Not so long ago, the idea of an American company directly paying a Chinese company like DeepSeek to use artificial intelligence would have sounded unlikely, to say the least. Not because there was a lack of alternative models, but because the enterprise AI board seemed dominated by the big names of Silicon Valley and by a growing concern around data, security and technological dependency. But the expense is starting to weigh. And when a technology becomes expensive to maintain at scale, some companies begin to look beyond the usual vendors.
The data. The specific signal appears in the monthly list of Ramp, a New York-based company that processes business expenses and ranks the software providers its customers buy for the first time. In June 2026, DeepSeek ranked first in that ranking. The data was also collected by SCMP, which presented it as part of a movement by some American companies towards more affordable AI options compared to alternatives such as OpenAI and Anthropic.
What Ramp measures. The nuance is important, because this ranking does not say that DeepSeek has surpassed OpenAI or Anthropic in total enterprise use. As we say, Ramp classifies the suppliers that its clients buy for the first time, which serves to detect early signs of interest, but not to automatically convert them into consolidated market share. In other words, DeepSeek appears as a trending provider within that spending universe, not as the new leader of enterprise AI in the United States.
It’s not just open source. Ramp’s precision is relevant because it separates two very different scenarios: using an open source model within the infrastructure itself or contracting DeepSeek directly as a provider. In this case, Kharazian assures that the spending data points to the latter and summarizes it like this: “Companies are sending and receiving data directly through DeepSeek.” That nuance explains why the movement attracts so much attention. We are not just talking about companies testing Chinese technology in isolation, but about direct payments and use of the service.
The underlying explanation is in the cost. Kharazian notes that companies are adopting more disciplined management of AI spending and that he expected more interest in open models or cheaper options from OpenAI, Anthropic and Google. What he did not expect, he explains, was that American companies would end up using DeepSeek. Therefore, the Chinese service remains part of a business conversation marked by invoices that are increasingly difficult to ignore.
Proportion matters. DeepSeek now appears high on the monthly list of trending providers, but its previous figures within the Ramp AI Index show that we are still talking about a small phenomenon. According to Kharazian, the Chinese company went from 0.3% enterprise adoption in January 2025 to 0.1% shortly after, and in April 2026 it was still around that level. In that same index, Anthropic and OpenAI concentrated 34.4% and 32.3%, respectively. The reasonable reading, therefore, is not that DeepSeek has caught up with the leaders, but that it has re-entered the radar of some American companies.
The complete photograph. According to the firm, companies are not only looking towards Chinese models, but also towards open models and model inference and deployment platforms such as Fireworks AI, fal AI and DeepInfra. In any case, the message for Silicon Valley is clear: some American companies are willing to look at alternatives that recently seemed much more difficult to imagine.
Images | with Nano Banana
In | France has been determined to rob Spain of its position as a data center power in Europe
