Chevron

An inspection round on a Chevron wellpad in western Colorado used to mean a technician on the road, because the reading had to be taken in-person standing next to the equipment. Russell Robinson, a program manager on Chevron's Facilities and Operations of the Future team, points to those drive times, and to what they look like in snow and ice, as the reason for putting autonomous drones over the wellpads. 

A quadcopter now flies the round from a dock that charges it and uploads its data between missions, launched from Houston, West Texas or Percepto's operations center in Florida. That is one inspection round in one basin, and it is the kind of change Chevron has learned to put a number against. 

Ten months ago the company made a much larger claim. On 12 November 2025 Chevron put its five-year plan in front of the market, and AI ran through the whole session. That investor day remains Chevron's fullest public account of its AI program. 

Chairman and CEO Mike Wirth told the room the company expects AI to enable the next leg in this journey, naming efficiency, asset productivity, optimization and exploration outcomes. Mark Nelson, Vice Chairman of Chevron, went further and said Chevron is already seeing AI generate value today, forecasting billions in annual value beyond current forecasts by the end of the decade. 

For scale, this is the company making it.

Billions a year is a large claim to make to analysts who will hold you to it. Here is what Chevron has put on the table underneath it.

Two days before that investor day, Chevron published the clearest numbers it has ever given on AI-assisted work. Since 2024 its robotics initiatives have saved the company more than $92 million while eliminating more than 143,000 at-risk hours. Two programs carry itemized figures inside that total. 

Using robots for tank inspections has saved more than $25 million and approximately 43,000 working hours since 2024. Using robots to clean tanks has saved more than $6 million while cutting approximately 28,000 working hours. Both figures come from Chevron's own newsroom, quoting Trey Caylor, then the robotics program manager at its Technical Center.

At the investor day itself, Vice Chairman Mark Nelson added two more. AI maintenance planning tools in Australia are reducing critical task planning time by 25%. Digital twins have, in his words, "already realized millions of dollars in value", with no further figure attached. Caterpillar, which we covered in August, does attach one: a single turbocharger digital twin that surfaced a $7 million sales opportunity.

One thing links those four numbers. Each covers a defined piece of work that carried a duration or a cost before AI touched it, which is what allows a figure to be attached at all. Nothing about that test requires a refinery.

Start underground, where Chevron has been at this longest. Nelson told investors that AI is "embedded across our workflows and growing in impact", from seismic processing through to inventory characterization, and he named APOLO, which in his words "uses multivariate machine learning to analyze data from more than 50 thousand wells" to map basin prospectivity. The compute underneath it goes back to 2008, when Chevron first put GPUs on seismic processing.

Above ground sit the robots and the drones. Tank inspection and cleaning robots carry the two dollar figures above. The drone program ran in the Permian Basin and western Colorado with optical gas imaging and RGB cameras on each aircraft, and the layer feeding all of it is Azure IoT Operations on Azure Arc, pulling cameras, sensors, robots and drones at remote sites into a single control plane. 

In June 2026 Chevron and DataRobot announced work to move from robots that execute an approved mission to agents that assess conditions continuously and judge whether the mission should proceed. Chevron and DataRobot describe it as work in progress rather than a finished rollout.

Maintenance is the quieter half of the same story. On the second quarter earnings call on 31 July 2026, CFO Eimear Bonner described early-warning maintenance already deployed and delivering results, catching anomalies earlier than a scheduled check would, and named real-time facility optimization as part of a newer set of maintenance and reliability approaches in the shale and tight business.

Then the desks. Bonner's finance organization runs to about 3,500 people, and she has cataloged roughly 15 enterprise AI workflows across it, covering investor relations research, audit work supporting SOX controls and forecasting, on Microsoft Copilot and Anthropic's Claude. 

The example Nelson chose to give investors was smaller and more telling than any of that. A permit application used to mean finding the information, sifting it, then writing the thing. Now, in Nelson's account of a well plug-and-abandon job in Australia, the team in Chevron's India engineering center "had the permit application ready and all orders placed by morning".

Chevron has had a dedicated AI team since 2023, and Gustavson told investors that "well over half of our employees have access" to current tools. Chevron joined the Responsible AI Institute in July 2023, while the work was still mostly research. 

A responsible AI board reviewing hundreds of use cases a month, as UnitedHealth runs, is the version of this that comes later. The cost program starts counting from a 2024 baseline, and at that same November 2025 investor day Bonner raised the target from $2-3 billion to $3-4 billion of annual run-rate savings by the end of 2026. The published numbers arrived in the same week.

Look closely at the sequence. Chevron discussed these programs well after the field work started, and it has published nothing it tried and abandoned, which is true of almost every enterprise AI disclosure.

Chevron notes that well over half of its workforce has access to AI tools, but no figure sits beside that sentence. Every number Chevron breaks down sits further down, at the level of a tank, a maintenance plan, a permit. The $92 million total is published but never itemized past $31 million of it. 

That gap is the most useful thing in the disclosure for anyone running a rollout. Access is easy to report and hard to value. A tank inspection had a duration and a cost before the robot arrived, so the saving is arithmetic. Tools spread across half a workforce have no such baseline unless somebody took one before the tools landed. 

The structural cost program is the frame around all of it. By the end of the second quarter Chevron had hit $3 billion of annual run-rate savings since 2024, six months early against its end-2026 deadline, reported with its results on 31 July. 

The $3-4 billion target is company-wide. It covers restructuring, supply chain and capital efficiency, and Chevron has never broken out how much of it AI is responsible for. What Wirth did say is that the company is "leaning in" on technology and AI to do its work safer, faster and at a lower cost. AI is named as a lever rather than as the number.

Chevron has never said what the AI work cost. The savings figures are gross, with no investment figure beside them, which is the one number an executive would most want before copying any of it. UnitedHealth is the counter-example here: it has earmarked $1.5 billion for AI in 2026 and runs more than a thousand use cases in production, with the spend stated out loud. 

You do not operate a Permian shale position or a refinery network. Five of these transfer anyway.