Highlights
DeepSeek’s AI models are gaining quiet traction among businesses despite national security concerns.
The allure is DeepSeek’s cost-effectiveness, which makes AI adoption more appealing to businesses.
Enterprises can safely deploy DeepSeek by self-hosting, but consumer app use faces risks of data exposure to China.
After making a big splash — and raising national security concerns — in January, DeepSeek has largely faded from the headlines. Its V3 chatbot and R1 reasoning models were at once praised and vilified, with several countries including the U.S. banning their governmental use.
So it might surprise many that the Chinese artificial intelligence (AI) company’s models have not crashed and burned but actually rank second or third in several cases only to OpenAI’s ChatGPT and Google’s Gemini. Notably, it has outpaced Perplexity, xAI’s Grok and Anthropic’s Claude.
Consider the following data from Similarweb:
Total website traffic in May 2025
Generative AI traffic market share
Six months ago (before DeepSeek hit the headlines)
As of June 10:
According to The Wall Street Journal, U.S. companies are interested in checking out DeepSeek. CIOs tell the news outlet that they are “thrilled” about the prospect of it driving down AI costs in the U.S. as developers adopt DeepSeek’s techniques.
DeepSeek claimed to have spent $5.6 million training one of its AI models, compared to $100 million to $1 billion cited by Anthropic CEO Dario Amodei, according to the Journal.
Read more: German Data Protection Authority Asks Apple and Google to Stop Distributing DeepSeek App
DeepSeek’s popularity appears to be based on its cost efficiency, according to a Bain & Co. blog post. “DeepSeek’s performance appears to be based on a series of engineering innovations that significantly reduce inference costs while also improving training cost.”
DeepSeek’s mixture-of-experts architecture activates only 37 billion out of 671 billion parameters for processing each token, which reduces computing costs without sacrificing performance, the authors said. DeepSeek also “optimized distillation techniques” that let reasoning capabilities of larger models be transferred to smaller models.
What’s surprising is that an “open-source Chinese startup has managed to close or at least significantly narrow the performance gap with leading proprietary models,” according to the blog post.
As for national security concerns, it depends on how the user accesses and uses the models. If a company or user downloads the model weights and runs inference locally, data doesn’t leave the device. Companies can also run DeepSeek models on premises or in private cloud environments to avoid data being sent to foreign servers.
On the other hand, using the DeepSeek mobile app or on the web means sending data to servers in China, which would be subject to the purview of the Communist government.
Therefore, enterprises can prevent data from leaving the country but consumers or individual users typically cannot. Hence, DeepSeek apps faced usage bans for government workers in the U.S., Italy, South Korea, Australia, Taiwan and other nations.
The private sector, however, has forged ahead.
In January, Perplexity offered DeepSeek R1 for doing deep research. The AI startup said the model is hosted on U.S. and EU data centers, so user data doesn’t go to China. It also created a version of DeepSeek R1 that provides “unbiased, accurate and factual information.” The model is called R1 1776 – after the year of the U.S. Declaration of Independence.
Cloud giants AWS, Microsoft Azure and Google Cloud have all added DeepSeek’s R1 AI model to their marketplaces. Cloud company Snowflake said it began offering DeepSeek to customers after getting a flurry of customer inquiries, according to Reuters.
As open-source models, they are also available through code repositories Hugging Face and GitHub.
Notably, Chinese users have been the most enthusiastic adopters. According to People’s Daily, the government’s flagship paper with an English-language version, Chinese companies including telecom operators, brokerages and automakers are rushing to use DeepSeek.
Read more: DeepSeek Upgrades AI Reasoning Model to Rival OpenAI and Google
Read more: DeepSeek Offers Steep Off-Peak Discounts to Spread Out Demand
Read more: DeepSeek Faces Ban on US Government Devices
We’re always on the lookout for opportunities to partner with innovators and disruptors.
Learn More
