Research by Imagine
Enterprise AI diffusion report
The Enterprise AI diffusion report is Imagine's research on how large companies deploy AI inside their organizations: who owns AI, whether the company built or bought its platform, which models run where, and how many employees use it.
Verified See methodology
How enterprises have deployed AI
- Embedded
Klarna
Payments and digital banking
Klarna built its customer service assistant (Feb 2024) and Kiki, its internal knowledge assistant (Jun 2023), on OpenAI models, and was among the first companies to adopt ChatGPT Enterprise for staff in Aug 2023.
80%
Customer service chats handled by AI
- Embedded
Salesforce
Enterprise software
Salesforce runs its own products on itself before customers get them, under a program it calls Customer Zero: Agentforce on help.salesforce.com since Oct 2024, Employee, Manager and TechForce agents in Slack, and Slackbot since Jan 2026.
99%
Workforce using internal agents
- Scaling
Goldman Sachs
Investment banking and financial services
The GS AI Platform, which Goldman Sachs built and launched in 2024, is the firm's gateway to outside models behind its firewall and carries the GS AI Assistant, translation, data search and agent work.
~40,000
Weekly GS AI Assistant users
- Scaling
IBM
Technology, software and consulting
IBM builds its internal AI on its own software: the enterprise AI platform rests on watsonx Orchestrate, watsonx.data and watsonx.governance (Matt Lyteson, Jan 2026), and IBM's CIO organization runs on Red Hat OpenShift in IBM Cloud.
80,000+
IBM employees using Bob
- Scaling
JPMorgan Chase
Banking and financial services
LLM Suite, a portal JPMorgan Chase built in 2024 to give staff secure access to outside models, now carries more than 200 internal applications (Lori Beer, Oct 2026).
$19.8B
2026 technology budget
- Scaling
Moderna
Biotechnology
ChatGPT Enterprise, rolled out across Moderna in 2024 after the in-house mChat chatbot, carries more than 4,000 employee-written GPTs (Jan 2026); Moderna's research teams build their own AI on AWS.
100%
Knowledge workers using ChatGPT daily
- Scaling
Walmart
Retail
Element, which Walmart built in 2022, carries every associate AI tool, from My Assistant for corporate staff to the WIBEY developer agent, and Code Puppy lets associates outside engineering build their own tools (2023 to 2026).
900,000+
Weekly users of the associate AI assistant (Jun 2025)
- Piloting
Apple
Consumer technology
Enchanté is an internal ChatGPT-style assistant Apple reportedly rolled out to office staff from Nov 2025, and Enterprise Assistant answers policy and IT questions; Apple has not announced either tool (Macworld, Jan 2026).
$11.7B
R&D expense, Jun 2026 quarter
- Piloting
AutoZone
Retail (automotive aftermarket parts)
AutoZone has no named internal AI platform: it moved most applications to Google Cloud and left its legacy data centers by Apr 2026, uses Gemini Enterprise in early stages, and an Oct 2026 job posting names Vertex AI Search among the platforms behind AutoZone.com search.
Early stages
Gemini Enterprise rollout
- Piloting
Berkshire Hathaway
Diversified holding company: insurance, rail, energy, manufacturing, retail
Berkshire Hathaway has no group-wide AI platform; BNSF and GEICO build their own tools in-house, and brk tech, a group technology unit, writes shared reference architectures for applied AI and hybrid cloud (2026).
35M+
Wayside detector readings BNSF's AI reviews daily
- Piloting
Burlington Stores
Retail (off-price apparel and home)
Burlington Stores has no AI platform of its own and buys AI inside vendor tools: Paradox's Conversational ATS with its assistant Allison for store and distribution center hiring since May 2024, HireVue assessments for early-career hiring since Aug 2020, and Microsoft Copilot for employees (Jul 2026).
