🧠 ARTIFICIAL INTELLIGENCE IS NOT JUST OPENAI AND ANTHROPIC
When people talk about the return on investment in AI, the discussion often narrows to ChatGPT and Claude. How many subscribers they have, how much they spend on computing power, and whether they can meet their obligations.
These are important questions. But you cannot assess the entire AI industry by looking at the financial performance of these two companies. Let’s look at where AI is already being sold, who is paying for it, and what they get for their money.
🏥 HEALTHCARE – GE HEALTHCARE
GE HealthCare ($GEHC) sells medical devices and software that use AI. AIR Recon DL uses deep learning to reconstruct magnetic resonance images. It also offers other algorithms through a Smart Subscription for CT.
The healthcare provider pays. It buys better imaging and tools for more efficient examinations. GE explicitly states that AI-enabled devices contribute to orders and revenue. This is already a commercial product, not a promise of future use.
🌾 AGRICULTURE – JOHN DEERE
Deere ($DE) offers See & Spray. Cameras and algorithms identify weeds, and the system applies spray precisely where needed.
Farmers pay for the technology and its use. Deere offers area-based pricing as well as an annual license. Under its Application Savings Guarantee program, the fee is tied to the savings achieved.
Here, return on investment is not measured by the number of chatbot subscribers, but by the cost of treating a field. AI has to deliver value in a specific task.
📣 ADVERTISING – APPLOVIN AND META
AppLovin ($APP) uses its proprietary AI system, Axon, to optimize advertising campaigns. In its 2025 annual report, it describes its business model directly: it earns revenue primarily from fees paid by advertisers, and Axon Ads Manager accounts for the vast majority of revenue.
Meta ($META) uses AI to deliver, target, and measure ads. The advertiser pays for the advertising service. AI is part of the technology that provides it.
So it would be a mistake to look for all AI revenue under the line item “AI subscriptions.” But it would also be a mistake to label all advertising revenue as revenue generated exclusively by AI.
🏭 ENGINEERING AND MANUFACTURING – SIEMENS
Siemens sells Inspekto, an AI-powered visual quality inspection system. It inspects castings, plastic parts, electronic components, and the accuracy of assembly, for example.
The manufacturer pays for the inspection system. Its economic benefits can be assessed in terms of defective products, repair costs, and manufacturing quality.
Even the way it learns is different from language models. During setup, Inspekto learns from good product samples and looks for deviations. It is not a universal model that has to be able to answer everything.
🎨 GRAPHICS AND VIDEO – ADOBE
Adobe ($ADBE) is developing its own family of generative models, Firefly. It offers paid plans and generative credits for creating and editing content, including images and video. Its offering also includes access to selected models from other developers.
Designers, agencies, or companies pay. They are buying a work tool. Adobe is an example of a company that develops its own models and makes them commercially available in its products.
💼 BUSINESS PROCESSES – SALESFORCE
Salesforce ($CRM) sells Agentforce to automate tasks in business processes and customer support. It offers paid options based on usage as well as user licenses.
The customer pays for tasks to be performed in the system. So when assessing the return on investment, I need to know what work the agent can handle, how much it costs, and how much human oversight it still needs.
🚕 TRANSPORTATION – WAYMO
Waymo, part of Alphabet ($GOOGL), provides paid autonomous rides. The customer pays for transportation.
This is another way AI is turned into revenue without customers buying access to a chatbot. But paid rides alone do not yet prove that the entire business is profitable.
☁️ PROPRIETARY MODELS AND CLOUD – AMAZON
Amazon ($AMZN) has its own family of Nova models. Through Amazon Bedrock, it also charges for the use of models from multiple developers, including Amazon, Anthropic, Meta, and Mistral.
So it is important to distinguish who develops a model, who operates it, and who builds the final application from it. These do not have to be the same company. What’s more, a single customer payment can become revenue for multiple links in the chain; they cannot simply be added together as final demand.
🔎 WHAT DOES THIS MEAN FOR INVESTORS?
OpenAI and Anthropic are important players, but they represent only one part of a much broader industry. Their results are not the income statement for all of AI.
At the same time, the existence of useful paid products does not prove that every investment in data centers will pay off. Not every task needs the largest model or the same amount of computing power.
That’s why I look at the whole chain: who pays, what value they receive, who keeps the margin, and whether revenue covers operating costs as well as invested capital. Revenue, customer savings, and supplier profit are three different things.
This is my personal view, not investment advice; an investment’s value can rise or fall, you may lose your capital, and past performance is no guarantee of future results.