canyon workers. PG&E has published no return it credits to AI alone, and CFO Carolyn Burke said in Jul 2026 that AI cost savings had barely begun.
- What does PG&E use AI for?
- PG&E runs AI in production in wildfire detection (AI-enabled cameras), meteorology and outage prediction, continuous grid monitoring with machine learning on smart meter and sensor data, asset inspection, contact center Agent Assist and Call Tagging, a PolyAI voice agent, and Neutron document search at Diablo Canyon. PG&E has published no return it credits to AI alone for any of them: its largest figure, $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026, belongs to continuous monitoring as a whole, and PG&E reports no firm-level AI figure.
- Who leads AI at PG&E?
- AI at PG&E is led by Ajay Waghray (EVP and Chief Information Officer, PG&E Corporation and Utility); David Leach (SVP, Chief Security Officer and Chief Data & Analytics Officer, Pacific Gas and Electric Company). Ownership of AI on the org chart: EVP and CIO Ajay Waghray; SVP and Chief Data & Analytics Officer David Leach runs the central AI team.
- What AI platform does PG&E use?
- PG&E builds its own machine learning models for weather, outages and grid monitoring, and buys generative AI from vendors: Agent Assist built with AWS, Atomic Canyon's Neutron at Diablo Canyon and a PolyAI voice agent on its phone lines.
- Which AI models does PG&E use?
- PG&E has named no general-purpose model provider; Neutron runs Atomic Canyon's FERMI models on NVIDIA GPUs at Diablo Canyon, and the weather, outage and smart meter models are PG&E's own.
- How many PG&E employees use AI?
- PG&E has given no company-wide AI user count; Agent Assist reached all 900 contact center agents by May 2026 and Neutron all 1,300 Diablo Canyon workers by fall 2025, out of about 29,000 employees.
- What return has PG&E reported from AI?
- PG&E has reported no business return it credits to AI alone since May 2025. PG&E's largest figure, $11.2 million saved through lower-cost repairs from Jan 2025 to Jun 2026 (Jul 2026 earnings release), is credited to continuous monitoring capabilities that combine tens of thousands of grid sensors with machine learning, with no share for AI; Agent Assist's 4% shorter calls (May 2026) is a task speedup with no cost figure; Neutron's safety-valve document search falling from about 180 days to 40 (Mar 2026) is one task timing; and the earnings decks credit O&M savings to planning, execution, automation and AI together. PG&E has not reported a company-wide rollout with a user count either: Agent Assist covers 900 contact center agents and Neutron 1,300 Diablo Canyon workers, out of about 29,000 employees. PG&E has recent AI work on the public record, so it sits at Experimenting; an AI-only return in one use case, or a company-wide AI tool with a stated user count under a named owner, would move it to Piloting.
Open questions
- How much of the $11.2 million that continuous monitoring saved from Jan 2025 to Jun 2026 comes from machine learning rather than the sensor networks it reads?
- What share of PG&E's O&M savings in the planning, execution, automation and AI category comes from AI?
- What has Agent Assist saved in dollars or staffing, beyond the 4% cut in call time PG&E reported in May 2026?
- Does PG&E give employees a general-purpose generative AI assistant, and how many use it? PG&E has announced none.
- Which models power Agent Assist on AWS, and does PG&E use any OpenAI, Anthropic, Google or Microsoft models?
- Who leads PG&E's Artificial Intelligence and Analytics Insights group, and whom do Ajay Waghray and David Leach report to?
- Are PolyAI's claims of 35,000 labor hours saved and 67% containment for PG&E current, and does PG&E stand behind them?
- What does PG&E spend on AI, and does it still run Palantir Foundry at the scope described in 2021 and 2024?
Sources
PG&E
- Customer at the heart, powered by AI, Vincent Davis (May 22, 2026)
- How PG&E is catching wildfire risks before they start (May 2026)
- PG&E using machine learning to build a smarter grid, GridFWD (Nov 7, 2025)
- The science behind wildfire safety (Jun 20, 2025)
- 2024 R&D Strategy Report: The AI-Enabled Utility (Nov 2024)
- Artificial intelligence takes center stage at PG&E Innovation Summit (Nov 14, 2024)
- Senior Director, Artificial Intelligence job posting (Mar 2026)
Filings, releases and investor materials
- PG&E 2025 annual report and Form 10-K
- Form 8-K on leadership changes (Dec 17, 2025)
- Q2 2026 earnings release (Jul 23, 2026)
- Q2 2026 earnings presentation (Jul 23, 2026)
- Q3 2026 earnings release (Oct 22, 2026)
- Q3 2026 earnings presentation (Oct 22, 2026)
- Q2 2026 earnings call transcript (Jul 23, 2026)
- Q3 2026 earnings call transcript (Oct 22, 2026)
- Q1 2026 earnings call transcript (Apr 23, 2026)
- Q3 2025 earnings call transcript (Oct 23, 2025)
- Continuous Monitoring Center release (May 1, 2026)
- 2025 wildfire season readiness release (Jun 6, 2025)
Partners, vendors and regulators
- Neutron at Diablo Canyon, PG&E and Atomic Canyon release (Nov 13, 2024)
- PolyAI case study: Pacific Gas and Electric (undated)
- Palantir and PG&E Foundry announcement (Jan 2021)
- PG&E testing AI on wildfire cameras (Nov 2021)
- Energy Safety approves PG&E's 2026 to 2028 wildfire mitigation plan (Feb 5, 2026)
- Utility Analytics Week: Governing Generative AI at PG&E (Oct 22, 2026)
News and interviews
- Business Insider: Diablo Canyon turns to AI for its paperwork (Mar 2026)
- World Nuclear News: AI solution deployed at Diablo Canyon (Nov 2024)
- Metis Strategy: Ajay Waghray on trusted data for predictive AI (Sep 10, 2026)
- Fortune: Patti Poppe on AI and the grid (Oct 23, 2026)
- T&D World: PG&E leadership changes (Dec 2025)
- AIM: David Leach appointed SVP, CSO and CDAO
- IoT M2M Council: PG&E monitoring centre (May 2026)
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
PG&E profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 5, 2026
