Yet, Marc Andreessen, the Silicon Valley venture capitalist and tech entrepreneur, declared on January 24 on X that this new Chinese model was a “profound gift to the world.”[i] Certainly, he is an investor in US technology companies. How can he possibly say that, given the selloff in the market this morning?

There’s an inherent contradiction here. So, let’s review quickly. DeepSeek developed a similar technology more cheaply than their competitors. Therefore, the market sells off because of the implication that tech and communication companies need neither these faster, more expensive chips, nor all the data centers to boot. But Andreesen calls it a “profound gift.” How does that work?

The answer is simple.

DeepSeek’s work is an open-source architecture. Obviously, they found a way to simplify the computational model that makes their AI so effective and thereby do it for less money. But since it is open source…then that means all American AI companies have access to their protocols, can study them, and then create their own compatibilities.[ii]

OpenAI, Google, Microsoft, and all these smart people were probably doing over-nighters this weekend to figure something out. In the likely event they are successful, what do you think happens?

If they now have a model that incorporates a quicker way to solve a problem, do you think they’ll want to use cheaper, slower chips, or more expensive, faster ones? One of the values of AI is its speed. Therefore, if you have a model that is faster and chips that are faster, then who wins? What do you think happens to US computational speed and accuracy?

If you answered, “It goes up,” then you win the proverbial “Christmas turkey.” And this is why it is a gift.

Here’s a couple of things to consider, though:

1. Yes, DeepSeek claimed they used slower and older Nvidia chips, but the world outside of China doesn’t know that this is the case.
2. Yes, DeepSeek says they spent far less to develop these capabilities, but that is also unverifiable. And,
3. Yes, DeepSeek is fast and competitive in specific cases, but their model is slower in other applications. So, their models do have limits and, therefore, they are not the “be all and end all” for this industry.

Because of these considerations and the expertise at American AI companies, I suspect this selling and over-reaction will dissipate.

And no. DeepSeek will not deep-six American technology companies. If anything, the demand by consumers, corporations, and governments for these speeds and uses will be driven higher.

[i] I’m sure there are intellectual property concerns, but if they are using the ideas and creating their own code, as opposed to copying it (which probably wouldn’t work anyway) then they could potentially avoid these concerns. Of course, I’m neither a technologist nor a lawyer, but the US does have smart ones.

Thank you for reading this analysis. If you have any questions, comments, or critiques, I’d be happy to receive them through my email at brian@broberginvestmentgroup.com .

[i] https://x.com/pmarca/status/1882719769851474108

[ii] I’m sure there are intellectual property concerns, but if they are using the ideas and creating their own code, as opposed to copying it (which probably wouldn’t work anyway) then they could potentially avoid these concerns. Of course, I’m neither a technologist nor a lawyer, but the US does have smart ones.

^{These are the opinions of Brian Broberg and not necessarily those of Cambridge- are for informational purposes only- and should not be construed or acted upon as individualized investment advice. Investing involves risk. Depending on the types of investments- there may be varying degrees of risk. Investors should be prepared to bear loss- including total loss of principal.}