DeepSeek shook up the AI world on Monday with claims of a low-cost alternative to large language models (LLMs) from OpenAI and others – then claimed “large-scale malicious attacks” on its services just hours later.
DeepSeek’s announcement sent shares of leading AI chip suppliers NVIDIA (NVDA) and Broadcom (AVGO) plunging 17% on Monday. NVIDIA set a single-day record with a $589 million market capitalization loss, more than double its own previous record set in September.
While the DeepSeek claims are unverified, the startup’s AI Assistant quickly rocketed to the top of Apple’s App Store, overtaking rival ChatGPT.
DeepSeek Claims Welcomed by Rivals
DeepSeek claims its AI models – which include DeepSeek-V3, DeepSeek-R1 and Janus-Pro – can perform on a par with models from AI rivals OpenAI, Stability AI and others at a fraction of the cost and using a fraction of the chips too.
The startup submitted a paper to the arXiv pre-print service last week that said the DeepSeek-R1 reasoning model “achieves performance comparable to OpenAI-o1-1217 on reasoning tasks.” The company has open-sourced that model.
While those claims remain unproven, one researcher who ran the numbers declared DeepSeek’s claims to be “plausible.”
The researcher, Ben Thompson of Stratechery, wrote: “The key implications of these breakthroughs — and the part you need to understand — only became apparent with V3, which added a new approach to load balancing (further reducing communications overhead) and multi-token prediction in training (further densifying each training step, again reducing overhead): V3 was shockingly cheap to train.”
Meanwhile, OpenAI and NVIDIA welcomed the new AI market entrant.
OpenAI CEO Sam Altman said in a statement on X (formerly Twitter) that “deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price. we will obviously deliver much better models and also it’s legit invigorating to have a new competitor!”
In a statement to Bloomberg, an Nvidia spokesperson called DeepSeek an “excellent AI advancement” that shows how a company can create new AI models while “leveraging widely-available models and compute that is fully export control compliant.”
If true, DeepSeek’s breakthrough claims would be good news for AI users and potentially for the environment too, as the power demands of AI data centers have threatened to derail the climate pledges made by leading tech companies.
Lower-cost AI models could also spur wider adoption of AI cybersecurity tools at a time when that capability is greatly needed to defend against growing cyber threats.
Unspecified ‘Malicious Attacks’ Halt DeepSeek Signups
Amidst all the excitement, DeepSeek had to halt signups after what it said were unspecified “large-scale malicious attacks.”
A message on the company’s sign-up page said:
“Due to large-scale malicious attacks on DeepSeek’s services, registration may be busy. Please wait and try again. Registered users can log in normally. Thank you for your understanding and support.”
DeepSeek’s service status page reported on January 28 that “A fix has been implemented” but added that the company is “temporarily limiting registrations to ensure continued service” (image below).
It is not clear what caused the service disruption, perhaps hacktivists or an overwhelming number of suspicious sign-ups. Some IT observers wondered if the incident was caused by the company’s infrastructure simply becoming overwhelmed by the massive spike in interest. The Cyber Express attempted to contact DeepSeek and will update this article with any response.
Meanwhile, one AI security researcher has already reported that they were able to jailbreak DeepSeek with surprising ease.
It may be some time before all the facts surrounding DeepSeek are known, but one thing is clear: on a late January Monday in 2025, the company shook up the AI market in ways that won’t be quickly forgotten.
