AI in Corporate America: What 15 Big Companies Actually Disclosed

The companies deploying AI hardest are hiring

Alphabet added nearly 12,000 employees over a year, going from 187,103 to 198,933, while running one of the largest AI infrastructure buildouts in the world. Eli Lilly went from 47,000 to 50,000 in the year it announced a major AI supercomputer partnership.

Wells Fargo put the point most directly: it measured a 35% increase in code written and stated it had not reduced resources because of the efficiencies gained, saying instead that it is doing more than it would have been able to do in the past.

Real adoption is measurable, and it is not universal

Bank of America reports nearly 200,000 teammates enabled with AI tools and approximately 150,000 active users against 213,000 staff, generating more than 1.5 million prompts a week.

Wells Fargo publishes the most useful version of all: about 93% of staff have access and roughly 12% use the internal platform. Access and adoption are different numbers, and almost nobody reports both.

The investment is enormous, and openly disclosed

Microsoft put approximately $115.9 billion of cash into additions to property and equipment in its 2026 fiscal year against $64.6 billion the year before, up about 80%, and that line excludes finance leases. Meta guided to $130 to $145 billion. Amazon reported a $66.1 billion increase in property and equipment purchases which it stated primarily reflects investments in artificial intelligence.

These are not leaks or estimates. They are audited numbers a company can be sued for getting wrong, which makes them the most reliable figures available on what AI actually costs.

Bank of America Shipping

Erica and internal AI tools. A consumer virtual assistant, plus AI tools on employee desktops.

The cleanest internal adoption funnel published by any bank. Nearly 200,000 teammates are enabled with AI tools on their desktops, approximately 150,000 are active users, and together they generate more than 1.5 million prompts per week against 213,000 teammates - roughly 70% active use at about ten prompts each per week, which is a realistic picture of enterprise AI rather than a launch statistic. Externally, Erica has had more than 3.2 billion interactions since its 2018 launch and more than 693 million in 2025 alone, helping more than 20 million clients. Note the coding assistants are a separate, much smaller group of about 18,000 technology staff.

Source: Bank of America 2025 annual report (SEC) · Mar 23, 2026

Wells Fargo Shipping

Employee generative AI. Assistants for staff, plus an internal AI platform.

The most honest adoption funnel published by any large employer, and worth more than a headline number: roughly 190,000 employees have access to Microsoft Copilot Chat, about 40,000 engage with it, and around 25,000 use the internal platform. That is 93% access and roughly 12% deep use. Any vendor telling you enterprise adoption is universal has not read this.

Source: Wells Fargo 2025 annual report · Mar 18, 2026

Alphabet Shipping

Growing while deploying. Google's workforce through its heaviest AI investment period.

Alphabet went from 187,103 employees to 198,933 between mid-2025 and mid-2026, adding nearly 12,000 people. That is the company doing the most aggressive AI infrastructure buildout in the world, hiring through it. Whatever AI is doing inside Google, it is not showing up as a shrinking payroll.

Source: Alphabet Q2 2026 results (SEC) · Jul 22, 2026

Wells Fargo Shipping

AI coding, and what it did not do. AI assistance for software development.

The single most useful sentence in this research for anyone worried about their job. Wells Fargo reports a 35% increase in code written, and then says plainly that it has not reduced resources because of the efficiencies gained - it is doing more than it would have been able to do in the past. A large employer measuring a real productivity gain and explicitly declining to convert it into headcount reduction.

Source: Wells Fargo 2025 annual report · Mar 18, 2026

Eli Lilly Shipping

AI supercomputer and a co-innovation lab. Large-scale computing applied to drug discovery.

Lilly grew from 43,000 employees in 2023 to 47,000 in 2024 to 50,000 in 2025, adding 3,000 in the year it announced one of the industry's most powerful AI supercomputers with NVIDIA. A separate co-innovation AI lab announced in January 2026 is described as up to $1 billion over five years and is a joint investment with NVIDIA, not Lilly alone. Worth noting honestly that Lilly grew faster the year before, so this is growth alongside AI rather than growth caused by it.

Source: Eli Lilly FY2025 annual report (SEC) · Feb 18, 2026

Amazon Shipping

AI investment. Compute and infrastructure across its cloud and retail businesses.

