By Akemi Kondo Dalvi, CPA/PFS, CFP
Last week the stock market roiled at the introduction of a new artificial intelligence (AI) system by DeepSeek, a privately held Chinese company that develops open-source large language models (LLMs). An LLM is a machine-learning model that utilizes artificial intelligence to process, understand, and generate human language.
An LLM gathers vast amounts of available data to answer questions, generate written content, analyze text, or translate between languages. LLM was most notably introduced to the market by OpenAI’s release of ChatGPT in 2022. ChatGPT impressed users by its ability to write essays, analyze data, create art, and write new music in seconds.
Since 2022, money has poured into the AI industry, leading to BLOOM (BigScience Large Open-science Open-access Multilingual Language Model) and Meta AI Llama (Large Language Model Meta AI), to name a few.i
However, the stock market was rocked when the new AI system by DeepSeek introduced a more efficient way to manipulate an immense amount of data at a lower cost. This sent darling of the U.S. stock market, Nvidia’s (NVDA) share price tumbling 16%, and sparked wider concerns that Big Tech’s investment by OpenAI, Softbank and Oracle in the pricey AI market could be a misstep.ii
Put simply, current AI leaders use one gigantic system to process swaths of data using an impressive amount of computing power. DeepSeek’s R-1 model provides the same output at a reduced cost of $6 million compared to $100 million spent for Open AI’s GPT-4 model, and using just a tenth of the computing power of Meta’s comparable LLaMA 3.1.iii
DeepSeek was able to achieve this by adjusting inference time computing to activate only the most relevant parts of the AI model for each query. Doing so allowed DeepSeek to process queries faster and with less computational overhead. Inference time computing is how fast and efficiently AI can take what it has learned, and apply it to solve a problem.
Traditional models like GPT-4 currently use 1.8 trillion calculations, trying to answer every query request all at once. Comparatively, DeepSeek’s R-1 model only expends 37 billion calculations to complete the same task, focusing on a specific set of queries that enable computing efficiency.
Furthermore, DeepSeek runs a leaner operation. DeepSeek currently employs 200 team members and was developed for an estimated $5 million. Comparatively, OpenAI has 4,500 employees and was funded with $6.6 billion in capital.iv
Interestingly, DeepSeek is open-source, which means anyone can build on it. Analysts suspect DeepSeek will rapidly take market share away from the more antiquated AI systems in the current market. At the same time, DeepSeek uses a model that others can copy, allowing the technology to be broken down for different types of projects, queries and models into smaller, more customized AI specialists.
This is not the end of the AI growth, nor the beginning of a tech bubble burst. Rather, this is the realization that what has been does not have to be. In fact, Alibaba has already announced its 2.5-Max AI model, which it claims is faster and more efficient than DeepSeek.
Technology moves fast and it is impossible to predict all the winners and avoid all the losers in a quickly evolving world. Diversification is often the best way to incorporate exposure to a market sector experiencing positive growth and at the same time mitigate concentration risks, maximizing consistent long-term returns. Reach out to your Certified Financial Planner (CFP) for guidance in the techy world ahead.
i. https://en.wikipedia.org/wiki/Large_language_model
ii. https://www.investopedia.com/watch-these-nvidia-stock-price-levels-amid-continued-slump-for-ai-favorite-deepseek-tariffs-8784704
iii. https://en.wikipedia.org/wiki/DeepSeek#
iv. Bob Veres Insider Information
https://www.bbc.com/news/articles/c5yv5976z9po
https://www.cbsnews.com/news/what-is-deepseek-ai-china-stock-nvidia-nvda-asml/
https://www.computerworld.com/article/3813072/alibaba-introduces-qwen-2-5-max-ai-model-claims-edge-over-deepseek.html
The opinions expressed above are solely those of Kondo Wealth Advisors, Inc. (626-449-7783 info@kondowealthadvisors.com), a Registered Investment Advisor in the state of California. Neither Kondo Wealth Advisors, Inc. nor its representatives provide legal, tax or accounting advice.