1,200+
Stores hiring through Paradox's AI assistant
- Piloting
Costco
Retail (membership warehouse clubs)
Costco has no named AI platform: AI sits inside its own systems, such as the pharmacy inventory system and costco.com product pages, while partners such as AskVet and Travelport supply member-facing AI, and Costco's IT runs on a Microsoft Azure hybrid cloud (Oct 2026).
>98%
Pharmacy in-stock rate with AI reordering
- Piloting
Delta Air Lines
Airlines
Delta Air Lines builds operational AI in-house, such as Baggage AI in Atlanta, runs most systems on AWS, and buys specialist AI: Fetcherr for fare recommendations and Airbus's Skywise platform for predictive maintenance.
100%
SkyMiles members with Delta Concierge
- Piloting
Ross Stores
Retail (off-price apparel and home)
Ross Stores has no named AI platform: AI work sits inside its IT organization, and job postings from Aug 2025 to Sep 2026 describe Microsoft Copilot Studio, Azure OpenAI and Azure AI Search for assistants and agents, and GitHub Copilot for engineers.
Daily
Software developers' AI use
- Experimenting
Texas Roadhouse
Restaurants (casual dining)
Texas Roadhouse has no named AI platform: its disclosed systems are a digital kitchen and a guest management system built in-house, both in every domestic restaurant by late 2025, with analytics on Microsoft Fabric and Power BI (2025 and 2026 job postings).
None
AI use cases Texas Roadhouse has named
Side by side
Every company in the report is measured on the same six dimensions, so profiles compare directly.
| Company | Ownership | Platform | Models | Hosting | Reach | Stage |
|---|---|---|---|---|---|---|
| Klarna | CEO Sebastian Siemiatkowski; no chief AI officer disclosed | Hybrid: Assistant and Kiki built in-house on OpenAI; ChatGPT Enterprise bought | OpenAI, plus Google media models and ElevenLabs voice | AWS as primary cloud; Google Cloud trains fraud models; generative AI runs through OpenAI and ElevenLabs services | 96% used generative AI daily (Aug 2024) | Embedded |
| Salesforce | President Joe Inzerillo (Enterprise and AI Technology), on the executive team; Rohan Kumar runs engineering | Hybrid: Own Agentforce, Slack and Data 360; coding agents and frontier models bought | Anthropic, OpenAI and Google models plus Salesforce's own, chosen by task | Hyperforce on AWS and Google Cloud plus Salesforce-run infrastructure in third-party data centers; Claude via Amazon Bedrock in Salesforce's VPC; Koa on Salesforce's own infrastructure; coding agents vendor-hosted | 99% of employees use internal agents (Sep 2025) | Embedded |
| Goldman Sachs | CIO Marco Argenti, on the Management Committee; OneGS 3.0 developed by President John Waldron | Hybrid: GS AI Platform built in-house; coding tools and agents bought | OpenAI, Google and Anthropic behind one gateway, plus open-weight models | Behind the firm's firewall; GenAI Platform team works on AWS; the clouds and data centers serving the models are not disclosed | About 40,000 weekly users of 46,200 staff (Sep 2026) | Scaling |
| IBM | SVP Transformation and Operations Joanne Wright, under the CFO; CEO chaired the steering committee | Built: Own watsonx stack: Ask agents, AskIBM and Bob | Claude, Mistral and Granite in Bob; others not disclosed | IBM Cloud: Red Hat OpenShift on IBM Cloud VPC, GPU-capable; where Bob's Claude models run is not disclosed | 80,000+ on Bob (Apr 2026); AskIT for about 270,000 | Scaling |
| JPMorgan Chase | CTO Scot Baldry, under Global CIO Lori Beer; becomes CDAO at the end of 2026 | Hybrid: LLM Suite built in-house; coding tools bought | OpenAI and Anthropic behind one portal, plus open-weight models | Own data centers and cloud providers, AWS and Microsoft Azure included; bank data barred from model training | About 150,000 weekly users (Feb 2026) | Scaling |
| Moderna | Chief People and Digital Technology Officer, on the Executive Committee | Hybrid: ChatGPT Enterprise bought; research AI built in-house | OpenAI for employees; in-house machine learning in research | OpenAI's hosted ChatGPT Enterprise for employees; AWS for research, data and machine-learning workloads; no company data centers disclosed | All knowledge workers use ChatGPT daily (Aug 2025) | Scaling |
