This chart shows where enterprise generative AI budgets are actually flowing, dividing $37 billion of 2025 spending into six categories across two layers. The data comes from Menlo Ventures’ third annual State of Generative AI in the Enterprise report, which surveyed approximately 500 U.S. enterprise decision-makers and combined their insights with a bottoms-up market model spanning model APIs, infrastructure, and applications.
On the left of this spectrum, infrastructure spending totals $18 billion, dominated by foundation model APIs at $12.5 billion, with model training ($4 billion) and AI infrastructure ($1.5 billion) making up the rest. On the right, application spending reaches $19 billion, split between horizontal AI tools like ChatGPT Enterprise and Claude for Work ($8.4 billion), departmental solutions led by coding tools ($7.3 billion), and vertical AI targeting industries like healthcare ($3.5 billion).
The year-on-year growth rates beneath each bar reveal where momentum is strongest. Horizontal AI grew 5.3x, followed by departmental AI at 4.1x, while foundation models grew 3.6x despite already being the largest category by absolute spend. Model training, by contrast, grew just 1.3x, suggesting that enterprises are increasingly consuming pre-trained models rather than building their own (a finding that supports the data on LLM usage in our piece this week on the first year of agents).
Three years after ChatGPT’s launch, this $37 billion market has become the fastest-scaling software category in history.