Enterprise spending continues to be the driving force behind frontier AI labs' growth.

Data published on Tuesday by VC firm Menlo Ventures finds that enterprise AI spending has increased fivefold in the last year, hitting $177 billion. The report claims that more money is spent on enterprise AI deployment than on developing the models themselves, a cost that came to $102.5 billion this year, accounting for training compute, data and services.

The report, which surveyed nearly 400 decision makers at US enterprises responsible for more than $4 billion in AI spend, creates tangible evidence of the skyrocketing spending that enterprises are committing to, especially as agents take an increasing number of actions. And companies are spending far more than they anticipated: Though many planned to nearly double their budget in 2026, on average, enterprises are spending 4.7 times more than they were last year.

Here are some of the main takeaways:

• Agents are largely driving spend, accounting for 80% of all tokens on OpenRouter, or seven times human-driven token spending. Of the AI engineers surveyed, 74% said they now use agents for workflows and tasks, with 32% using multiple agents at once.
• Frontier labs still dominate the market, with OpenAI, Anthropic and Google capturing 69% of the market, up from 50% the prior year. Claude and Codex alone account for $24 billion in spend. Of those three, Anthropic and OpenAI are pulling ahead of Google, with Anthropic's market share hitting 55% from 40% the previous year, OpenAI's hitting 31% from 27%, and Google's falling to 9% from 21%.
• Open source models are seeing moderate gains, handling 19% of tokens, up from 11% last year. And these models are significantly cheaper, only accounting for 4% of model spend.
• The SaaSpocalypse might be overblown. The report found that AI is causing "modestly slowing growth for traditional [Software-as-a-Service]." Only 22% of enterprises' software budgets were redirected towards AI. Of that, 13% of that cash still went to traditional software companies for new AI products, such as Microsoft for Copilot licenses.

"Far from the SaaSpocalypse some feared, our study finds AI creates abundance in every market it touches — across frontier and open-source models, labs and independent apps," the report reads.

Our Deeper View

Though Menlo Ventures' report is likely music to the ears of frontier labs, the question remains: how long will this last? The report measured spend on deployment, rather than actual returns and value gained as a result, an open problem that many enterprises are still navigating. And though open source technology currently only makes up a fraction of usage, this report provides proof that these models are significantly cheaper, with average open source token costs dropping year on year. Additionally, while enterprise spending is currently outweighing the cost to develop expensive, frontier models, that might not always be the case, with AI giants like Anthropic poised to spend nearly $900 billion on AI infrastructure to feed the growing demand. To put it simply: Frontier labs are banking on enterprise spend, but it remains to be seen whether they eventually begin to tighten their purse strings.