DeepSeek Questions the Traditional Thinking on AI Expenses

While Nvidia faced the largest one-day decline in stock market history, other major tech players like Microsoft, Alphabet, and Amazon, who are heavily investing in competing AI technologies such as ChatGPT and Gemini, also experienced significant losses.

The sell-off was triggered by investors’ astonishment at the alleged capabilities of DeepSeek’s new R1 chatbot, which was reported to surpass its competitors in advancement and be cheaper to develop.

DeepSeek R1 has skyrocketed, becoming the top free downloaded app on Apple’s app store, coinciding with a steep decline in US technology and related stock prices.

Why tech stocks plummeted

The market was taken aback by DeepSeek’s introduction of technology that promises lower costs while delivering comparable performance.

This shift has significantly altered market expectations regarding computing power, showcasing that more can be accomplished for less money. It has also put the US tech companies’ current AI products and advancements at a disadvantage.

Stock valuations are primarily influenced by market expectations. The proclaimed capabilities of DeepSeek R1 have led to a substantial reevaluation of what is technologically feasible and how affordably AI can be developed and operated.

Investors quickly adjusted stock prices to reflect the emergence of a low-cost Chinese AI competitor, anticipating that this new player could disrupt the market and reduce the competitive edge of current leaders.

Who is DeepSeek and what is R1?

Founded in 2023 by the Chinese hedge fund High Flyer, which has utilized AI in trading since 2021, DeepSeek specializes in developing large language models (LLMs) that serve as the foundation for chatbots and various AI-based tools. R1 represents the most recent evolution of DeepSeek’s chatbot and technology, building on prior generative AI models created by the company, while featuring a remarkably enhanced performance at a reduced cost.

Technology investors believe R1 either matches or outperforms competing solutions, such as OpenAI’s ChatGPT 4.01, across numerous benchmarks.

However, several key distinctions exist:

1. The R1 model operates in a less resource-intensive manner, resulting in lower development and operational costs, requiring less data and computational power.
2. Despite the US export ban that prevents Chinese companies like DeepSeek from acquiring chips from US firms like Nvidia, R1’s training was achieved. The Biden administration has enacted regulations restricting the sale of specific computer chips and equipment to China, aimed at inhibiting its access to advanced technology.
3. The training data and R1’s operational data are stored on servers located in China. Given existing concerns regarding data privacy and intellectual property related to US-based companies, having data under the jurisdiction of the Chinese Communist Party (CCP) raises even greater concerns.
4. The code for the chatbot program is available for free download, reading, and modification, in contrast to ChatGPT; however, this apparent transparency may be misleading, as the underlying model is of greater significance than the chatbot code itself.
5. R1 is known to censor its responses to align with the values of the Chinese Communist Party.

The future of AI and technology stocks

The current decline in tech stock prices raises questions about whether it reflects an irrational panic that may reverse or if it represents an accurate reassessment of value. The future costs and benefits associated with AI remain uncertain.

This situation poses both technological and economic questions.

From a technological perspective, it remains to be seen whether R1 actually necessitates fewer resources for computing and less data for training and use.

Econimically, certain stakeholders may emerge as winners while others face losses. AI users might benefit from more affordable AI access, particularly LLMs, which could bolster adoption rates and enhance productivity. In contrast, existing producers like Nvidia may face challenges in a market once characterized by limited competition.

On a broader scale, society could gain from less computationally demanding and thus more energy-efficient AI solutions. However, the geopolitical risks of a single country dominating the market, coupled with apprehensions related to data privacy, intellectual property rights, and censorship, may overshadow these advantages.

Michael J. Davern, Professor of Accounting & Business Information Systems, The University of Melbourne and Matt Pinnuck, Professor of Financial Accounting, The University of Melbourne

This article is republished from The Conversation under a Creative Commons license. Read the original article.