Chevron says AI is worth hundreds of millions in 2026. Exploration cycle time is down 50%

Chevron's head of investor relations just put a price and a speed on the company's AI work. Jeanine Wai told TD Cowen's Jason Gabelman that 2026 is "the inflection," that Chevron is capturing "hundreds of millions of dollars of real value" this year, and that AI tools are cutting exploration cycle time by 50%.

The podcast was recorded September 24 at TD Cowen's energy conference in Austin. Wai said Chevron has been building AI capability for years, then named three places the company is going hardest: exploration, shale and tight, and commercial.

On exploration, she said the tools reduce cycle time by half by assessing "millions and millions of files" that used to take someone weeks or months to comb through. Chevron has separately described ApEX, an in-house generative AI platform launched in August 2024, as searching more than a million files for explorationists in the Gulf of America.

What she actually said

"We're capturing real value, hundreds of millions of dollars of real value in 2026, and that's only gonna increase as we scale across the organization," Wai said.

"For AI, the tools, they help to reduce the exploration cycle time by 50%," she said. That cut, in her telling, comes from putting those files "right at their fingertips" instead of leaving someone to comb them by hand.

On shale and tight, she pointed at the Permian: about a million barrels of oil equivalent a day, and working-interest insight into one out of every five wells in the basin. In the second quarter, Chevron said Permian CapEx per BOE improved 25% year over year, with capital expected below $3.5 billion. Wai said that improvement came from sweating the assets "but also with the AI and the technology overlay."

She did not give a single number for how much of the hundreds of millions comes from exploration versus the Permian versus commercial. She did not say what share of the 25% CapEx improvement is AI versus drilling and completion efficiency.

Why the file search is the easy half

Cutting prospect evaluation in half by reading what is already on disk is a real win. It is also the part of the job that leaves a trail. The weeks that used to go into combing files are exactly the kind of work a model can take once the files are there.

The rest of an exploration or Permian workflow does not always leave a trail. Which analog the team trusts, which failure it has already seen, which number someone cleans before the morning meeting — that still lives with people. Wai's third bucket, commercial optimization, depends on the same thing: integrating what the systems know with what the traders and schedulers already know in their heads.

Hundreds of millions is a result. It is not a map of which handoff still happens by hand.

The money behind the claim

"Hundreds of millions" is the only value figure she offered for 2026. The 50% is a cycle-time claim for exploration, not a dollar claim. The 25% is CapEx per BOE in the Permian, and Chevron's own earnings language puts sweating the asset and drilling and completion efficiency in the same sentence as the technology overlay.

Investors heard a clear story. They did not hear a line-item AI P&L. On commercial work, Wai said AI helps integrate production, infrastructure and market data so the company can forecast better and find constraints in a volatile market. That is a use case. It is not a published savings number of its own.

The lesson for everyone else

A company that already owns a million-barrel Permian position and a million-file exploration archive will get a different return from AI than one that buys a seat and asks for summaries. The sequence still matters. Put the tools on the work that leaves files. Then find the steps that never made it into a file. Budget the second category by workflow, not by how many people have a login.

The smaller version does not require Chevron's data advantage. Pick one recurring job. Talk to the people who did it last week, not the owner of the process. Find the handoff that still happens by hand. That is the step worth software. Searching the archive is the easy half.

What we still don't know

How much of the hundreds of millions is AI versus other technology. Whether the 50% cycle-time cut is measured from idea to drill decision, or only the document-search step. How many people use ApEX. Whether the commercial gains are booked the same way as the Permian CapEx number.

Wai said the value "is only gonna increase as we scale." She did not say what scaling looks like in headcount or spend. TD Securities published the episode on October 15 and still listed the full written transcript as forthcoming on its own page, so this account draws on the audio transcript of that conversation.

This story draws on Jeanine Wai's remarks on the TD Cowen Insights podcast recorded September 24, 2026, published on TD Securities' site, and on Chevron's Q2 2026 earnings commentary and ApEX newsroom post.