DeepSeek Shakes Up AI Market with Low-Cost Innovations

DeepSeek, a China-based AI research lab, originated from High-Flyer, the nation’s top hedge fund, in 2023. The fund has utilized AI for developing trading algorithms over several years.

The team at DeepSeek has discovered methods for creating large language models (LLMs) at a fraction of the cost incurred by major U.S. AI firms. This development sent shockwaves through the U.S. stock market on Monday, causing investors to reassess the future prospects for chip suppliers like Nvidia (NASDAQ: NVDA) and leading developers such as OpenAI, backed by Microsoft.

Stay Ahead of Financial News! Get the latest updates with Breakfast News delivered to your inbox every market day. Sign Up For Free »

Potential Game Changer in AI Competition

This moment may transform the AI landscape. DeepSeek’s innovations not only show great promise but there is also at least one other Chinese AI startup reportedly achieving comparable results. The implications for Nvidia and OpenAI could be significant.

Image source: Getty Images.

AI Models Becoming More Accessible

Ilya Sutskever, co-founder of OpenAI, had long believed that ample data and computing resources were essential for developing superior AI models. This notion, known as pre-training scaling, favoured well-funded developers with expansive data centers and cutting-edge chips. However, in November 2024, he revealed to Reuters that the effectiveness of this method has plateaued.

Since then, OpenAI has pivoted to focusing on models that excel in “reasoning,” spending additional time to generate better responses, known as test-time scaling. The latest versions (GPT-4o1 to GPT-4o3) have shown marked improvement in solving complex problems, bringing AI closer to resembling human intelligence on an academic scale.

OpenAI invested around $20 billion to achieve these advancements, largely funded through investors since 2015. In stark contrast, DeepSeek’s recently launched V3 model, produced for just $5.6 million, demonstrates competitive performance against OpenAI’s GPT-4o models across various benchmarks.

After the U.S. restricted Nvidia from selling its latest GPUs to Chinese companies, DeepSeek leveraged older models like the H100 and H800 to develop V3. To make up for reduced hardware capabilities, DeepSeek innovated with more efficient software algorithms and data processing techniques.

Moreover, DeepSeek employed a method known as distillation, training a smaller model using existing models like GPT-4o1 to generate a similar outcome. This approach accelerates the training process of competitive LLMs and hints at future commoditization. This suggests the market may see a proliferation of hundreds of LLMs with similar features that are often interchangeable.

This scenario poses a genuine risk for OpenAI and Nvidia. OpenAI may struggle to sustain its edge, while reduced demand for LLM training could impact Nvidia negatively.

DeepSeek’s Cost Advantages Highlight Industry Changes

Training efficiency is just one aspect of AI development; inference—the process of generating accurate responses based on prompts—also plays a crucial role. Typically, lower costs can lead to lower prices for consumers.

Currently, DeepSeek charges only $0.14 per 1 million input tokens, significantly undercutting OpenAI’s price of $2.50 for the same amount. Input tokens are calculated based on the total number of words in a user’s prompt.

DeepSeek is not alone in this landmark shift. Kai-Fu Lee, former head of Google operations in China, launched 01.ai, which claims to have models that rival those from DeepSeek. They charge just $0.10 per 1 million input tokens, even cheaper than both DeepSeek and OpenAI’s rates.

Implications for OpenAI and Nvidia

The trend towards commoditization presents significant challenges for OpenAI. Its closed-source models restrict developer flexibility, potentially discouraging users once more competitive open-source models become widely available.

In contrast, DeepSeek’s open-source strategy allows developers to customize models to meet their specific needs and securely manage their data.

While OpenAI grapples with uncertainties, Nvidia may benefit from reduced inference costs, which could balance out diminished GPU demand in training scenarios.

Consider how the cellphone industry has evolved. When users faced costs for every text message or internet activity, usage was limited. Now, unlimited plans allow consumers to use their phones extensively for a small monthly fee, illustrating how low costs can drive higher usage.

AI could follow a similar trajectory. Increased usage may boost demand for Nvidia’s GPUs to manage inference tasks, especially as reasoning capabilities advance, necessitating greater computational power.

The short-term outlook remains uncertain. Will Nvidia’s clients cut back on data center spending as they refine their training approaches like DeepSeek has? Only time will tell, and insights should come with the start of the quarterly earnings season as companies report their results.

Is Investing in Nvidia a Smart Move?

As you consider investing in Nvidia, take note:

The Motley Fool Stock Advisor analyst team recently identified their view of the 10 best stocks to purchase now…and Nvidia didn’t make the list. The featured stocks may offer substantial returns in the years to come.

Imagine, had you invested $1,000 in Nvidia back when it was recommended on April 15, 2005, you’d currently have $758,099*

Stock Advisor provides a straightforward guide for investors, including insights on portfolio building, regular updates from analysts, and two new stock suggestions each month. The Stock Advisor service has delivered returns over four times that of the S&P 500 since 2002*.

Discover more »

*Stock Advisor returns as of January 27, 2025

Suzanne Frey, an executive at Alphabet, serves on The Motley Fool’s board of directors. Anthony Di Pizio holds no positions in the stocks listed. The Motley Fool maintains positions in and recommends Alphabet, Microsoft, and Nvidia. It also recommends long January 2026 $395 calls on Microsoft and short January 2026 $405 calls on Microsoft. For transparency, the Fool has a disclosure policy.

The views and opinions expressed herein are those of the author and do not necessarily reflect those of Nasdaq, Inc.