US companies are spending less on AI even as their actual model usage climbs to record levels, according to the latest Ramp AI Index from Ramp economist Ara Kharazian. AI consumption has jumped about 50 percent since corporate spending hit a ceiling in July, reaching a new high at the end of September. This downturn is not the first blip in enterprise AI budgets, but it is certainly sticking around longer than the usual accounting noise.
As Ara Kharazian points out, this budget contraction is driven almost entirely by the cage match between OpenAI and Anthropic. The pricing pressure comes from direct cuts to flagship models alongside a flood of cheaper, more efficient standard and lite tiers. Together, the two frontier labs effectively own the enterprise market: in the final week of September captured by the data, Anthropic captured 51 percent of token spending, while OpenAI took 44.5 percent.
Meanwhile, open-source architectures have completely missed out on commercial budgets during this deflationary cycle. According to Ara Kharazian, open-source models still account for less than five percent of total business spending. The Ramp AI Index tracks transaction data from more than 70,000 US companies, with token granularity drawn from a dedicated subsample.
Open-source advocates spent the last year pitching freely licensed weights as the inevitable economic equalizer that would shatter proprietary vendor lock-in and bleed API bills dry. Instead, the proprietary oligopoly simply slashed prices on lite and flagship tiers, leaving open-source models with crumbs while corporate token usage surged 50 percent. If this is what vendor lock-in looks like, finance departments will gladly take it.