| Walmart | EVP AI Acceleration Daniel Danker, on the Executive Council; Global CTO Suresh Kumar runs the platform | Hybrid: Element built in-house; ChatGPT Enterprise and Claude bought | OpenAI, Anthropic and Walmart's own Wallaby LLMs, chosen by task and cost | Google Cloud, Microsoft Azure and Walmart's own data centers, plus edge computing in stores; ChatGPT Enterprise and Claude are vendor-hosted | 900,000+ weekly assistant users (Jun 2025) | Scaling |
| Apple | SVP Software Engineering Craig Federighi, with VP of AI Amar Subramanya | Hybrid: Enchanté and Enterprise Assistant built in-house; coding tools bought | Apple Foundation Models with Anthropic Claude and Google Gemini, as reported | Apple data centers and private servers; Google Cloud with Nvidia GPUs since Jun 2026; AWS for search inference | Not disclosed; Enchanté reported across office staff (Jan 2026) | Piloting |
| AutoZone | No named AI chief; SVP and CIO Michelle Borninkhof speaks for AI plans | Hybrid: Google's Gemini Enterprise bought; ALLDATA built its own AI tool | Google's Gemini Enterprise; ALLDATA's model not disclosed | Google Cloud, after exiting legacy data centers (Apr 2026); some on-premises servers and a mostly vendor-run IBM mainframe remain; ALLDATA's AI hosting not disclosed | Not disclosed; Gemini Enterprise in early stages (Apr 2026) | Piloting |
| Berkshire Hathaway | No group AI chief; subsidiary CEOs own AI, with brk tech (Hari Govind) as shared help | Hybrid: No group platform; subsidiaries build in-house and buy point tools | Not disclosed group-wide; Claude reported in GEICO engineering, in-house models at BNSF | GEICO: private cloud in two colocation data centers plus public cloud, reported as mostly Microsoft Azure; brk tech: Azure and Oracle Cloud; others undisclosed | Not disclosed; no group-wide employee AI tool | Piloting |
| Burlington Stores | No named AI chief; CIO Varadheesh Chennakrishnan runs IT, and the Audit Committee oversees AI governance | Bought: Vendor AI: Microsoft Copilot, Paradox and HireVue | Not disclosed; each vendor's own models | Not disclosed; Copilot, Paradox and HireVue run as vendor services; data engineering on Snowflake and Microsoft Azure (job posting); Oracle Retail on premises, moving to Oracle's cloud | Not disclosed; Copilot licensed by approval | Piloting |
| Costco | No named AI chief; CIDO Javier Polit runs technology and reports to the CEO | Hybrid: AI built into its own systems; partner AI for members | Not disclosed; Google Cloud Vertex AI reported by Constellation | Not disclosed; Microsoft Azure hybrid cloud and Costco data centers, with Google Cloud Vertex AI reported; partner AI runs on vendor systems | Not disclosed; no employee AI assistant announced | Piloting |
| Delta Air Lines | Amala Duggirala, Chief Digital & Technology Officer, reporting to the CEO | Hybrid: In-house tools on AWS; Fetcherr pricing and Airbus Skywise bought | Not disclosed; Amazon Bedrock used to test models (2023) | AWS as preferred cloud plus Delta's own data centers; vendor platforms for pricing (Fetcherr) and maintenance (Airbus Skywise) | All SkyMiles members for Delta Concierge (Aug 2026); staff use not disclosed | Piloting |
| Ross Stores | No named AI owner; AI work sits in IT, with no separate AI unit planned | Hybrid: No named platform; Microsoft Azure AI and GitHub Copilot in job postings | Not disclosed; Azure OpenAI named in job postings | Not disclosed; job postings point to Microsoft Azure, including Azure OpenAI and Azure AI Search, alongside on-premises applications | Not disclosed; software developers use AI daily (Aug 2026) | Piloting |
| Texas Roadhouse | No named AI owner; CTO Hernan Mujica runs technology and reports to the CEO | Bought: None named; 10-K cites third-party AI models and vendors | Not disclosed; 10-K cites dependence on third-party AI models | Not disclosed for AI; IT on Microsoft Azure (job postings), third-party cloud and hosting providers and a redundant data center (10-K) | Not disclosed; no employee AI tool announced | Experimenting |
Method
What the report measures
- Ownership
- Who owns AI on the org chart, and how close that person sits to the chief executive.