Amazon's trailing twelve-month free cash flow swung to an outflow of $7.6 billion from an inflow of $18.2 billion a year earlier, driven by a $66.1 billion increase in purchases of property and equipment, net of proceeds from sales and incentives - which Amazon states primarily reflects investments in artificial intelligence. That is a company naming AI as the direct cause of a cash-flow reversal, in its own results.

Source: Amazon Q2 2026 results · Jul 30, 2026

Microsoft Shipping

Microsoft 365 Copilot. A paid AI seat sold on top of Office. Microsoft is selling AI here as well as using it.

Microsoft 365 Copilot has passed 30 million paid seats, which is a genuinely large number. It is worth knowing what you cannot calculate from it: Microsoft has not disclosed its total commercial seat base, so the attach rate - what share of its customers actually bought AI - is not public.

Source: Microsoft FY26 Q4 results · Jul 29, 2026

Microsoft Shipping

Datacenter capital spending. The physical buildout behind every AI product you use.

Microsoft put approximately $115.9 billion of cash into additions to property and equipment in its 2026 fiscal year against $64.6 billion the year before, an increase of roughly 80%. Read that line precisely: it is the cash-flow figure and it excludes finance leases, so Microsoft's total capital commitment is higher still and is discussed separately. Whatever you think of AI, this is one of the largest private infrastructure programs in history, and it is disclosed in an audited filing rather than a press release.

Source: Microsoft FY26 Q4 results · Jul 29, 2026

Meta Shipping

AI infrastructure. Compute for AI products and advertising systems.

Roughly $31.1 billion of capital expenditure in a single quarter, with full-year guidance of $130 to $145 billion, narrowed from a prior range starting at $125 billion. The cost shows up immediately: free cash flow fell to $784 million from about $8.5 billion a year earlier. Meta is converting nearly all its spare cash into fixed assets, and saying so in the filing.

Source: Meta Q2 2026 results · Jul 29, 2026

Oracle Shipping

Cloud AI infrastructure. Selling AI computing capacity to other companies.

The demand signal is real and enormous: remaining performance obligations of about $638 billion, which is contracted business not yet delivered. The gap is the thing to watch - annual cloud infrastructure revenue is around $18.1 billion, so under 3% of that backlog has been realized so far. Capital spending reached roughly $55.7 billion and free cash flow was negative $23.7 billion.

Source: Oracle FY2026 Q4 results (SEC) · Jun 10, 2026

Salesforce Shipping

Agentforce. Agentic AI sold inside its customer software.

Read the headline number carefully, because the company states the composition itself: Agentforce and Data 360 annual recurring revenue reaches nearly $3.4 billion, up over 200% year on year, including $1.1 billion Informatica Cloud ARR and $1.2 billion Agentforce ARR. Roughly a third is a data business Salesforce purchased rather than built, and Agentforce itself is the $1.2 billion.

Source: Salesforce Q1 FY2027 results (SEC) · May 27, 2026

Pfizer Shipping

AI in discovery and operations. AI used to compress drug development timelines and reduce costs.

The most honest sentence found in this entire research, and it comes from the company rather than a critic. Its chief executive wrote that while AI has made our work more efficient and helped significantly compress timelines, no medicine designed primarily by AI has been approved for patients. The same piece is headed by the principle that Pfizer measures outcomes, not effort, and states plainly that demonstrations and benchmarks do not impress him.

Source: Pfizer, how Pfizer thinks about AI · Aug 6, 2026

Microsoft Changed

A new risk disclosure. What Microsoft's lawyers added that its earnings calls had not said.

After a period of consistent public confidence, Microsoft's 2026 annual report added language about its AI investments being made at significant scale and on an accelerated timeline, in advance of fully developed revenue streams, and warning that the associated revenue may not be realised in the expected timeframes or at expected levels. Both phrases appear in no other Microsoft filing ever, which is what makes it notable. This is not a scandal - it is a company being appropriately careful in the one document where it is legally required to be.

Source: Microsoft FY2026 annual report (SEC) · Jul 29, 2026

Nike Shipping

AI-powered shopping on Google. Buying Nike products without leaving Google - browsing and checking out inside the Gemini app and AI Mode on Search.

