Rafee Tarafdar, the chief technology officer at Infosys, wasn’t caught off guard by last month’s news of a low-cost artificial intelligence model by DeepSeek. The India-based IT services company has been keeping an eye on the Chinese AI company for more than a year.
“We’ve been fairly closely watching all of the developments,” says Tarafdar. “The reasoning model that they have launched is something we found to be very good.”
Infosys is a little over two years into its generative AI journey and has built the company's architecture to support the use of various AI models from the likes of OpenAI and Google, matching those large language models to best fit use cases that range from coding assistance to content developed for employee upskilling. DeepSeek could potentially join that group of AI vendors though Infosys isn’t yet ready to work with it, citing concerns that surfaced after conducting tests evaluating privacy violations and how the system responds to fraud.
“These are some areas that we have identified as risk areas that need to be mitigated,” says Tarafdar.
CTOs and chief information officers have only had a couple weeks to evaluate DeepSeek’s latest AI model, which upended prior assumptions about the training and inference costs to develop AI and presents a new case in support of open-source AI. And with rising geopolitical tensions between the U.S. and Chinese governments, heightened by the latest salvo of tariffs, corporate tech leaders must balance a wide spectrum of factors as they assess the industry's hot new AI model.
“This comes from a country with sanctions,” says Carl Froggett, CIO at cybersecurity provider Deep Instinct. “That’d be a pretty risky move, in my opinion, to really start embedding it and using it.”
For CIOs at companies thinking about trying DeepSeek, the most important first step involves addressing concerns that they and their legal department may have around data privacy. Many tech leaders Fortune spoke to stress the distinction between DeepSeek’s AI app, which collects and stores personal information from users on servers in China, compared to running the DeepSeek-R1 model via model catalogs offered by Microsoft Azure and other cloud providers, which is a more secure environment. “For companies that do choose to use DeepSeek, it’s important that they understand the privacy policy and terms of service," says Graeme Thompson, CIO of enterprise cloud data management company Informatica.
For organizations that deem DeepSeek safe, they will then need to evaluate if its AI models are a good fit for their companies’ generative AI use cases and the outcomes they hope to achieve. The fact that DeepSeek is open source is a big plus for organizations looking for greater control or lower costs as they incorporate AI into their operations. Informatica's Thompson cites the ability to fine-tune the R1 model on proprietary datasets as an important benefit, for example.
For Patrick Richards, CIO of fleet management software provider Motive, greater clarity on pricing is a potential draw. “We evaluate all the most interesting and potentially valuable technologies in the marketplace to see if they have applicability, if they could be run safely at lower cost,” he says. “With that mindset, DeepSeek is no different.”
As Motive has ongoing conversations with vendors like cloud-based software giant Salesforce and AI-powered search startup Glean, Richards is still sorting through how best to evaluate AI's pricing structure. Some AI vendors are experimenting with consumption-based pricing, which is similar to how cloud computing is billed, but comes with uncertainty as businesses like Motive don’t know how much AI they’ll end up consuming in the long run.
"We’re very concerned and very closely tracking the cost of compute,” says Richards. “Things like DeepSeek are exciting because they represent dramatic decreases in that cost.” Open source AI isn't exactly "free," as there are plenty of costs associated with data preparation, ongoing maintenance to keep the AI systems up to date, development costs as engineers design and implement the AI model to match each use case, and computing power.
OutSystems CIO Tiago Azevedo says he will evaluate DeepSeek with the playbook he developed to evaluate OpenAI and to later assess major AI hyperscalers like Google and Amazon. “The type of risks that we always assess is how much of the information that we incorporate is used to train the model,” says Azevedo.
DeepSeek, he adds, will ultimately put more pressure on U.S.- and European-based AI companies to lower the AI's costs. “The competition that [DeepSeek] is generating is healthy,” says Azevedo.
How that competition plays out amid the fraught political backdrop remains unclear however. In recent months, there have been job cuts in China by major American employers like IBM; sanctions imposed against 16 Chinese chip manufacturers; Beijing’s antitrust probes that have targeted Nvidia, Google, and other U.S. tech giants; and state and federal bills that are already seeking to ban DeepSeek from government devices.
“If I’m using DeepSeek’s open source and I’m hosting it, the issue you then have is, you don’t know how it was trained and its biases,” says Froggett. “Do you trust Facebook and Llama, or Mistral or whatever open-source you want, or do you trust China and DeepSeek?”
John Kell
Send thoughts or suggestions to CIO Intelligence here.
