AI Exits and Acquisitions

What buyers look for when they acquire an AI company

Large software firms have paid billions for AI agent companies. What they were buying, what they check before they pay, and what founders should prepare long before a sale.

The Infercline Desk · Friday, 6 November 2026 Sign in to save

rectangular brown wooden table with chair lot inside building — Photo by Benjamin Child on Unsplash

Buying AI companies has become one of the main ways large technology firms add agent capability. The deals of the past two years show what buyers value, and they also show the risks that can stop a sale. For founders thinking about an exit, and for operators whose suppliers may be bought, it helps to understand what acquirers are paying for and what they check.

Some deals buy a finished agent product and its customers. In March 2025 ServiceNow agreed to buy Moveworks, which builds AI agents for employee support, for 2.85 billion US dollars in cash and stock, subject to adjustments. In September 2025 Workday agreed to buy Sana, whose products include no-code AI agents, for about 1.1 billion dollars. In both cases a large platform bought a product that fitted its existing customers and could be sold to them quickly.

Some deals buy data, because agents are only as good as the information they can reach. Salesforce agreed in May 2025 to buy Informatica, a data management company, for about 8 billion dollars in equity value, saying the deal would strengthen the trusted data foundation needed for agentic AI. Some buy talent and technology rather than a whole company. In July 2025 TechCrunch reported that Google had taken a non-exclusive licence to Windsurf's technology and hired its chief executive and top researchers, after a planned acquisition by OpenAI fell apart; Cognition then acquired Windsurf's product, brand and remaining team. In June 2025 Scale AI said Meta's investment valued it at over 29 billion dollars, and CNBC reported Meta was putting in 14.3 billion dollars for a 49 per cent stake without voting power.

Smaller deals follow the same logic. In March 2026 Promptfoo, which makes tools for testing AI applications, said it was joining OpenAI. For a buyer, a specialist tool with a strong user base can be quicker to buy than to build.

Before paying, buyers check a familiar list, with some AI-specific additions. Revenue quality comes first: recurring revenue, growth, how many customers stay and whether a few large clients make up most of the income. Then margins. An agent business pays for AI model usage every time its product runs, and those costs can be high and changeable. Buyers look closely at the cost of running the product per customer and how it moves with model prices.

Dependence on suppliers is the next test. Many agent products rely on one or two AI model providers. If a provider changes its prices, terms or products, the business can be hurt. OpenAI's June 2026 note that it was winding down its Agent Builder product from 30 November 2026 is the kind of change buyers now ask about. A company that can switch models without rebuilding its product is worth more.

Intellectual property and data rights matter more than in most software deals. Buyers ask who owns the code, the prompts and any fine-tuned models, what data was used to build them, and whether the company had the right to use it. They check data protection compliance in every market the company serves, since customer data held without proper consent is a liability that transfers with the business.

Regulators can shape or stop a deal. In the UK, artificial intelligence is one of 17 sensitive areas where some acquisitions must be notified to the government under the National Security and Investment Act, according to GOV.UK. Cross-border deals face scrutiny elsewhere too: the Irish Times reported in April 2026 that China had blocked Meta's planned purchase of the AI start-up Manus, which Bloomberg Law had reported was worth more than 2 billion dollars. Founders in any of the seven Infercline editions should expect national security and competition reviews where buyers or technology cross borders.

Finally, buyers look at people and proof. They want to know whether key staff will stay, whether customers would renew under a new owner, and whether the product's results are measured and documented rather than claimed.

Why it matters to your business

For founders, an exit is often the largest financial event of their working lives, and the value is set by what buyers can verify. Clean revenue, clear ownership of code and data, controlled model costs and the ability to switch suppliers all raise the price and shorten the process. For operators, a supplier being bought can change its pricing, roadmap or terms, so it pays to know who owns the agents your business relies on.

What to do on Monday

If you run an AI company, write down which AI model providers your product depends on and what it would take to switch each one.
Calculate the model cost of serving each customer for the last six months, and keep it updated monthly.
Confirm in writing that your company owns its code, prompts and any trained models, and that you had the right to use the data behind them.
Check whether your business falls within sensitive sectors for investment screening in the countries where you or likely buyers are based.
If you buy agents from a smaller vendor, ask what happens to your contract, data and pricing if the vendor is acquired.

What buyers look for when they acquire an AI company

Why it matters to your business

What to do on Monday

Sources