The most revealing split on this page, because the same company says two opposite things in the same quarter. In May 2026 Nike's newsroom announced a multi-item cart you use inside Google's Gemini app and AI Mode, launching in the US that June, and called it Nike's position at the forefront of AI-assisted commerce. Eight weeks later its annual report described that identical behavior as a threat: to remain competitive, we may need to develop, integrate and scale AI-enabled search, discovery, personalisation, recommendation and other digital commerce capabilities, and warned about consumers using agentic shopping to discover, evaluate or purchase products. All 11 AI mentions in that filing sit in Risk Factors. Nothing in the business description, nothing in Human Capital. Worth knowing where this actually runs: Google supplies the model and the checkout, Nike supplies the catalog, so Nike is the merchant integration rather than the builder.

Source: Nike newsroom, AI-powered shopping on Google · May 19, 2026

Disney Changed

OpenAI licensing agreement. A three-year character licensing deal alongside a planned $1 billion equity investment.

Announced in December 2025 covering more than 200 Disney, Marvel, Pixar and Star Wars characters, and expressly subject to the negotiation of definitive agreements. Five months later a shareholder letter filed with regulators said that, as widely reported, OpenAI opted to shut down Sora, and as a result Disney would not proceed with its previously planned investment in the company - while adding that it continues to explore potential commercial opportunities with OpenAI and others. Note the relationship was not ended, only the investment.

Source: Disney Q2 FY2026 shareholder letter (SEC) · May 6, 2026

ExxonMobil No verified AI

Nothing verifiable. The phrase machine learning appears zero times in its 2025 annual report.

A full-text search of every Exxon filing returns only a handful of machine learning mentions, none of them in an annual report. For a company that operates some of the most data-intensive physical infrastructure on earth, the absence in the document it is legally accountable for is the finding.

Source: ExxonMobil FY2025 annual report (SEC) · Feb 25, 2026

Mastercard Changed

Decision Intelligence. Its long-running AI fraud-detection brand.

A nine-year brand that quietly disappeared. Decision Intelligence appears in Mastercard's annual reports every year from 2016 through 2024, and is absent entirely from the 2025 filing. Mastercard's whole disclosure about using AI internally is one clause saying it uses AI to enhance its operations and its employee productivity - no tool, no adoption rate, no number.

Source: Mastercard FY2025 annual report (SEC) · Feb 11, 2026

AT&T No verified AI

No AI revenue line, no AI capital figure and no named AI product anywhere in its annual report.

Two things worth pairing. Its filing says there can be no assurance that the usage of AI will meaningfully enhance our products or operations, or be beneficial to our business - careful, lawyer-drafted, and notably non-committal. Meanwhile its own data and AI chief said publicly the company is centered on one word, agentic. That word returns zero hits across every AT&T filing.

Source: AT&T FY2025 annual report (SEC) · Feb 1, 2026

AI is the most-cited reason for layoffs, in a year layoffs fell sharply

Headlines saying AI is destroying jobs, alongside a job market that does not feel like it is collapsing.

Both are true at once. The firm that counts US job cuts by stated reason found AI was the leading cited reason for five consecutive months, with 112,713 AI-cited announcements year to date, roughly 24% of all cuts. Over the same January to July period total announced cuts were 477,033 against 806,383 a year earlier, down about 41%, and announced hiring plans rose to 107,500 from 86,132.

Practical read: cited is not caused, and hiring plans are not hires. Read those two numbers as what they are - a company's chosen framing for cuts, and an intention to hire. They are still the best primary data available, and they point somewhere less alarming than the headlines.

Source: Challenger, Gray and Christmas job cut report · Aug 6, 2026

When a big company cuts staff, it usually does not blame AI

A large layoff, followed by everyone assuming AI was the reason.

Meta reduced headcount by approximately 8,000 people and booked $1.18 billion in severance, and its results give no causal reason at all. A company with every incentive to credit AI for efficiency simply did not make the connection in the document where a false statement would carry legal liability.

Practical read: when you read that AI eliminated a role, check whether the company said so somewhere it could be held to. Executives speak freely on podcasts and stages. Filings are different, and the gap between the two is usually where the real uncertainty sits.

Source: Meta Q2 2026 results · Jul 29, 2026

The heaviest AI adopters in this research grew their workforces

The assumption that deploying AI and cutting staff are the same decision.

Alphabet added nearly 12,000 people over a year while running the largest AI buildout in the world. Eli Lilly went from 47,000 to 50,000 in the year it announced a major AI supercomputer partnership. Wells Fargo measured a 35% increase in code written and stated it had not reduced resources because of the efficiencies gained, saying instead that it is doing more than it would have been able to do in the past.

