Are businesses actually using DeepSeek?

I was drafting a post for Ramp’s top new SaaS vendors list for January 2025 and was surprised, where was DeepSeek? 

‍

The Chinese artificial intelligence start-up’s new low-cost AI model triggered a selloff in U.S. tech stocks last month. But for all the commotion, DeepSeek didn’t come close to our list of trending vendors. Ramp’s latest estimate of the top AI vendors shows, as of the end of January 2025, only 0.2 percent of U.S. corporations have used DeepSeek. 

‍

This low rate of business adoption stands in stunning contrast to DeepSeek’s popularity with consumers. Within days of its launch, DeepSeek rose to the top of the App Store, surpassing ChatGPT. DeepSeek’s model is also cheaper, open-source, and still performant relative to options from OpenAI and Anthropic. Why aren’t businesses using DeepSeek?

‍

‍‍

1. Since launch, DeepSeek has struggled to keep services up and running

It’s not clear how many companies (or users) have actually been able to use DeepSeek. DeepSeek’s status page shows a major or partial outage every day over the last week on their API. The free chat service is slightly more available, but as a chat UI, it’s not a scalable option for companies to integrate in their product, codebase, or employee workflow.

‍

‍

DeepSeek said the outages were caused by cyber-attacks and announced a temporary limit on new registrations, and still hasn’t announced a fix. Regardless of the reason, DeepSeek’s inability to provide reliable service will deter adoption by corporations, where reliability and uptime are critical, particularly in selecting AI models integrated into company products or internal operations. In other words, businesses don’t want to buy from other businesses subject to regular cyberattacks.

‍‍

2. Companies are blocking DeepSeek due to security concerns

Bloomberg reported last week that “hundreds” of companies have already moved to block DeepSeek due to concerns about data leaks. These measures, which block DeepSeek’s services completely from company devices, are so far-reaching that they also prevent employees from experimenting with DeepSeek’s free models, let alone signing paid contracts.

‍

DeepSeek is open-source, so nothing is stopping employees from setting up their own version, running it locally, and never connecting it to the internet, but running a model locally is impractical for companies that need scalable solutions. That said, we’ve seen some businesses on Ramp use DeepSeek with cloud solutions, like together.ai or Hugging Face.

‍

‍

3. The cost of AI tokens is coming down everywhere

DeepSeek’s key innovation was the ability to perform as well as OpenAI’s models at a significantly lower cost. DeepSeek was now a major competitor to OpenAI and Anthropic. If companies could achieve similar performance for a fraction of the cost, why not switch to DeepSeek?

‍

It’s likely that U.S. corporations held off on switching while they waited for a response from OpenAI, anticipating either, 1) broad price reductions on all models, or 2) the introduction of a new, cheaper, reasoning model. On January 31, just 11 days after the launch of DeepSeek-R1, OpenAI released o3-mini, its most cost-efficient reasoning model yet. It’s not as cheap as DeepSeek, but it’s pretty close.

‍

Ramp data also found, recently, that the cost of AI tokens has been falling more broadly, driven both by user-switching to lower-cost models and price reductions for frontier models. 

‍

‍

We expect the trend to continue. American AI companies, like OpenAI and Anthropic, will likely continue to introduce lower cost models and invest further in cost efficiency, driving down prices for companies and users. Ironically, these rapid changes in pricing will also deter companies from switching immediately, as they await competitive responses from their current providers.

‍

In conclusion

Concerns about uptime, privacy, the friction of setting up a local model, and decreasing cost of using high-performance models from OpenAI and Anthropic, may be enough to keep businesses off DeepSeek for now. Early signs, such as the release of o3-mini from OpenAI, show that U.S. AI companies are responding competitively. We expect U.S. corporations to continue to prefer options from OpenAI and Anthropic, if trends hold.