Explore our Topics:

What do healthcare executives need to know about DeepSeek?

DeepSeek, a new AI model from China, is sweeping headlines due to claims of its power and affordability. What does it mean for healthcare?
By admin
Jan 29, 2025, 3:58 PM

In late January, DeepSeek, an artificial intelligence company based in Hangzhou, hit the bigtime with the launch of its latest model that experts say rivals the results of OpenAI’s Chat-GPT – but at significantly lower costs to train and power. The new AI model has shocked stocks and flabbergasted founders across Silicon Valley, who up until now thought they had a pretty firm lock on the world of large language models (LLMs). 

While many of the technical differences between DeepSeek and Chat GPT are pretty in the weeds for the average app user looking for help writing emails or answering trivia questions, there are some potentially significant impacts in the business realm, including the healthcare environment.  

Faster, cheaper, and more efficient AI could help accelerate adoption of advanced models in patient experience applications, clinical decision support, and revenue cycle management use cases, all of which could lead to reduced costs and better outcomes for the health system as a whole. 

But there are a number of intermediary steps that would have to take place, including gathering additional unbiased evidence that DeepSeek’s successes are really as remarkable as they seem and navigating the thorny geopolitical implications of adopting foreign technology. 

For now, let’s break down the hype to better understand what DeepSeek is, what makes it special, and what executives need to know now about this intriguing development in the battle for the best AI. 

What is DeepSeek?

 DeepSeek is an AI company owned by a Chinese hedge fund whose co-founder, Liang Wenfeng, serves as CEO. In early January, the company released chatbot app based on a trio of complementary models (R1, V3, and Janus-Pro-7B) that have matched the performance capability of OpenAI’s latest models in industry benchmark testing.  

That in itself wouldn’t cause such a kerfuffle, except DeepSeek claims to have hit these markers for just a fraction of the cost and computing power OpenAI has spent on its development. 

Sources say that compared to OpenAI’s $100 million investment in its models, DeepSeek has only spent $6 million getting its tools off the ground. The company also reportedly used dramatically fewer computing chips to get there – and not even the most cutting edge ones – which is adding further fuel to the political fire in light of a recent ban on exporting certain US-made chips in an effort to curb China’s AI progress. 

That’s because DeepSeek’s models use some resource-efficient techniques, such as the “mixture of experts” approach, to solve problems with less compute power. With this strategy, the model is divided into sub-networks, each of which handle its own component of the problem, then combine to jointly complete the task.  

This is more efficient than having to activate the entire model for every task, and thus saves on compute time, power, and expense. For example, according to tech news site CCN, DeepSeek has 671 billion total parameters, but only 37 billion are active at any time to handle specific tasks.  

The models also employ efficiency techniques such as load balancing and multi-token prediction training which allows them to predict multiple pieces of information concurrently to accelerate calculations. 

Why is it such a big deal?

There are a few reasons why the launch rattled AI experts, alarmed politicians, and wiped hundreds of billions of dollars off big tech stocks, including Nvidia, the world’s primary chip maker for AI applications. 

First, China has been a digital adversary on many fronts for many years, and state-backed cyberthreats are a major concern for the healthcare industry and other critical sectors of the US economy. Seeing DeepSeek rocket to the top of the free downloads list immediately on the heels of the controversial TikTok ban is not a good look for US officials who have been trying to prevent American users from freely giving all their data to a rival country’s cyber forces.  

Second, the financial implications for chip manufacturers and AI developers could be massive. The New York Times says that DeepSeek required only about 2,000 Nvidia chips to get to this point, compared to the 16,000 that other companies typically use to train their chatbots.  

Extrapolated to the larger market, that’s a big hit in volume for a sector that has been growing exponentially based on a certain expected level of demand. On the heels of the release, Nvidia’s market cap plummeted by about $600 billion, and hasn’t yet recovered from fears that the tech sector is in for a major, long-term shakeup. 

It’s important to note that some experts are questioning the data coming from the company itself around its costs and chip usage, as well as how much of its architecture has been siphoned from existing sources and recycled into the app, so it might take some time to clarify these actual numbers. 

Depending on one’s perspective, however, the anticipated need for fewer chips could actually be a good thing.  AI is very thirsty in terms of electricity, water, and other resources, so lower computing requirements could be better for the planet. 

And smaller, more nimble, more specialized AI companies could start to do more with less and prevent a monopoly on AI by the big guys, including OpenAI, Meta, and Google. This could bode well for industries with specific needs, and could lead to more choices and more ethical AI development. 

What does DeepSeek mean for healthcare?

Right now, there probably aren’t any immediate impacts for the healthcare system, except to repeatedly remind all staff never to share protected health information (PHI) or other sensitive data with chatbots of any kind, domestic or foreign. 

Over the next few months and years, however, the implications could be significant. 

If AI can be developed and trained on less extensive and less expensive infrastructure, health systems, payers, and pharmaceutical companies could do much more in-house to design models that are specifically tailored to the complex needs of the sector. This could jumpstart many of the projects that have been limited by questions of safety, accuracy, and appropriateness, such as bedside AI clinical decision support and more effective predictive analytics tools. 

DeepSeek itself is designed to be better at solving narrower, deeper problems, compared to Chat-GPT’s broader approach, which could further healthcare’s successes. However, it’s important to remember that stakeholders may face legal and compliance issues if they try to build on top of DeepSeek’s architecture, especially during a time of heightened sensitivity around China’s access to US data stockpiles. 

Instead, they should invest in learning from what DeepSeek can accomplish and use those findings to develop their own models that meet the compliance, privacy, and security parameters for the highly regulated healthcare field. 

Overall, competition from unexpected quarters can be a good thing for catalyzing innovation and keeping leading companies honest. End users, including healthcare stakeholders, are likely to benefit from the international race for better, faster, and cheaper AI – if they have an investment and implementation strategy that helps them stay nimble in the face of extraordinarily rapid innovation. 

It’s hard to tell how much of the hype and uproar is just a gut reaction to an unanticipated development and how much DeepSeek will really permanently change the landscape. But by now, healthcare leaders know they should be prepared for anything, because digital transformation never stops for a moment. 


Jennifer Bresnick is a journalist and freelance content creator with a decade of experience in the health IT industry.  Her work has focused on leveraging innovative technology tools to create value, improve health equity, and achieve the promises of the learning health system.  She can be reached at [email protected].


Show Your Support

Subscribe

Newsletter Logo

Subscribe to our topic-centric newsletters to get the latest insights delivered to your inbox weekly.

Enter your information below

By submitting this form, you are agreeing to DHI’s Privacy Policy and Terms of Use.