DeepSeek Surpasses ChatGPT as Top App Store Download, As R1 Reasoning Model Beats OpenAI's o1

DeepSeek, a Chinese artificial intelligence startup, has taken the top spot on Apple’s U.S. App Store last weekend, surpassing OpenAI’s ChatGPT in downloads.

The milestone comes after the January 20 release of DeepSeek’s flagship reasoning model, R1, which has quickly gained recognition for its ability to rival advanced AI systems while operating on a fraction of the resources typically required.

DeepSeek R1 provides cutting edge performance, while being at the same time being censored according to CCP rules.

The R1-powered app’s rapid rise reflects DeepSeek’s innovative engineering and strategic use of Nvidia H800 GPUs, which are restricted for export to China due to U.S. sanctions.

By developing efficient training methods, the Hangzhou-based company has demonstrated that AI advancement is possible even under geopolitical constraints.

This development challenges the perception of U.S. dominance in artificial intelligence and raises questions about the effectiveness of export restrictions aimed at curbing China’s technological capabilities.

Building AI Under Restriction: A Resourceful Approach

DeepSeek’s R1 model was trained using just 2,048 Nvidia H800 GPUs at a total cost of under $6 million, according to a research paper the company released in December 2024.

These GPUs are intentionally throttled versions of the H100 chips used by U.S. companies like OpenAI and Meta. Despite the hardware limitations, DeepSeek’s engineers developed novel optimization techniques that allowed R1 to achieve results comparable to models trained on far more powerful infrastructure.

Founder Liang Wenfeng, a former hedge fund manager, explained the company’s approach during an interview with 36Kr. “We need to consume four times more computing power to achieve the same effect,”

Liang said. “What we need to do is continuously narrow these gaps.” Liang’s foresight in stockpiling Nvidia GPUs before U.S. restrictions took effect was a critical factor in the company’s ability to innovate under challenging circumstances.

DeepSeek’s engineers also focused on reducing memory usage and computational overhead, enabling high accuracy despite hardware constraints. Dimitris Papailiopoulos, a principal researcher at Microsoft’s AI Frontiers lab, highlighted the efficiency of R1’s design.

“They aimed for accurate answers rather than detailing every logical step, significantly reducing computing time while maintaining a high level of effectiveness,” he told MIT Technology Review.

Performance Benchmarks and Industry Recognition

R1’s performance has been particularly strong on technical benchmarks, earning scores of 97.3% on MATH-500 and 79.8% on AIME 2024. These results place R1 alongside OpenAI’s o1 series, demonstrating that DeepSeek’s resource-efficient model can compete with industry leaders.

Beyond its flagship model, DeepSeek has also released smaller versions of R1 capable of running on consumer-grade hardware. This accessibility has widened the model’s appeal among developers, educators, and hobbyists. On social media, users have shared examples of R1 handling complex tasks such as web development, coding, and advanced math problem-solving.

DeepSeek’s achievements have drawn praise from prominent figures in the AI field. Yann LeCun, Meta’s Chief AI Scientist, emphasized the role of open-source collaboration in DeepSeek’s success. ““DeepSeek has profited from open research and open source (e.g., PyTorch and Llama from Meta). They came up with new ideas and built them on top of other people’s work.” LeCun wrote on LinkedIn. Because their work is published and open source, everyone can profit from it. That is the power of open research and open source.”

Similarly, Marc Andreessen, co-founder of Andreessen Horowitz, described R1 as “one of the most amazing breakthroughs I’ve ever seen.” These endorsements highlight the global impact of DeepSeek’s resourceful approach to AI development.

Affordability and Open-Source Ethos

Unlike proprietary platforms such as OpenAI’s ChatGPT, DeepSeek has embraced an open-source philosophy. The company has made the R1 model’s weights, training recipes, and documentation publicly available, allowing developers worldwide to replicate or build upon its work. This transparency has set DeepSeek apart in an industry often characterized by secrecy.

Affordability has also been a key factor in R1’s popularity. The app is free to use, and API access is priced significantly lower than competitors’ offerings. These pricing strategies, combined with the model’s robust capabilities, have made DeepSeek an attractive option for individuals and businesses alike.

Geopolitical Implications of DeepSeek’s Success

DeepSeek’s rise comes at a time of heightened geopolitical tensions between the United States and China, particularly in the field of artificial intelligence.

Since 2021, the Biden administration has expanded restrictions on the export of advanced chips to China, aiming to limit the country’s ability to develop competitive AI technologies. However, DeepSeek’s achievements suggest that such measures may not fully prevent innovation.

The company’s success has prompted debates within U.S. technology circles about the unintended consequences of export controls. Some executives argue that these restrictions may be driving resourceful innovation among Chinese firms. Liang’s strategy of stockpiling GPUs and focusing on efficiency has proven that constraints can spur creative problem-solving rather than stifling it entirely.

A Broader Movement in Chinese AI

DeepSeek’s open-source approach aligns with a broader trend in China’s AI sector. Other companies, including Alibaba Cloud and Kai-Fu Lee’s 01.AI, have also prioritized open-source initiatives in recent years. Liang has described the need to address what he calls an “efficiency gap” between Chinese and Western AI ventures, explaining that local firms often require double the resources to achieve comparable results.

In July 2024, Liang stated, “We estimate that the best domestic and foreign models may have a gap of one-fold in model structure and training dynamics. For this reason alone, we need to consume twice as much computing power to achieve the same effect. In addition, there may also be a gap of one-fold in data efficiency, that is, we need to consume twice as much training data and computing power to achieve the same effect. Together, we need to consume four times more computing power. What we need to do is to continuously narrow these gaps.”

His leadership has earned DeepSeek recognition both within China and internationally. In 2024, he was invited to high-level meetings with Chinese officials to discuss strategies for advancing the country’s AI capabilities.

Future Challenges and Opportunities

As DeepSeek continues to refine its models, the company faces both opportunities and challenges. While its achievements have proven the viability of resource-efficient AI, questions remain about whether such approaches can scale to compete with the massive investments of tech giants like OpenAI and Meta.

In a post after the DeepSeek R1 release, Mark Zuckerberg, Meta’s CEO, has highlighted the importance of large-scale investments in AI infrastructure, sying “This will be a defining year for AI. In 2025, I expect Meta AI will be the leading assistant serving more than 1 billion people, Llama 4 will become the leading state-of-the-art model, and we’ll build an AI engineer that will start contributing increasing amounts of code to our R&D efforts. To power this, Meta is building a 2GW+ datacenter that is so large it would cover a significant part of Manhattan.

We’ll bring online ~1GW of compute in ’25 and we’ll end the year with more than 1.3 million GPUs. We’re planning to invest $60-65B in capex this year while also growing our AI teams significantly, and we have the capital to continue investing in the years ahead. This is a massive effort, and over the coming years it will drive our core products and business, unlock historic innovation, and extend American technology leadership. Let’s go build!”

For now, DeepSeek’s success with R1 has demonstrated that innovation is not solely the domain of the most well-funded players. By prioritizing efficiency, transparency, and accessibility, the company has made a lasting impact on the global AI industry.