- Platform
- Whether the company built its own AI platform, bought one, or runs a mix of both.
- Models
- Which model providers the company uses, and whether it can switch between them.
- Hosting
- Where inference runs, and whether company data leaves the company's own environment.
- Reach
- How many employees can use AI tools, and how many use them routinely.
- Stage
- How far AI has spread through the company, from experimenting to embedded.
Stages of diffusion
From experimenting to embedded
1Experimenting
Experimenting means AI work lives in proofs of concept run by a central team or a few enthusiasts. Employees have no sanctioned general tool, and nothing runs in production at scale.
2Piloting
Piloting means a sanctioned assistant or platform is live for selected groups, a handful of use cases are in production, and the governance rules are still being written.
3Scaling
Scaling means a company-wide AI platform is open to most employees, dozens to hundreds of use cases run in production, and a named executive owns AI and reports its results.
4Embedded
Embedded means AI is part of how core work gets done: most employees use it every week, workflows and roles have been redesigned around it, and the company reports its financial effect.
Methodology
How a profile is built
- Primary sources first. Annual reports, shareholder letters, investor days, earnings calls and the company's own announcements set the facts.
- Reporting corroborates. News coverage and executive interviews fill in what filings leave out, and a claim with one secondary source is marked unconfirmed.
- Every number is graded. Each figure is marked confirmed, estimated or unconfirmed, and estimates say what they are derived from.
- Returns are business results. A use case shows a return only when the company reported one and credited it to AI, in revenue, cost, headcount or output, with its date. Usage figures and task speedups do not count.
- Every profile is dated. The verification date is the day each fact was last checked against its source.
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
Questions
About the report
- What is the Enterprise AI diffusion report?
- The Enterprise AI diffusion report is Imagine's research on how large companies deploy AI inside their organizations: who owns AI, whether the company built or bought its platform, which models run where, and how many employees use it.
- How does the Enterprise AI diffusion report decide a company's stage?
- Imagine places each company on one of four stages: Experimenting, Piloting, Scaling, Embedded. The stage follows from evidence in the profile: whether a sanctioned AI platform exists, how many employees can use it, how many use cases run in production, and whether the company reports the financial effect.
- Where does the data in the Enterprise AI diffusion report come from?
- Every profile in the Enterprise AI diffusion report is built from public sources: annual reports and shareholder letters, investor day presentations, earnings calls, company announcements, executive interviews and news reporting. Each figure is marked confirmed, estimated or unconfirmed, and each profile shows the date its facts were last checked.
- Which companies are in the Enterprise AI diffusion report?
- The Enterprise AI diffusion report currently profiles Klarna, Salesforce, Goldman Sachs, IBM, JPMorgan Chase, Moderna, Walmart, Apple, AutoZone, Berkshire Hathaway, Burlington Stores, Costco, Delta Air Lines, Ross Stores, and Texas Roadhouse. Imagine keeps adding companies, each measured on the same six dimensions: ownership, platform, models, hosting, reach and stage.
- How can I suggest a company or report a correction?
- Email data@imagine.ai with the company name or the correction and a source. Imagine checks every correction against its source before changing a profile, and the profile's verification date moves when it does.