Robert Maire Semiconductor Advisors
- - DeepSeek story is negative for OpenAI not so much for Nvida
- - We are dubious about order of magnitude efficiency improvements
- - More/smaller transistors always better- Everyone wants a better chip
- - Unclear whether DeepSeek is a clever look alike or true competitor
DeepSeek is primarily a better algorithm/software not better hardware/chip
DeepSeek claims are essentially a more efficient algorithm to make up for the fact that they (China) has less powerful semiconductor capabilities than Nvidia and the US.
It is certainly not a claim of a better or more powerful semiconductor but rather a "work around" to make up for having less powerful chips (read that as not nearly as good as Nvidia).
If you are limited in how powerful your engine is you try to be more creative about how you use what little power you do have.
However you can't ever fully make up for basic lack of power and transistor count by using clever software ......if you could, there would be no Moore's Law and we would write more clever software every year rather than make more powerful chips as making more clever software is way, way cheaper.
Better hardware will make better software even betterer.......
Necessity is the mother of all inventions
If you can't get good chips, you find a way to work with what you have got. China has been limited in its ability to get Nvidia devices or make its own Nvidia-like devices due to restrictions. So obviously China has no choice but to try to make up for that lack by working on the part of AI that it can impact, namely software. When you have 200 Billion transistors on an Nvidia chip your code may not have to be as good as compared to only having 10 Billion transistors to work with.
However, we highly doubt that you can get an order of magnitude improvement in efficiency that would imply that DeepSeek is as good as an OpenAI/Nvidia solution....its just not possible and coders at OpenAI and Nvidia are just not that stupid nor is China that lucky or brilliant.
The real question is can you get something that looks and smells close enough to scare everyone into believing you have a magical elixer. Or get something that works and is close enough for most casual observations.
AI is about "more is always better"
AI LLMs (large language models) are all about manipulating huge amounts of data very quickly. This needs a lot of memory (read that as HBM , high bandwidth memory) and the ability to move it and process it very quickly, in many parallel operations.
Simply put, more transistors in an Nvidia chip allows for more data to be moved and processed more quickly.
If you are working with a smaller set of data you can try to use smarter software to draw better conclusions out of lesser data but reliability and certainty usually suffer.
If I had a choice, I would always want my models to have more data.....
OpenAI is the victim of DeepSeek not Nvidia
It seems clear to us that the true target/victim of the DeepSeek concerns is OpenAI which claims the throne of all AI software and algorithms, which DeepSeeks claims to have out done.
Does the Emperor have no clothes? Is OpenAI really Empty AI with poorly designed, brute force , inefficient code?
We doubt that. But could an upstart do things a different way and get a usable result? Absolutely.
Even if OpenAI is an empty suit, the demand for bigger, better faster chips to run DeepSeek's better software on remains 100% intact
Competing AI players will always want an advantage in better faster hardware to run on.
We will never, ever, hear that "my software is better so now I don't need to buy hardware that is that good".
Companies vying for king of the hill in AI will get the most powerfull software along with the most powerful chips (Nvidia) to get a performance advantage over the competition.
We don't think AI players are going to start canceling their Nvidia orders any time soon nor do I think the current sold out condition will change any time soon.
We would however imagine that they might rethink their algorithmic approach or at least try to figure out the alleged magic in DeepSeek.
DeepSeek works but we don't know what's inside the "black box" yet
We have tried DeepSeek ourselves and it works. We have not tried a deep, long trial, nor have we tried to figure out what makes it work. But on simple minded tasks it seems work (at least on the surface)
We have seen similar movies before- color us "dubious"
We would point out that we have heard many, many claims of China's technology breakthroughs only to find out later that they weren't all that they were cracked up to be.
This is especially true of sectors that they have been shut out of and claim to have done better on their own.
We have heard of numerous breakthroughs on EUV technology that fell flat......
Right there is a lot of hysteria over unclear yet spectacular claims that sound a bit "too good to be true"
We don't think this is a binary situation with DeepSeek either lying or the next best thing since sliced bread but more likely somewhere in between with a different approach that is likely unique though not likely a panacea.
The Stocks got nuked
This relatively unknown company with a lot of unverified information decimated the chip stocks and took the market down with it.
It feels like the chips stocks which have been so hot for so long and had a few brushes with down drafts finally got hit by something that was difficult to shake off because there was very little substance to go on.
The chips stocks were boxing with a shadow and got sucker punched......
We think the reaction across the board was well overdone......
We think that we have a case of the fear of the unknown being greater than the fear of the known.........
Its just like the movies, fear of the unknown causes a greater reaction than a known, understood, established threat.......
We will likely eventually figure out the real threat DeepSeek presents and odds are it likely won't be as bad as presumed in today's market reaction.....
About Semiconductor Advisors LLC
Semiconductor Advisors is an RIA (a Registered Investment Advisor), specializing in technology companies with particular emphasis on semiconductor and semiconductor equipment companies. We have been covering the space longer and been involved with more transactions than any other financial professional in the space. We provide research, consulting and advisory services on strategic and financial matters to both industry participants as well as investors. We offer expert, intelligent, balanced research and advice. Our opinions are very direct and honest and offer an unbiased view as compared to other sources.
www.bbc.com
