A few weeks ago, DeepSeek, a Chinese AI research lab, introduced its open-source AI model, DeepSeek-R1, which has captured significant attention within the technology sector. According to a research paper released by the lab, DeepSeek-R1 has demonstrated superior performance compared to leading AI models like OpenAI’s o1 and Meta’s Llama across a range of benchmarks.
This achievement is particularly remarkable, considering that DeepSeek-R1 is not only open-source and cost-effective but also requires far less computational power than its competitors. This development could have wide-ranging implications, especially for Nvidia, a dominant player in the AI hardware market, as it sets the stage for possible disruption in Silicon Valley’s approach to AI development.
DeepSeek’s Innovative Approach
DeepSeek’s approach to AI research represents a fundamental shift in the traditional reliance on powerful hardware. The company has reportedly redesigned AI model foundations with an emphasis on software-driven resource optimization, aiming to reduce the dependence on hardware. Although DeepSeek still uses tens of thousands of Nvidia’s H100 and H200 AI GPUs for training its models, it has faced significant challenges due to U.S. export controls that limit access to the latest chips. To overcome these constraints, the lab has implemented a series of innovative engineering adjustments. These include custom communication schemes between chips to enhance data transfer efficiency, memory-saving techniques, and reinforcement learning strategies that minimize the computational power required.
These optimizations have enabled DeepSeek to dramatically lower the costs associated with developing large AI models, in stark contrast to the traditional, more hardware-intensive approaches used by other companies in the industry.
DeepSeek-R1 Pricing and Market Challenges
The cost-efficiency of DeepSeek-R1 is evident in the model’s API pricing structure. The cost for using DeepSeek-R1 is significantly lower than that of competitors. At just $0.55 per million input tokens and $2.19 per million output tokens, DeepSeek offers a far more affordable alternative to OpenAI’s API, which charges $15 per million input tokens and $60 per million output tokens. However, while this pricing structure may give DeepSeek an edge in terms of cost, the company’s commercial growth could face obstacles.
The U.S. export ban on advanced chips could limit DeepSeek’s ability to scale up its operations, and geopolitical tensions between the U.S. and China could also create trust issues among potential users of the model, particularly in Western markets. These factors could hinder DeepSeek’s penetration outside of China, posing challenges to its global ambitions.
DeepSeek and Implications for Nvidia and the AI Hardware Industry
DeepSeek’s innovative approach could disrupt traditional AI strategies, forcing big tech companies to reconsider their reliance on massive investments in computing power. Nvidia, which has seen substantial growth due to the AI boom, could be significantly impacted by this shift. The company’s GPUs have been regarded as the best for training and deploying AI models, and as a result, Nvidia has benefited from the explosion in AI development, with its revenue increasing by over 125% in fiscal year 2024, reaching $61 billion, and its net margins approaching 50%.
However, if more companies follow DeepSeek’s lead and adopt cost-efficient, resource-optimized models, demand for high-powered AI hardware could slow down. This change would be a significant shift in the AI ecosystem, which has been under economic strain, with many of Nvidia’s customers struggling to generate returns on their investments. If companies begin to gravitate towards cheaper AI models, it could significantly reduce the demand for Nvidia’s GPUs, leading to a potential decline in the company’s revenue and stock performance.
Volatility of Nvidia Stock
Nvidia’s stock performance over the past four years has been marked by significant volatility, with annual returns swinging dramatically. In 2021, Nvidia saw a remarkable 125% increase in stock price, followed by a 50% drop in 2022. The company rebounded in 2023 with a 239% gain, and in 2024, its stock rose by 171%. Despite the impressive gains, the stock has not been a smooth ride, with its performance often diverging from the broader S&P 500 index.
The uncertainty surrounding the macroeconomic environment, including rate cuts and geopolitical tensions, raises the question of whether Nvidia will face a similar downturn in the coming year, like it did in 2022, or if it will continue its upward trajectory. For those seeking less volatility, portfolios such as the Trefis High Quality Portfolio, which includes a selection of 30 stocks, have offered more consistent performance and have outperformed the S&P 500 over the same period.
Also Read: DeepSeek AI Disrupts US Tech Stocks, Nvidia Faces $432 Billion Market Cap Loss