Chinese AI Startup DeepSeek Challenges U.S. Giants

Chinese AI Startup DeepSeek Challenges U.S. Giants with Low-Cost Breakthroughs

Developers at leading U.S. AI firms are both impressed by and skeptical of the rise of DeepSeek, a Chinese AI startup whose innovative models have rapidly gained prominence. DeepSeek’s achievements are raising questions about whether their low-cost technology could challenge the dominance of multi-billion-dollar systems developed by American companies.

On Monday, DeepSeek caused a stir in the U.S. stock market, sparking a selloff as its free AI assistant overtook OpenAI’s ChatGPT as the top-ranked app on Apple’s (AAPL.O) U.S. App Store. DeepSeek’s model was reportedly trained using Nvidia’s (NVDA.O) H800 processors, which are less powerful than higher-end chips, and achieved this with a training budget of under $6 million.

While some U.S. AI experts applauded the DeepSeek team’s strong research and innovative approach, others dismissed concerns, arguing that the startup’s methods remain limited compared to the capabilities of leading U.S. AI labs. According to individuals familiar with the thinking at four major AI firms, who spoke anonymously, the news hasn’t significantly shaken their confidence.

OpenAI CEO Sam Altman acknowledged the achievement, stating on X that DeepSeek’s R1 model—one of several released recently—was “an impressive model, particularly around what they’re able to deliver for the price.” Nvidia, for its part, noted that DeepSeek’s accomplishment demonstrates the increasing demand for its AI chips.

Adoption and Caution in the Industry
The U.S.-based software company Snowflake (SNOW.N) responded quickly, announcing it would include DeepSeek’s models in its AI marketplace after a surge of customer inquiries. Snowflake’s executive vice president of product, Christian Kleinerman, shared that while the company considered the risks of offering Chinese-developed AI, it ultimately chose to proceed, ensuring transparency with its customers. “We decided that as long as we are clear to customers, we see no issues supporting it,” Kleinerman said.

Meanwhile, U.S. developers are analyzing DeepSeek’s V3 model, which underpins its popular app. Though DeepSeek released a research paper detailing the model in December, questions remain about the full development costs. Experts noted that the $6 million budget cited in the paper referred specifically to the final training run, not the total expenses of the project.

The Hidden Costs of Development
DeepSeek’s V3 model was reportedly trained using 2,048 Nvidia H800 chips, designed to comply with U.S. export restrictions implemented in 2022. These controls, however, have done little to impede China’s progress in AI, according to industry experts. While the paper claimed the final training phase was completed on a relatively modest budget, sources at two major AI labs estimated that earlier stages of development likely required a far larger investment—potentially exceeding $1 billion.

“The training run is just the tip of the iceberg in terms of costs,” said executives from two leading AI labs. They highlighted that designing an optimal training process involves significant upfront expenses.

The Rise of Open Source AI
DeepSeek’s decision to release its models as open source has also garnered praise within the AI community. Open source technology allows other developers to freely modify or utilize the models, contrasting sharply with the tightly controlled, proprietary systems from companies like OpenAI.

Venture capitalist Marc Andreessen described DeepSeek’s R1 as “one of the most amazing and impressive breakthroughs I’ve ever seen—and as open source, a profound gift to the world.”

Industry experts noted that DeepSeek’s success demonstrates the viability of open source AI as a cost-effective alternative to the high-priced systems dominating the market. This has further challenged the assumption that only the wealthiest tech giants, capable of massive investments in infrastructure, can lead the AI race.

The Road Ahead
While DeepSeek has made significant strides, challenges remain. As excitement around its free release strategy grows, the company could soon face a shortage of chips to meet demand, one expert predicted.

Still, the emergence of DeepSeek underscores how quickly China has closed the gap with U.S. firms, narrowing the timeline from 18 months behind state-of-the-art models to just six months. This rapid progress is sure to intensify competition in the AI space, especially as Wall Street begins reevaluating the dominance of major American tech firms in the field.

With several U.S. giants set to release their quarterly earnings soon, the success of DeepSeek highlights a growing shift in the industry—one that could reshape the future of AI innovation worldwide.