AI Spending at Big Companies Just Slowed — Is the AI Boom Finally Meeting a Budget Limit?
Corporate spending on AI is still growing, but new data suggests the pace may be cooling. That does not necessarily mean businesses are abandoning AI. It may instead mean that cheaper models, more selective usage and budget discipline are starting to reshape what companies actually spend.
Table of Contents
01 Event
TechCrunch reported on August spending data from Ramp, whose payments platform covers about 70,000 companies. The share of Ramp customers paying for AI products reached 56% in August, up only 0.4 percentage points from July.
Among the top 1% of AI-spending companies in Ramp’s sample, spending per employee fell by nearly 10% to about $7,205. That is notable because the heaviest users had been expected to drive a large portion of future AI revenue growth.
02 What Changed?
The slowdown comes as the cost of using major AI models has been falling. Ramp economist Ara Kharazian said average token costs had declined to about $0.68 per million tokens from a 2026 peak of roughly $1.15 in March.
That creates an unusual situation: usage can remain healthy while revenue growth slows because each unit of AI work costs less. Companies may also choose older, cheaper models for routine tasks instead of paying for the most powerful frontier model every time.
03 Why It Matters
Hundreds of billions of dollars are being invested in AI data centers, chips and power infrastructure. Those investments assume that businesses and consumers will eventually generate enough recurring revenue to justify the buildout.
Lower AI prices are good for customers, but they put pressure on providers to make up the difference with more usage, more customers or higher-value products. If price falls faster than demand expands, revenue can disappoint even while AI becomes more widely used.
04 What It Means for You
For businesses buying AI, this is generally positive. Falling token costs make experimentation cheaper and reduce the penalty for using AI in lower-value workflows. The key is measuring whether the tool saves labor, increases output or improves quality rather than assuming more AI usage is automatically better.
Companies should also separate “AI adoption” from “AI value.” Paying for five AI tools does not prove that five tools are useful. A mature AI budget should track actual outcomes by team and use case, then remove overlapping subscriptions or expensive models where cheaper alternatives perform adequately.
This is the same pattern that eventually appears in most enterprise software categories. Early adoption is driven by experimentation and broad access. Later, finance teams ask which licenses are actually used, which workflows produce measurable returns and whether several products are solving the same problem. AI is moving into that second phase faster because usage-based pricing makes waste visible quickly.
That does not have to be negative for the industry. Cheaper inference can unlock use cases that were previously uneconomical, just as cheaper cloud computing expanded the number of applications companies were willing to run. The risk comes when infrastructure investment assumes that every dollar of lower price will automatically be replaced by a larger volume of usage.
The healthiest signal would therefore be falling unit prices combined with rising total productive usage. If businesses spend less because the same work is cheaper, AI is becoming more efficient. If they spend less because employees stop using the tools, that would tell a very different story.
05 Numbers + Context
Ramp’s data gives several useful markers: 56% of its customers paid for AI products in August, adoption rose only 0.4 percentage points month over month, and spend per employee among the top 1% of AI users fell nearly 10% to about $7,205.
Average token costs fell from roughly $1.15 per million tokens in March to about $0.68. That is a decline of around 41%. For customers, the same amount of model usage can therefore produce a much smaller bill than earlier in the year.
Ramp’s customer base skews toward technology companies, so the figures should not be treated as a perfect representation of the entire economy. U.S. Census data cited by TechCrunch showed a much lower overall business AI adoption rate of about 22%.
06 Earnyx Takeaway
A decline in AI spending is not automatically evidence that the AI boom is ending. It may be evidence that AI is becoming cheaper.
That distinction matters. For buyers, lower prices and more model choices improve value. For AI providers and infrastructure investors, the challenge is harder: they need usage to expand quickly enough to offset falling unit prices. The next useful metric is therefore not just how many companies use AI, but how much profitable revenue each unit of usage actually creates.
