DeepSeek turned into a market and technology story at the same time. A Chinese AI assistant climbed above ChatGPT in the iPhone App Store’s “Free Apps” category, while concern over its R1 reasoning model helped fuel a 17 percent drop in Nvidia stock on Monday.
The reaction was not just about one app becoming popular. It was about what DeepSeek R1 seemed to suggest: that a low-cost, freely available AI model could challenge the assumptions behind the current American AI business model.
What made DeepSeek R1 stand out
The key moment began around January 20, when Chinese AI startup DeepSeek announced R1. The company described it as a simulated reasoning model and claimed it could match OpenAI’s o1 in reasoning benchmarks.
Like o1, R1 is designed to work through a simulated chain of thought process before providing an answer. That approach can potentially improve accuracy or usefulness for some user questions, depending on the task and the prompt.
The existence of another reasoning model was not shocking by itself. Google is also pursuing simulated reasoning models, and OpenAI has announced an upcoming SR model called “o3” that can surpass o1 in performance.
What changed the conversation was the combination of cost, timing, and openness. DeepSeek allegedly trained R1 for only $6 million, reportedly about 3% of the cost of training o1, as a so-called “side project.” The company also used less powerful Nvidia H800 AI-acceleration chips because of US export restrictions on cutting-edge GPUs.
R1 also appeared just four months after OpenAI announced o1 in September 2024. Most importantly, DeepSeek released the model weights for free with an open MIT license, allowing anyone to download it, run it, and fine-tune it.
Why investors reacted so sharply
For American AI companies and investors, DeepSeek raised an uncomfortable question: what happens if the most advanced AI capabilities become cheaper, faster to reproduce, and easier to distribute?
Companies such as OpenAI and Google have so far built much of their AI strategy around proprietary, closed models. DeepSeek R1 pushed observers on social media to argue that those companies may have “no moat,” meaning that technological leadership, access to advanced hardware, or large funding reserves may not be enough to stop challengers.
The anxiety quickly moved beyond AI labs. Nvidia stock dove 17 percent on Monday amid worries over DeepSeek, even though DeepSeek used Nvidia chips for training. That contradiction is part of what made the sell-off notable: the market reaction reflected fear about the broader economics of AI, not simply a direct rejection of Nvidia hardware.
Several facts fed the concern:
- DeepSeek R1 was presented as a model that could challenge OpenAI’s o1 in reasoning benchmarks.
- It was allegedly trained for only $6 million.
- It used Nvidia H800 chips rather than cutting-edge GPUs.
- Its model weights were released for free under an open MIT license.
- The DeepSeek app overtook ChatGPT in the iPhone App Store’s “Free Apps” category.
That mix made DeepSeek feel less like a normal product launch and more like a challenge to the cost structure of the AI industry.
The App Store surge amplified the story
DeepSeek’s app gave the broader public an immediate way to test the company’s technology. The assistant lets users experiment for free with both R1 and DeepSeek’s V3 conventional large language model.
Over the weekend, the app jumped to the top of the US iPhone App Store. That visibility mattered because it put DeepSeek in direct comparison with ChatGPT for everyday users, not only researchers and investors.
The surge also spread across social media and Reddit. Multiple AI-related Reddit threads became filled with DeepSeek-related posts, which led to so-far unfounded accusations that someone in China was astroturfing support for the company.
Venture capitalist Marc Andreessen added to the attention on Friday when he wrote on X that DeepSeek R1 is “one of the most amazing and impressive breakthroughs I’ve ever seen” and a “profound gift to the world.” His comments helped push the discussion further into the mainstream technology conversation.
Open models versus closed models
Some observers framed DeepSeek as part of a larger US versus China technology contest. The source article notes that this framing may be overblown, according to some experts.
Meta Chief AI Scientist Yann LeCun, who often supports open-weights AI models and open source AI research, argued on LinkedIn that the more important issue is openness. He wrote, “To people who see the performance of DeepSeek and think: ‘China is surpassing the US in AI.’ You are reading this wrong. The correct reading is: ‘Open source models are surpassing proprietary ones.’”
That interpretation shifts the focus away from nationality and toward business models. If open-weight models continue improving, closed AI companies may face pressure from tools that are cheaper to access, easier to modify, and available outside a single vendor’s product ecosystem.
This is why DeepSeek R1 matters even if some claims require more formal evaluation. Benchmarks can be gamed, and they do not always show how a model performs in everyday scenarios. But a model does not need to be perfect to change expectations about what AI development should cost or how widely powerful tools can be shared.
What this means for AI’s next phase
Informal testing described in the source article found DeepSeek-V3 and DeepSeek-R1 to be roughly equivalent to OpenAI’s ChatGPT models, though results can vary dramatically depending on use and prompting. DeepSeek’s AI assistant can also search the web like ChatGPT through chat.deepseek.com.
The broader lesson is that AI may be following a familiar path in computing. Technologies that begin as expensive, specialized products often become cheaper, smaller, and absorbed into larger products. The source article compares this pattern with software components that were once sold separately and microprocessors that replaced large, expensive computer systems before becoming embedded into everything.
DeepSeek R1 does not settle the future of AI. It does, however, show why investors and AI companies are watching open weights, training costs, and reasoning models more closely. If a cheaply trained open weights AI model can match America’s best commercial models, closed-source AI companies face a real strategic threat.
The panic around DeepSeek may fade. The question it raised will not: how durable is an AI moat when capable models can become cheaper, more open, and easier for others to build on?