Impact on the secondary market
(1) Short-term impact on the US AI market
- DeepSeek is causing turbulence in the short term , especially in the USA.
- Stock prices are affected because demand for pre-training is growing more slowly and post-training and inference scaling have not yet scaled fast enough.
(2) Influence on hardware and chip manufacturers
- DeepSeek uses FP8 instead of FP16 , which allows for optimized use of limited computing resources .
- Last week, DeepSeek had a big impact in North America : Mark Zuckerberg raised the forecast for Meta’s investments. NVIDIA and TSMC fell , while Broadcom emerged as the only winner .
(3) Impact on NVIDIA and market structure
- Analysts fear gaps (“air pockets”) in NVIDIA’s transition from H to B cards , which could put pressure on the share price in the short term.
- Nevertheless, this could represent an opportunity for investors in the long term .
- DeepSeek’s low training costs ensure short-term market reactions (e.g. to NVIDIA’s share price).
- In the long term, AI will remain a growth market , especially if CUDA remains the preferred standard – then the need for high-performance hardware will also remain high.
Open Source vs. Closed Source: “If the capabilities are similar, it’s a challenge for closed source”
(1) DeepSeek as a catalyst for the open vs. closed source debate
- DeepSeek is attracting attention because it is exacerbating the conflict between open and closed source models.
- OpenAI and others may hold back their best models to protect their technological leadership.
- But after DeepSeek’s disclosure, other companies may be forced to disclose more as well.
(2) Impact on tech companies and cloud providers
- Amazon & Co. have not made any changes yet , they continue to follow their existing plans.
- Open and closed source models currently coexist , especially because cloud providers support both.
- Universities and small labs are likely to choose DeepSeek because it is freely accessible – but there is no direct competition from major cloud providers.
- DeepSeek is not yet as mature as Anthropic in areas such as tool use and security , which would be important for long-term acceptance in Western markets.
(3) Open source as a potential threat to closed source models
- Open source puts price pressure on the entire market.
- When open source models achieve 95% of the performance of closed source models , the question arises whether expensive closed source models are still competitive.
- If open source and closed source become technologically equivalent, this would pose a major threat to closed source approaches.
DeepSeeks importance for China: “The vision is more important than the technology”
(1) DeepSeek proves China’s AI progress
- DeepSeek’s success shows that China is not two years behind the US in AI, but only 3-9 months – or in some areas even further ahead.
- Historically, China has often broken through technologies where it was blocked by the US – and AI could be a similar example.
(2) International perception and technological leadership
- DeepSeek is not a product of chance , but the result of continuous development.
- The success of R1 has raised awareness of China’s AI advances in the United States , especially at the highest levels of decision-making.
(3) China’s advantage: efficient use of limited computing power
- China can compensate for its lack of computing capacity through engineering.
- Chinese AI teams must find ways to remain competitive with fewer computing resources – this could influence the future dynamics of the AI race.
(4) The next big breakthrough: Reasoning and infinite context lengths
- China is currently reproducing existing technologies , but the crucial race is to see who can develop the next big reasoning model.
- Infinite reasoning could be the next milestone.
(5) Why vision is more important than technology
- The difference between AI labs is not in the technology, but in their long-term vision.
- Technology alone is not enough – the crucial question is what vision an AI lab pursues.
- Ultimately, the vision is more important than the technology.