Huggingface is now in process of replicating Deepseek R1
Deepseek saved open source AI. The genie is out of bottle now. When everyone has free access to SOTA AI, then US has no moat....
Hardware is now the remaining bottleneck rate limiter (GPUs, EUV, etc)
In a few years Huawei will be competing with Nvidia and its truly game over.
The fore runners on the LLM front are now Open AI (Microsoft), Google, Meta, Deep Seek, Anthropic (Amazon), Alibaba, and Byte Dance.
There's not enough market for all of these companies. The Western companies are mainly held up by scatter shot investment, where investors are hedging their bets by throwing money at all of them - similar to what they were doing during the internet bubble and bit coin bubble - so that they don't miss out on the winner. Open AI is currently the government favorite (Sam in the White House etc.), but Google probably has better long term fundamentals. There's also Musk's x AI, which could get a boost from Musk being in the government, but they've yet to show they're at the frontier.
If I had to bet, I would say that Google is most likely the eventual winner in the West, though Open AI is a close contender. The main reason I favor Google is because, despite both companies having great research teams, Google has the benefit of dominating search despite Microsoft's best attempts, and having that reliable, synergistic revenue stream outside of just selling chat bots makes their project much more sustainable. Nonetheless, Microsoft may come to dominate the LLM coding space, simply because they have existing leverage there with Visual Studio and GitHub. I don't much like Meta and Anthropic's chances - the chat bot business model sucks.
The situation is different in China. There is no "Chinese Google" because, we should be honest, Baidu sucks. The chances of them winning the LLM race are slim, despite the search advantage; and in any case, being the leading search engine in China is not the same, because the Chinese internet is fundamentally different from the English internet in that the former is much more diffused between various media formats, while the latter is arguably still concentrated in the text-based World Wide Web.
Productivity enhancement is going to be much more important, as such, in the Chinese market. Companies will favor the model that give them the most bang for the buck. LLM companies are currently competing on price, and in that domain, companies that have deep pockets and which can sustain low or even negative margins for a long time, has the advantage. Deep Seek has the benefit of being backed by their own hedge fund, but Byte Dance and Alibaba both also have their own revenue streams outside of AI. The question in this scenario is more around integrations - who will be able to capture the most third party partners and build the best ecosystem? So far it's not clear, but I think we will get much better clarity in 2025, specifically through answering questions like: 1) is Deep Seek going to beat the rest to o3, 2) will the next Qwen release be successful or a bust, and 3) how fast can Byte Dance leverage multi-modal capabilities and stand out from the rest?
2025 will prove to be an exciting year, and I hold onto my previous argument, which is that China doesn't have to
win the AI race, it just needs to
not lose it. Being a fast follower is fine - and more profitable - as long as you're close enough.
