Today, shares of KLA Corporation, ASML Holding NV, and Lam Research Corporation experienced a decline.
Over the weekend, Chinese AI startup DeepSeek made waves in the tech world by announcing the operation of a new open-source, free-to-use large language model (LLM) since December. This development has raised eyebrows and sparked conversations about China's AI capabilities and the demand for semiconductor chips in the AI sector.
DeepSeek's impressive feat cost only $6 million and took just two months to develop. The new LLM reportedly performs similarly to OpenAI's ChatGPT, throwing the spotlight on Western tech companies' supposed edge over China in AI development. This development has caused worries about the demand for advanced, high-speed semiconductor chips for AI, leading to a drop in shares for companies like Lam Research, KLA, and ASML.
After a robust period of growth, the flurry of purchases of semiconductor manufacturing equipment by China due to US sanctions and their efforts to build their own chips and AIs seem to have peaked in 2024. The country has sufficiently purchased the necessary equipment and sanctions have curtailed further imports, causing sales to slump.
However, the real concern is whether China has found a way to build LLMs with fewer or less advanced chips. If this is true, it could significantly impact the demand for semiconductor manufacturing equipment, causing a slowdown in sales worldwide.
Given the uncertainty, it's challenging to determine if this risk is already factored into the stocks' prices. Shares of Lam Research, KLA, and ASML are valued at 25x, 34x, and 40x respectively, but their long-term earnings growth estimates might be revised following the DeepSeek news. Investors should exercise caution and wait for further developments before making any significant investment decisions.
As for the broader impact on the AI industry, this presents an opportunity for companies to reevaluate their strategies. DeepSeek's development using optimized training and inference techniques has highlighted the potential for more efficient AI development without relying on high-end GPUs, potentially disrupting traditional revenue streams for companies like Nvidia.
In China, the overcapacity and misdeployment of chips have prompted a government response aimed at consolidating and professionalizing the industry to prevent future chaos. The transition from training to inference demand is also causing shifts in the market, offering opportunities for companies to adapt and seize new opportunities.
Sources:
[1] (https://www.reuters.com/business/us-ai-startup-deepseek-debuts-large-language-model-2023-07-01/)
[2] (https://www.information-age.com/china-semiconductor-market-hits-tipping-point-123506788/)
[3] (https://www.techradar.com/news/deepseek-chatbot-just-used-us-sanctions-to-create-an-ai-model-different-from-chatgpt/)
[4] (https://www.wsj.com/articles/deepseed-ai-firm-challenges-nvidia-with-laser-focus-on-tech-giant-s-heartland-5e9a2eed7d39e67aa0f05a33)
The significant reduction in development costs for advanced LLMs like DeepSeek's could shift the financial landscape of AI investing. With DeepSeek's model performing similarly to ChatGPT using less advanced chips, it might impact the demand for high-end semiconductor components, potentially impacting the profits of companies like Nvidia that rely on high-end GPU sales.
