Menlo Says Only 22% of New Enterprise AI Spend Came From Software Budgets
Menlo Ventures puts 2026 enterprise AI spend at $176.8 billion. Of the 22% of new spend that came from software budgets, incumbents kept 13 points and AI-native vendors took 9, or $12.6 billion.
TL;DR
- Menlo Ventures estimates enterprise AI spend hit $176.8 billion in 2026, with 55% of the increase funded by new budgets.
- AI sold as a cheaper swap for a seat-priced tool competes for the 22% of new spend that came from software budgets, and incumbents kept most of it.
- Ramp's data shows token spend nearly flat in August and September, and its next AI Index will show whether new budgets last.
Enterprise AI spend grew 5.4 times this year, and most of the increase didn't come out of software budgets.
Menlo Ventures published its 2026 State of Enterprise AI report on November 3. It estimates deployment spend at $176.8 billion, up from $33 billion. Of the new spend, 55% came from budgets that didn't exist before, according to the roughly 400 US decision-makers Menlo surveyed with Morning Consult in September. Another 22% came out of existing software budgets.
A write-up headlined "only $12.6B went to new vendors" put the AI-native figure up front, but that $12.6 billion is only the AI-native cut of the software slice. An AI tool sold as a cheaper swap for a seat-priced product is chasing the smallest pool of money in the report.
New vendors took far more than $12.6 billion
Menlo splits the 22% software-budget share in two. Incumbents turning their own licenses into AI products kept 13 points. AI-native vendors took the other 9, which Menlo puts at $12.6 billion.
That slice says little about the AI-natives' overall position. By Menlo's estimate, they hold 77% of application spend and 90% of infrastructure spend. The model labs alone hold 69% of total spend, or $122.7 billion. Set against the 9 points, that suggests most of what AI-native vendors earn arrives from somewhere other than an existing software line.
The new and software budgets leave about 23% of the increase. SaaStr's Jason Lemkin puts that remainder down to labor and other budgets. Menlo sizes the sales category at $1.4 billion, up 3.6 times, and names Clay and Monaco, a Menlo portfolio company.
Replacing a seat means beating the incumbent at its own renewal
The buyer already has a contract, an owner and a renewal date for that line. The incumbent gets first claim by folding AI into the license it already sells, and incumbents kept 13 of the 22 points that way. A new vendor pitching a cheaper replacement usually has to win on price.
With a new budget, no renewal is at stake, so the vendor sells an outcome the buyer wasn't paying for. Labor budgets go further still, because a role that isn't backfilled frees a salary. For most roles, a salary runs well above a seat fee.
Lemkin turns this into an investment rule. "Products that have to win a software budget are competing for the smallest pool," Lemkin wrote. Lemkin won't back an agentic B2B company unless it opens budget beyond software, with deal sizes at least twice those of pre-AI rivals. SaaStr's own agent spend came from headcount it didn't backfill, and three people there now work alongside more than 20 agents.
Ramp sees the new money concentrating
Per TechCrunch, Ramp's token-spend data shows a sharp slowdown and a partial rebound:
• July: up 42% month over month
• August and September combined: up 1.5%
• September 23 to October 23: up 8.9%
Ramp lead economist Ara Kharazian told TechCrunch that "almost the entire spend on AI is from venture-backed, high-growth tech companies." Ramp's October AI Index restates its earlier finding that 80% of OpenAI and Anthropic spend is driven by 1% of firms.
If that holds, the new-budget pool may be a small group of heavy spenders and not a line most buyers will carry. Menlo's own numbers lean the same way. Average spend rose 4.7 times against 1.9 times planned, while the median rose 2.2 times against 1.8 times, a gap we covered when the report came out.
Ramp tracks what its customers pay model vendors, so its data speaks to the size of the new pool and leaves the budget-source split alone. The median buyer still spent more than planned, and software budgets still funded only 22% of the increase. The next Ramp AI Index will show whether token spend kept climbing after the October rebound.
Most new AI money sits outside the software line
A tool sold against a software line meets the incumbent's own AI add-on at renewal. A tool sold against a new budget or an unfilled role has no renewal to beat. New budgets alone supplied 55% of the new money.
By Menlo's split, 9 cents of every new dollar of enterprise AI spend this year went to an AI-native vendor out of a software budget.
Products that have to win a software budget are competing for the smallest pool.
Ramp's next AI Index will show whether token spend kept climbing after its October rebound. That will test whether the new-budget pool Menlo describes is widening or staying with a few heavy spenders.
