Enterprise AI budgets are tilting from building in-house tools to buying them — but how big is that shift?

This asks how much enterprise AI money goes to outside vendors rather than in-house builds. The short answer is that this collection doesn't contain a percentage. If you need a dollar share, you'll need a market report. What the corpus does show is the direction: buying is gaining on building. In a16z's 2025 survey of 100 enterprise CIOs, leaders described a clear turn over the past year toward buying third-party AI apps instead of building their own Are enterprises abandoning custom AI apps for third-party solutions?. Their reason was not that vendors had better models. It was the work of keeping a homemade tool current. Models change constantly, each change means re-tuning, and an internal tool that needs that much upkeep stops giving the company any edge.

This changes what the build-vs-buy decision is about. The common view is that building is the hard part, and that cheaper, easier tooling will tip companies back toward making their own. Evans argues the bottleneck lies elsewhere Does easier tool-building actually solve enterprise adoption problems?. Most workers don't see their own tasks as something that could be automated. And adopting a tool across a company takes decisions that span departments and long timelines. Vendors sell more than code. They arrive with the use case already defined, which takes care of the "what should we even build?" question that internal teams struggle with.

The spending isn't evenly spread either. OpenAI's own usage data for ChatGPT Enterprise, covering 1,764 firms, shows adoption concentrated in larger companies that spend heavily on R&D. Inside those companies, marketing staff and early-career workers use it far more than executives do Who adopts enterprise AI first and how do they use it?. A separate study finds that the firms most exposed to AI replace outside freelance work with AI tools faster and more cheaply than other firms do Do firms substitute labor for AI at different rates?. That points to a possible split. Companies that have already built up AI skills may keep building, while everyone else buys. If so, one industry-wide "vendor share" figure would hide two quite different groups of companies.

One caution if you're tracking where the vendor money ends up. A vendor's numbers on a benchmark test don't tell you whether its product will pay off at work. An analysis of 960 real job workflows found that AI agents do well on contest-style tests but fail at long, multi-step professional tasks Why do agent benchmarks not predict real economic value?. So the shift toward buying may be running ahead of evidence that the purchased tools deliver. That makes "how much are we spending" a less useful question than "what are we getting for it."