Are we in an AI bubble?
I speak to investors often given my unique access to Ramp data on trends in software and AI spend. (I decline to share data — usually they’re interested in my methodology or what didn’t make the cut in my posts).
The most frequent question I get: is AI a bubble? Funds want to curb their AI investments if so, and double-down if not. And while most concerns about an AI bubble revolve around whether we are building too many data centers, the leading indicator will be when businesses stop spending on AI.
If you were an investor, here’s the Ramp data I would show you.
1. Paid AI adoption declined in September, for the second time this year.
According to Ramp AI Index, The share of businesses paying for AI models and services fell to 43.8% in September, a 0.7% drop from August. It was the second decline in 2025, and larger than the 0.1% decline we saw in June. The decline was modest (adoption is still above July levels), but what’s notable about this decline is how broad it was.
We saw large declines in small- and medium-sized businesses (adoption increased among large businesses). And adoption fell or stayed flat in all industries except technology and…restaurants, the lowest-adoption industry in our dataset.
It also wasn’t isolated to one model company. OpenAI adoption fell slightly (0.9% to 35.6% of businesses). Anthropic, the second-most popular provider, virtually flatlined, growing only 0.1% despite being the fastest growing model company this year.
2. That might be a short-term blip, because AI products are getting stickier.
In 2022, barely half of companies that subscribed to an AI product or service were still subscribed by the end of the year. That share has risen every year since. As of the end of 2024, more than 80% of companies subscribed to AI products and services remained subscribed through the end of the year.
I see three reasons why: First, there are more AI products and services available, and as the ecosystem matures, those products are better matching to company needs and delivering real value. Second, the products themselves are getting better, earning stronger retention rates from loyal customers. Third, a certain level of AI spend is now seen as a necessary line-item. “What’s your AI strategy” is not just a cliché. There is an expectation (by investors, executives, advisers) that firms make long-term investments in deploying AI at scale throughout their organizations.
3. AI contract sizes are growing.
In 2023, companies were signing contracts with AI companies worth around $39K on average. Contract sizes have multiplied every year since. To date, the average contract value for 2025 sits at $530K. We now project that the average contract value will reach $1M in 2026.
Our analysis includes contracts with AI products and services outside the leading model companies. It’s a sign of a maturing product ecosystem. The largest contracts in our view are often not tied to a specific model company: they’re AI customer service agents, observability platforms, and products and services built specifically for enterprise. These are longer-term investments likely to drive productivity gains beyond the enterprise chat subscriptions that are already commonplace.
So, is AI a bubble?
If it is, it’s not showing up in Ramp data yet (via software spend). There is more evidence than not that business spend on AI is increasing and will continue to increase. I will update my assessment when we see a reversal in these trends. Specifically, I’m looking for changes in the propensity to invest long term (contract sizes) and product fit with buyers (retention rates).
