top of page

DeepThoughts on Deep Seek

Writer's picture: Andrew BorgAndrew Borg

Updated: Feb 24

(cowritten with Chris McNulty)


If you haven’t heard the latest news about DeepSeek and its impact on the GenAI market today, take a look.  A Chinese startup, DeepSeek, has annouced a credible alternative to ChatGPT using open source and fewer Nvidia chips than expected.



In that spirit, we also wanted to share some new analyst research from Alan Pelz-Sharpe at Deep Analysis. (Their monthly newsletter is a great source for public research.)


DeepSeek is absolutely moving at warp speed across the indsutry. Today at Microsoft’s AI Tour in New York, Scott Guthrie annouced that Microsoft is now offering DeepMind AI models as part of Azure AI Foundry and GitHub (joining the other 1700+ AI models supported by Microsoft.)  For more, see: DeepSeek R1 is now available on Azure AI Foundry and GitHub


Scott Guthrie on stage at the Microsoft AI Tour in New York, January 30, 2025
Scott Guthrie on stage at the Microsoft AI Tour in New York, January 30, 2025

Apart from the immediate impact on certain tech stocks (NVidia dropped 17.44% on Monday), it’s not clear in the long run if DeepSeek, specifically will be a lasting answer to customer demands for cheaper, faster AI. But those demands will only grow louder moving forward.



It’s unlikely that Western companies will move en masse to hand over their sensitive data and IP to a China-based AI service - look at the history with TikTok. Observers have also raised legitimate questions about DeepSeek’s data handling and privacy policies.  But the price pressure, given what's suggested with standard chipsets and open-source LLMs, will be a real thing. At a minimum, this will freeze some deals and disrupt buyer decisions.

In the end, this may just be more FUD (Fear Uncertainty and Doubt) in a highly competitive AI market. We don’t yet know if DeepSeek is really running on cheap chips and open source as much as they’ve disclosed, so the next week will be interesting.

Emerging Considerations

An interesting and cautionary POV has emerged regarding the possibility that all the fuss about DeepSeek is actually a Chinese government-orchestrated psychological operation (psyop). DeepSeek's remarkably rapid success and close ties to the Chinese Communist Party (CCP) are fueling these suspicions. This blog by @Amuse on Substack assesses DeepSeek's technological advantages, its potential access to restricted resources, its recent impact on the US stock market (NASDAQ down 3.7% Monday, NVIDIA down 16% over the last two trading days), and the inherent cybersecurity and data privacy risks associated with its technology.


Expert opinions vary, highlighting the debate surrounding DeepSeek's true nature and intentions. It's open source and 1/10 the cost of equivalent power from OpenAI, and its performance cannot be ignored:


© 2024 DeepSeek
© 2024 DeepSeek

@mens_aurelius on X posits that it's part of a CCP strategy to undermine the huge infrastructure investments required by the US approach to AI (e.g. the recent StarGate announcements). The end goal would be to encourage the US to deprioritize these investments and convince "investors and boards to shift their focus to chasing an illusion."


 

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page