Practical read: this is the fact most missing from AI coverage, and it is in audited filings rather than opinion pieces. It does not prove AI never costs jobs. It does show that the two do not automatically travel together, and that some large employers are explicitly choosing not to make that trade.

Source: Alphabet Q2 2026 results (SEC) · Jul 22, 2026

In New York City you can check before you apply

Applying, hearing nothing, and suspecting software rejected you.

New York City law says an employer may not use an automated employment decision tool unless it has had a bias audit within the past year, the results are publicly posted, and candidates are notified at least ten business days before it is used. Enforcement began on 5 July 2023 and complaints go to the city's Department of Consumer and Worker Protection.

Practical read: before applying to a New York City role, search the employer's site for its posted bias audit. It costs you two minutes, and if there is no audit and no notice, that is a violation you can report.

Source: NYC Department of Consumer and Worker Protection · Jul 5, 2023

In Illinois, an AI video interview needs your consent

A recorded one-way video interview with no human on the other end.

Illinois law requires an employer using AI to analyze video interviews to notify you beforehand, explain how the AI works and what characteristics it evaluates, and obtain your consent. It may not evaluate applicants who have not consented, and on request all copies of the video must be destroyed within thirty days.

Practical read: you are entitled to ask in writing whether AI analyzed your recorded interview. If it did and nobody told you, that is a statutory violation - and separately you can require deletion of the recording and its backups.

Source: Illinois AI Video Interview Act · Aug 9, 2019

Regulators have found AI-powered to be untrue, with penalties

A product marketed as AI-powered that does not do what was claimed.

The securities regulator settled charges against two investment advisers for what it called AI washing. One was found not to have had the AI and machine learning capabilities that it claimed, paying $225,000. The other falsely described itself as the first regulated AI financial advisor, paying $175,000.

Practical read: AI-powered is an advertising claim and is regulated as one. If a product does not do what its AI marketing promised, that is reportable to the Federal Trade Commission, and to the securities regulator where an investment is involved.

Source: US Securities and Exchange Commission · Mar 18, 2024

Ask for the denominator.

30 million paid AI seats means little without the total customer base. The companies worth believing are the ones publishing both - one bank at roughly 70% active internal use against a stated headcount, another at 93% access but only 12% deep use.

Access is not adoption.

The single most useful distinction on this page. One bank gives 190,000 people access and sees about 25,000 using the platform properly. Both numbers are true, and only one of them means anything.

Growth and AI are not opposites.

The two companies here doing the most visible AI work both added thousands of employees while doing it. That is in audited filings, not commentary.

Read the risk factors.

They are the least exciting part of an annual report and the most honest. New language appearing there is a company telling you something it has not said out loud.

Separate selling AI from using AI.

Microsoft, Oracle and Salesforce sell it. Banks and pharmaceutical companies use it. Conflating the two is the most common error in coverage of this subject, and it inflates the picture badly.

Silence is data, but it expires.

An oil major's annual report contains zero instances of machine learning. A telecom's carries no named AI product at all. Those are deliberate choices by companies with excellent lawyers - and they are true until the day they are not. Nike was in this category until it shipped a product, which is why every entry here is dated.

Are big companies actually using AI, or just talking about it?

Both, and the split is measurable where companies publish denominators. Bank of America reports nearly 200,000 teammates enabled and approximately 150,000 active users against 213,000 staff, generating more than 1.5 million prompts a week. Wells Fargo publishes the fullest picture: about 93% have access and roughly 12% use its internal platform. Note how few companies give you a denominator at all - most publish a user count with nothing to divide it by.

Is AI causing the layoffs?

It is the most-cited reason in a year when layoffs fell substantially. The firm tracking US job cuts by stated reason recorded 112,713 AI-cited announcements year to date, about 24% of the total, while total announced cuts fell roughly 41% year on year and announced hiring plans rose. Cited is also not the same as caused - and Meta cut approximately 8,000 roles with $1.18 billion in severance while giving no reason at all in its results.

Which large companies are growing while adopting AI?

