The US market dropped $1.2T yesterday, in reaction to the release of DeepSeek R1, a reasoning model built on top of DeepSeek V3 (an LLM released just a month ago). This new model performed comparably to OpenAI's o1 in several benchmarks, but at a fraction of the cost. According to its creators, training the model cost less than $6M. These developments followed the announcement of Project Stargate, which secured $500B in funding from SoftBank and other backers.
The model is “open source” in the sense that its weights and architecture have been released to the public. Hugging Face is attempting to reverse engineer the entire system. As a result, the tech market dropped significantly, led by Nvidia, which recorded the largest single-day drop in dollar terms, losing nearly $600M.
If these results are legitimate (though it’s difficult to confirm whether training really cost so little), the implications could be bearish for hardware companies like Nvidia and ASML, due to a reduced demand for increasingly powerful GPUs. However, it could be bullish for companies building AI-powered services, as lower costs of goods sold (COGS) make these services more viable.
These AI products are extremely expensive to operate. While no public data exists, I believe that many of them run at a loss. With cheaper, open source models, building these services might become accessible to a broader range of companies, not just the large players that can afford to burn cash.
Big tech companies might dip because they lose their cash burning dominance. It will be harder for them to starve out challengers by underpricing at a loss. Overall, though, we may see an increase in the offering of AI-powered services, which is bullish for cloud providers.
Do you like this post? Of course you do. Share it on Twitter/X, LinkedIn and HackerNews