News

DeepSeek Debut Drives Increase in Self-Hosted Models, Says Cloud AI Report

• By David Ramel
• 03/05/2025

One interesting trend in a new report from Wiz about AI in the cloud is the disruption caused by the arrival of a DeepSeek model, which caused an uptick in self-hosted models.

That was one of the key trends in "The State of AI in the Cloud 2025" published recently by Wiz, a cloud security company. The report is based on data from hundreds of thousands of public cloud accounts.

China-based DeepSeek early this year released a revolutionary DeepSeek-R1 model with advanced reasoning capabilities and cost-effectiveness, addressing one of the primary challenges of advanced AI, which normally requires vast amounts of computing power and other resources to create models like those from OpenAI, Anthropic, and Google. However, perhaps influenced by geopolitical concerns, the debut caused a backlash along with some usage restrictions (see "Cloud Giants Offer DeepSeek AI, Restricted by Many Orgs, to Devs").

Detailing DeepSeek's "staggering growth," the Wiz report site said, "The release of DeepSeek-R1 prompted a surge in adoption, with its newest model netting around 130,000 downloads on HuggingFace. Roughly 7% of organizations using self-hosted AI models are currently using models developed by DeepSeek, representing a more than 2X increase in January 2025 alone."

That would be a trend to watch as it could have significant implications for the cloud security landscape, presenting new challenges and perhaps opportunities for established cloud AI leaders like Microsoft, AWS and Google, commonly referred to as the "Big Three" cloud giants. Organizations might rely less on cloud providers' proprietary AI services, potentially impacting these providers' revenue streams from AI offerings. It could also cause competitive pricing pressures.

Describing the overall trend, Wiz said, "AI adoption in cloud environments continues to rise, with 75% of organizations now using self-hosted AI models and 77% utilizing dedicated AI/ML software."

Noting the rise in self-hosted AI, the report indicated that among the most prevalent model types, BERT has become even more dominant, rising from 49% to 74% year-over-year. Other changes mentioned by the report are that Mistral AI joined the top 10 and Alibaba Cloud's Qwen2 emerged as a new contender.

Interestingly, DeepSeek also figures into another key focus of the report, security, in which Wiz specializes.

"DeepSeek quickly gained widespread attention for its cost-effectiveness and rapid pace of innovation," Wiz said. "However, as evidenced by the exposure of data in their service our research team uncovered in January 2025, the rise of DeepSeek also underscores a stark corollary of the AI gold rush: that innovation should not come at the cost of safety, and AI systems on the whole warrant much closer security oversight."

That link points to a report from Wiz Research about data exposures found in a publicly accessible database belonging to DeepSeek that allowed full control over database operations, including the ability to access internal data. The exposure includes over 1 million lines of log streams with highly sensitive information, the Jan. 29 blog post revealed.

Coincidentally, the Wiz Research data leakage report was released about the same time as another report on DeepSeek from the Cloud Security Alliance (CSA). At the time, we reported: "Wiz Research has identified a publicly accessible ClickHouse database belonging to DeepSeek, which allows full control over database operations, including the ability to access internal data," the Wiz post said. "The exposure includes over a million lines of log streams containing chat history, secret keys, backend details, and other highly sensitive information. The Wiz Research team immediately and responsibly disclosed the issue to DeepSeek, which promptly secured the exposure." (see "Cloud Security Org Weighs In on DeepSeek AI Just as Data Leakage Reported")

Further examining the security situation, one of the report's key findings notes that security continues to play catch up as threats continue to expand and new technology outpaces current solutions. "The rise of self-hosted AI models has also brought with it a corresponding increase in security incidents," the report said. "In January 2025, 1 in 5 organizations experienced a security incident related to their self-hosted AI models, up from 1 in 7 in January 2024. The most common incidents were data leaks, which accounted for 45% of all security incidents."

Wiz said these findings reinforce a key theme of the report: AI software brings massive opportunities, but also serious risks.

The detailed study also examines the mix of open and closed-source AI and details how AI managed service adoption has stabilized, among many other data points.

"This report shows a landscape in flux, with DeepSeek poised to trigger a major shakeup," Wiz said in its conclusion. "Innovation should not come at the cost of risk, and the major vulnerabilities our research team discovered in DeepSeek and other AI providers underscore the urgent need for stronger AI security. As self-hosted AI adoption accelerates, organizations must prioritize cybersecurity efforts."