An AI assistant by Chinese startup DeepSeek ousted OpenAI’s ChatGPT to become the most-downloaded free app on Apple’s U.S. App Store today, triggering market fluctuations and a rethink of Silicon Valley’s trillion-dollar AI playbook.

Founded by former hedge fund manager Liang Wenfeng, DeepSeek produces open-source large language models like DeepSeek-V3 to rival OpenAI’s GPT-4. DeepSeek operates at a fraction of the cost of their Silicon Valley counterparts, with the V3 model built for only $5.5 million — a rounding error compared to the billions poured into Western AI models. DeepSeek built its model on a budget and using restricted Nvidia chips, proving that the AI race isn’t just about who has the deepest pockets.

Founded by Liang in May 2023, DeepSeek operates out of Hangzhou, China, and enjoys the rare luxury of being fully funded by High-Flyer, the hedge fund that Liang previously led. This setup means the company can focus on AI research without worrying about quick returns.

One of DeepSeek V3’s standout features is the model’s ability to handle 128K tokens, making it better at long-context tasks. Thanks to a Mixture-of-Experts (MoE) system, the model activates only the parameters it needs for any given task, keeping the operation lean and efficient. Multi-Head Latent Attention (MLA) tackles nuanced relationships, and DeepSeek’s pure reinforcement learning approach creates a cycle of constant self-improvement.

DeepSeek’s biggest differentiator is affordability. With 95% lower costs per token compared to OpenAi’s GPT-4, and training tasks completed in just 2.8 million GPU-hours, DeepSeek has the potential to make high-performance AI accessible to more people than ever before.

The use cases for DeepSeek’s language models are broad, from coding and debugging in software development to personalized education tools. By staying open-source and cost-effective, DeepSeek is challenging the AI industry’s big players.

Silicon Valley elites seem equal parts impressed and nervous by the latest DeepSeek AI milestone. Satya Nadella, Microsoft’s CEO, called DeepSeek’s models “seriously impressive.”

Alexandr Wang, CEO of Scale AI, warned that this is America’s wake-up call, calling for more export control in a recent post on X:

VC legend Marc Andreessen labeled DeepSeek’s breakthrough the “AI Sputnik moment.”

Meanwhile, OpenAI’s Sam Altman commented, “It’s relatively easy to copy something that you know works. It’s extremely hard to do something new, risky, and difficult when you don’t know if it will work.”

Nvidia CEO Jensen Huang has not directly commented on DeepSeek’s efficiency claims. However, he recently made statements about quantum computing that some analysts view as potentially deflecting attention from DeepSeek’s achievements.

Meta’s AI Infrastructure Director, Mathew Oldham, is reportedly worried that the next version of Llama, Meta’s flagship AI, might not stack up against DeepSeek, according to two employees close to the matter interviewed for a recent article in The Information.

On Friday, Meta CEO Mark Zuckerberg upped the ante on the company’s AI ambitions, announcing that Meta’s spending on capital projects will soar by 60% this year, reaching somewhere between $60 billion and $65 billion, even as the company plans to lay off 5% of its workforce by February 10. 

Former Google CEO Eric Schmidt has been skeptical about China’s AI progress in the past — last May, he declared that the U.S. led China in AI by “two to three years, which in my world is eternity,” dismissing Europe as irrelevant and citing China’s chip shortages as a fatal flaw. By November 2024, after DeepSeek’s R1-Lite-Preview and Alibaba’s Qwen2.5-72B models were released, Schmidt admitted he was shocked and acknowledged that China is closing the gap despite the chip sanctions.

“I thought the restrictions we placed on chips would keep them back.” Schmidt remarked. A month later, at a Harvard forum, Schmidt doubled down, stating that “the U.S. is falling behind,” and crediting China’s state subsidies, chip breakthroughs, and foundational research focus.

By January 2025, Schmidt had a sobering take on ABC’s This Week: “I used to think we were a couple of years ahead of China, but China has caught up in the last six months in a way that is remarkable.”

Aravin Shrinivas, the co-founder and CEO of AI company Perplexity, said recently that chip restrictions helped DeepSeek’s breakthrough achievement in efficiency. “Necessity is the mother of invention,” he said, adding, ”Because they had to figure out workarounds, they actually ended up building something a lot more efficient.” Shrinivas also said that since DeepSeek’s models are open source, his company is already experimenting with them.

Renowned AI guru Kai-Fu Lee said in a post on X: “In my book AI Superpowers, I predicted that US will lead breakthroughs, but China will be better and faster in engineering. Many people simplified that to be ‘China will beat US’. And many claimed I was wrong with GenAI. With the recent Deepseek releases, I feel vindicated.”

Midjourney founder David Holz pointed out that DeepSeek has another, less obvious advantage in processing complex topics related to ancient Chinese philosophy and literature, as a large language model trained on long and continuous textual corpus:

Does DeepSeek’s meteoric rise signal a paradigm shift in the global AI landscape, challenging the assumption that innovation depends solely on capital and cutting-edge resources? By embracing open-source models and cost efficiency, DeepSeek has demonstrated that necessity breeds innovation, even under restrictive conditions.

As it bridges gaps between performance and accessibility, DeepSeek’s breakthrough highlights a growing shift in the balance of technological power. Whether this disruption will inspire collaboration or intensify competition remains to be seen, but one thing is clear: the AI race remains close.

Cover image from DeepSeek on X.