Ctrl-Altman-Delete
Necessity is the mother of invention and competition is the destroyer of monopolies
The announcement last month about Deepseek, the Chinese AI disrupter, prompted a vast amount of ‘instant expertise’ across the traditional and alternative media, ironically much of it likely written by Chat GPT. Many comments were quite technical and a lot of the rest was essentially chatroom stuff about Nvidia and why ‘everyone knew’ this was going to happen.
We have little to add in either area, but in the following attempt to tie in some broader themes with Ten Thoughts on Deepseek and Open AI.
The analogy that occurred to us (sat in the Alps) was that after developing and owning the technology around skis, bindings and boots, the US has suddenly noticed that the Chinese just invented snowboarding. Same, but different. Quicker learning curve and a lot cheaper.
As ever, none of this should be considered investment advice, please do your own research and talk to your advisor.
This is a strong message from Xi to Trump
The timing of the announcement had its own message. Released on US Inauguration Day, it is difficult not to see this as a signal to the new White House administration that necessity is the mother of invention and that by trying to limit China’s access to the fastest chips (from Nvidia) the US has simply caused them to rethink the process and come up with something just as good, but requiring cheaper and less sophisticated inputs. In other words, Xi is telling Trump that sanctions won’t stop China developing, but they might stop your own companies competing with the first movers, especially if, like Sam Altman and (no longer) Open AI, they try to get government to limit the competition.
Deepseek is apparently extremely impressive and as one of the equally most impressive (in our opinion) tech investors, Mark Andreeson, just put it “This is AI’s Sputnik moment”, which is a powerful analogue. The launch of the USSR’s sputnik satellite in 1957, let us not forget, is when the west realised it had been too busy believing its own propaganda that the Soviet Union was massively lagging the US in the space race.
The west has been so busy convincing itself that China can’t innovate and can’t compete, that they themselves can’t compute what has just happened.
Innovators rarely get to build a Moat.
In a post almost exactly a year ago (Friday Market Thinking Jan 24, 2024) we discussed a new ACCEL chip from China, which was reportedly not only many times faster but staggeringly more energy efficient. As we noted at the time…
During a lab experiment, the ACCEL chip demonstrated an impressive computing speed of 4.6 PFLOPS (peta-floating point operations per second), surpassing the speed of one of the most commonly used commercial AI chips, Nvidia’s A100, by a factor of 3,000.
Additionally, researchers noted that the Chinese chip’s energy consumption is a staggering 4 million times lower.
Techovedas.com
This was one of the reasons we were nervous about the massive momentum rally in AI stocks last year, not just the concentration effect, but the fact that first mover advantage in technology can disappear rapidly. The ACCEL chip is not the story today, but it is part of the bigger and rapidly developing picture.
It’s about third party markets
It’s not just the US market, but the third parties. The west might be able to use ‘National Security’ as a reason to shut down competition, but it can’t prevent ‘the rest’ from buying the cheaper/better product. For example, we also noted back in June that Huawei had produced a 7nm chip that was a competitor to the Nvidia H20 (the fastest one that they were allowed to sell in China) that was so good that Nvidia were having to cut prices to maintain market share. This was within 9 months of ‘experts’ saying it would take China 5-10 years to catch up.
As we also pointed out at the time, Xiaomi was also not only making things tough for Apple in phones, but also for German Automakers in EVs. The dramatic rise of Chinese EVs came as a similar ‘shock’ in the second half of last year as many western companies and analysts, confident in their own bearish views on China, appeared to be taken completely by surprise.
Deepseek isn’t the only competition.
It’s not even the only Chinese competition.
The nature of open source is the essence of scientific discovery, that others build on top of what is already there (hence Issac Newton’s quote that his achievements were built on the shoulders of the giants that went before him). Be in no doubt, that just as BYD is by no means the only Chinese EV company, so Deepseek isn’t the only Chinese AI company. Baidu, Alibaba, Tencent et al are all heavily invested here.
It’s about open source and competition, not closed source and regulated monopolies
The real issue here is not that Deepseek was apparently developed for a few million dollars while the US has spent hundreds of billions of dollars and just announced an investment of a further hundreds of billions in Stargate , it’s that this was developed as a ‘side-gig’ by some Chinese quant investors and that it is open source, so that the next set of smart quants can take it on to the next level.
This is the real point of the title; Sam Altman originally set out to provide AI as open source - hence the name Open AI, but subsequently changed direction to go full monetisation - which is behind his fallout with Elon Musk. The problem is that this can only work is it is closed source with a monopoly and a moat. Deepseek has just sunk that. In some senses this is a little like Tesla as an EV company, it has failed to capitalise on its first mover advantage and competition from China is now overwhelming it.
As an Economist, my thoughts jumped to Jevons Paradox, the idea that increased efficiency and lower prices increase demand for a commodity - increased fuel consumption leads to more travel not less. In this case, a (dramatically) lower cost of AI is going to increase mass adoption. We were clearly not alone as our friend Nick Glydon at Redburn Atlantic has his chart of the week as the number of stories on Bloomberg mentioning the topic; it jumped from one or two to over 500 last week! (Although that is likely because Satya Nadella of Microsoft used the term.)
But for this to happen raises the question of what problem the third party LLM’s were solving, otherwise it’s a bit like the proliferation of Apps in the wake of the internet arriving. The real winners were the products developed that used Apps.
There is a meme going round of one person using AI to create a paragraph to send to a client from a single bullet point. Meanwhile the client is also using AI - to convert the paragraph back into a bullet point. The equivalent of digging holes in the road and filling them in again.
It is difficult to see why Apple, Microsoft, Meta and Amazon now need to buy anything from Open AI or Anthropocene rather than build it themselves now. Or why anyone needs an expensive subscription.
Hubris? Meet Nemesis
One of the (many) discussions on the merits of Deepseek highlighted how a developer friend had run Deepseek’s R1 model on a 2021 MacBook Pro with an M1 chip (basically nothing fancy) and also that Deepseek also has a model (Janus-Pro-7B image generation and analysis model), which apparently outperforms both StableDiffusion and Open AI’s DALL-E 3. In other words it is targeting the whole Open AI product suite with open source software. It is also targeting Anthropocene, with its Claude model.
The reality appears to be that the US version of AI was incredibly expensive because nobody was incentivised to cut costs; the money raising machine found it easier to demands billions than millions and the hype expanded accordingly with seemingly nobody worrying about profitability. It was the Amazon model on steroids. Open AI had over $3bn of revenue and still lost $5bn. It is difficult to see how it will ever make money and now along comes someone who appears to be able to. It ran as an Oligopoly and it just got hit with competition. Its next fund raise is going to be a challenge.
Sam Altman of Open AI has won few friends in recent years and it looks like hubris has just met nemesis. Ctrl Altman Delete
I think your auto correct changed Anthropic to Anthropocene...