Alphabet added nearly 12,000 employees over the year, going from 187,103 to 198,933, while running one of the largest AI infrastructure buildouts in the world. Eli Lilly went from 47,000 to 50,000 in the year it announced a major AI supercomputer partnership. Wells Fargo reported a 35% increase in code written and said explicitly that it had not reduced resources as a result.

How much are companies actually spending on AI?

More than almost any private infrastructure program in history, and it is disclosed. Microsoft put approximately $115.9 billion of cash into additions to property and equipment in its 2026 fiscal year against $64.6 billion the year before, up about 80%, and that figure excludes finance leases. Meta guided to $130 to $145 billion for the year. Amazon reported a $66.1 billion increase in property and equipment purchases which it said primarily reflects investments in artificial intelligence.

Has AI actually produced a new medicine?

Not yet, according to one of the largest pharmaceutical companies in the world. Its chief executive wrote that while AI has made the company's work more efficient and helped significantly compress timelines, no medicine designed primarily by AI has been approved for patients. That is the company saying it about its own field, which makes it considerably more credible than a critic saying it.

Do I have any rights if AI screens my job application?

It depends heavily on where you are. In New York City an employer must have a bias audit from the past year, post the results publicly, and notify you at least ten business days before using an automated tool. In Illinois, AI analysis of a video interview requires notice, an explanation and your consent, and you can require deletion within thirty days. Much of the United States has no equivalent requirement.

Which large companies have no verifiable AI?

ExxonMobil's annual report contains zero instances of machine learning. AT&T's filing carries no AI revenue line, capital figure or named product, and the word its own AI chief called the company's focus returns zero hits across its filings. One caution about this category, learned the hard way: Nike sat here until it shipped AI-powered shopping through Google in June 2026, so a verified silence is a finding with a shelf life, not a permanent verdict. That is why every entry on this page carries a date.

What is AI washing?

Claiming AI capabilities a product does not have. The securities regulator has settled charges on exactly this, finding one investment adviser did not have the AI and machine learning capabilities it claimed and fining it $225,000, and fining another $175,000 for falsely describing itself as the first regulated AI financial advisor.

Filings and results releases, not press releases

Nearly every company here is publicly traded. A press release is marketing; an annual report and a results release carry legal liability. Every number on this page comes from the second kind of document.

No company claim here rests on a document older than January 2026

Every company entry carries the publication or filing date of the document behind it, and that date is 1 January 2026 or later. Anything that could only be supported by a 2025 or older source was removed rather than left standing with an old citation. The one deliberate exception is law: statutes, regulator orders and court rulings are dated to when they were enacted or decided, because a 2019 statute still in force is not out of date, it is simply old. Those are the only pre-2026 dates on this page.

That rule costs the page entries, and it cost it some good ones. It is worth it. A corporate AI claim from 2025 is not a small amount out of date - Nike went from verified silence to a shipped product in eight months, and Mastercard let a nine-year AI brand vanish from its filings without ever announcing it had.

What the three labels mean

Shipping means in production and documented in the company's own filing. Changed means rolled back, replaced, repriced or quietly dropped. No verified AI means the filings were searched and the only mentions were boilerplate. A fourth label, Announced, is used on the other guides for things committed but not generally available - nothing on this page currently qualifies, because everything that survived verification is either running or documented as changed.

This page was rebuilt after failing its own check

The first version of this page was written from earnings-call transcripts supplied by a third-party aggregator. A verification pass then searched every quote against the companies' actual filings, and a substantial number of them returned no match anywhere. One headline argument turned out to be flatly wrong - the company I said had retired a metric had in fact kept reporting it.

So the draft was discarded rather than patched, and this page was rebuilt using only claims that could be traced to a primary document. It is shorter than the original. That is the correct trade.

Absence is checked, not assumed

A No verified AI label means the company's own filings were searched and the only mentions were risk-factor boilerplate - language nearly every public company now carries whether or not it uses anything. Several entries rest on a phrase appearing zero times where you would most expect it.

Selling AI and using AI are different things

Microsoft, Oracle and Salesforce sell it. Banks, insurers and pharmaceutical companies use it internally. Conflating the two is the most common error in coverage of this subject and it badly inflates the picture of how much AI is actually running inside ordinary companies.

What is deliberately not here

No hypotheticals, no vendor-funded statistics presented as research, and no predictions about what happens if the AI buildout does not pay off. The spending is measured and is reported here; the consequences are not measured, and anyone claiming otherwise is guessing.