A few thoughts on DeepSeek – and why you can't trust what you read
thatstocksguy.substack.comThe article is interesting, raises doubts about DeepSeek’s narrative, particularly its claim of being built on a shoestring budget with outdated chips. Archer highlights the influence of the Chinese government in tech ventures and suggests that DeepSeek could be part of a strategy to disrupt U.S. tech markets, especially AI investments.
Archer also warns that if DeepSeek’s model is genuine, it could trigger a collapse of the AI bubble, similar to the dot-com crash, by flooding the market with unused Nvidia GPUs and slowing AI hardware demand. He concludes by questioning whether DeepSeek is a breakthrough or a strategic move by China to shift market dynamics.
> flooding the market with unused Nvidia GPU
Most GPUs are for inference time, they will be used for doing more things as the models become more capable and valuable in increasing numbers of situations
>disrupt U.S. tech markets, especially AI investments.
By open sourcing so that others can adapt their approaches and improve their models to be as cheap?
As I see it DeepSeek that totally dirupted USA tech industry PxQ projections and, the price part at least, exposing the fact that more efficient cheaper models are possible. They're coping saying that at a lesser P, inevitably there will be a bigger Q due to Jevons Paradox.
I'd say that won't happen immediately, as the industry needs to shift based on this new information, plus, this development erodes confidence in the possibility that industry leaders really know what they are doing.
>By open sourcing so that others can adapt their approaches and improve their models to be as cheap?
But can they (americans) make cheap models? In my view, this was China telling investors to look at them instead of the americans. Add that to the 1 trillion yuan in investments the government has promised to make in AI, China is looking to demonstrate that it is in the game.
>to look at them instead of the americans
What PRC dunking on US AI at fraction of cost, latter is key, is to ensure all the trillion dollar soverign funds DON'T invest 100s of billions in American AI because ROI looks bad now. Americans are/will still twist arms and mandating "either us or them" approach to ensure investors stay way from PRC AI... and as long as US have export controls on hardware US still retains strong leverage. But many US companies lost the snakeoil salespitch for obscene investments.
IMO PRC doesn't really need outside investers (FDI single digit %s anyway), the 1T AI investment from Bank of China isn't going to prioritize LLMs / short term "AGI" grift. They're focusing on industry... biopharma etc... transforming atoms because they've got the industrial supply chain to build that humanoid robot army powered by second mover advantage models that can't make up for the fact nice prose isn't going to make US build as effeciently as PRC short of speedrunning for grey goo nano machines.
If I had investments in AI I might write an article like this.
Source: trust me, bro.