Microsoft Corp. and OpenAI are examining whether data output from OpenAI’s technology was gleaned illegally by a group linked to DeepSeek, according to a Bloomberg News report – the latest demystification of the Chinese artificial intelligence (AI) startup in recent days.
Last fall, security researchers at Microsoft investigated people with alleged ties to DeepSeek who exfiltrated a large amount of data using OpenAI’s API. [OpenAI’s API is the primary way software developers and business customers buy OpenAI’s services.] Microsoft, which has invested $13 billion in OpenAI, notified the company of suspicious activity, the report said.
The counter-narrative comes amid professions from DeepSeek that it has developed a low-cost reasoning model that exceeds the performance of models from OpenAI, Anthropic, Google, Meta Platforms Inc., and others.
But some industry leaders insist DeepSeek’s technology isn’t what it claims to be.
White House AI and crypto czar David Sacks, in an interview with Fox News Tuesday, said it was “possible” DeepSeek stole intellectual property. “There’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI’s models,” Sacks said.
An OpenAI spokesperson said Chinese companies and others constantly attempt to replicate models of leading American AI companies, but did not name DeepSeek or any other company.
SemiAnalysis analyst Dylan Patel, meanwhile, said there is proof DeepSeek spent more than $500 million on GPUs over the course of its nearly 2-year history. DeepSeek says it spent about $5.6 million to develop its R1, an open-source reasoning model. By comparison, OpenAI and Meta have spent billions of dollars.
Among the cacophony of criticism is that DeepSeek failed to produce a scientific paper that detailed its advances.
“When did we decide that we’re going to just believe a paper that comes out of China?” Daniel Newman, CEO of The Futurum Group, told Yahoo Finance on Tuesday. While suggesting the company’s declarations require verification, he noted “It’s not unlike China to potentially try and play a little bit of PSYOP with Americans and the markets to see how we would react.”
Corporate America should have deep concerns using models like DeepSeek, warns Unmesh Kulkarni, head of generative AI at Tredence. “Certain facts in China’s history or past are not presented by the models transparently or fully,” Kulkarni said in an email. “The data privacy implications of calling the hosted model are also unclear and most global companies would not be willing to do that.”
Conversely, industry experts acknowledge the technological strides made by DeepSeek since it surfaced last month. “The models they’ve built are fantastic, they really are … but what they’re doing is not miraculous either or unknown to any of the other top-tier AI researchers or AI labs that are out there,” Bernstein managing director and senior analyst Stacy Rasgon told Yahoo Finance.
This much is clear, some tech experts admit: DeepSeek’s better performance per dollar upended leading firms’ insistence that success in large language models is capital intensive, bigger is better, and early winners will dominate. “DeepSeek continues to show that there is no moat for foundational AI models. In particular, anything that OpenAI does can be replicated with open source for a fraction of the cost,” Eric Lin, vice president of Applied AI at DynamoAI, said in an email.
Ironically, DeepSeek’s advances could ultimately help Microsoft, Meta, and others by accelerating AI demand with lower-cost products. In the long run, one argument goes that DeepSeek will likely benefit the overall AI market in lowering costs and thus leading to increased overall demand.
DeepSeek’s efficiency gains are significant, and yet another refinement in the ongoing push for lower costs and higher performance in AI, according to industry experts.
“I am particularly excited about DeepSeek’s potential to democratize AI development,” Sam Mahalingam, chief technology officer at Altair, said in an email. “By enabling developers to build domain-specific models with constrained/cost-effective resources and efficient training methods, it opens new avenues for innovation. The breakthrough, in my view, lies in the open-source licensing model.”
The excitement around DeepSeek is understandable, but it’s important to separate immediate market reactions from long-term impact, according to Sheldon Monteiro, executive vice president and chief product officer at Publicis Sapient. “The AI industry has a habit of overestimating short-term breakthroughs while underestimating their cumulative effects over time,” Monteiro said in an email.
Still, DeepSeek’s sudden emergence has stirred reverberations throughout Silicon Valley, Wall Street, and all points in-between. Underlying paranoia in tech circles may have been provoked even further Wednesday with news of AI startup Moonshot AI’s release of Kimi k1.5, a model that it says eclipses OpenAI’s ChatGPT-4o and Anthropic’s Claude 3.5 Sonnet in key benchmarks.
