DeepSeek: Who Are the Big Winners?
DeepSeek has shaken up the AI industry by offering a powerful, cheap AI model that challenges OpenAI and Nvidia. With innovative training, open availability, and extremely low prices, the Chinese upstart could change the rules of the game for the entire industry.
Chinese AI challenger DeepSeek has dropped a bomb on the AI industry, sparking stock market panic, political reactions, and uproar. The company's latest AI model, DeepSeek R1, quickly became one of the most downloaded AI apps in the world, even passing OpenAI's ChatGPT on the App Store. What makes DeepSeek especially notable is its low development cost, innovative training methods, and accessibility - factors that could change the entire AI landscape.
Cheap Yet Powerful AI Model
One of the biggest surprises with DeepSeek is how cheap it's been to train the models. By improving memory efficiency and removing human feedback from the training process, the company has drastically reduced costs. According to DeepSeek's own calculations, the latest training run of the V3 model cost less than $6 million - a fraction of what it normally costs to develop top models like those from OpenAI.
That figure doesn't include all prior research and development, but the low cost indicates that top-tier AI models can be trained for significantly less money than the industry previously believed.
New Hardware Solutions Challenge Nvidia
DeepSeek used Nvidia's H800 chip, a scaled-down version of the company's most powerful AI hardware, since the US has banned the export of top-tier models to China. This forced DeepSeek's developers to optimize communication between the chips, resulting in a more efficient training process.
The consequences of this were seen immediately on the stock market, where Nvidia's stock fell sharply. DeepSeek has shown that it's possible to train world-class AI models without using the most expensive, most powerful chips, something that could change the rules of the game for the entire AI industry.
Open AI Model Pushes Prices Down
DeepSeek's models are open, meaning anyone can download, modify, and build on them. This puts them in direct contrast to OpenAI, which offers its most advanced models behind paywalls.
On top of that, DeepSeek's app is completely free, unlike ChatGPT, which requires payment for the latest versions. The company also offers an API with extremely low prices - 55 cents per million input tokens and $2.19 per million output tokens, compared to OpenAI's $15 and $60 respectively for equivalent services.
Meta's AI chief Yann LeCun commented on the development with these words:
"If you think this means China is outpacing the US in AI, you're wrong. The correct interpretation is that open-source models are now outperforming proprietary ones."
China's Strategic AI Move
DeepSeek's breakthrough is also a geopolitical milestone. The AI race between the US and China has gained a new dimension, with DeepSeek seen as a Chinese response to US dominance in AI development.
At the same time, the company has a clear connection to the Chinese state, reflected in the fact that its chatbot censors questions that could be sensitive for the Chinese government. That aspect may be less relevant for many companies, though, since most use AI to build specific applications rather than to seek impartial information.
Consequences for the AI Market
DeepSeek's breakthrough could have far-reaching consequences for the AI industry:
Nvidia could be affected, but will be fine in the short term
Even though DeepSeek has shown that cheaper chips can be used to train AI models, demand for AI hardware will keep growing. Tech giants will step up their AI investments to avoid falling behind.
Tech giants' cloud services won't be significantly affected
For companies like Microsoft and Amazon, which run the cloud services AI is used on, it matters less which models customers use - as long as they do it on their platforms.
AI companies like OpenAI, Anthropic, and Mistral are the big losers
DeepSeek has pushed prices down close to zero, which is a threat to companies whose business model relies on selling generative AI as a product. This could force these companies to radically change their business strategies.
The US will likely tighten restrictions on China
DeepSeek's success could lead to new sanctions and restrictions from the US, especially if the company is seen as a threat to American AI dominance.
Ordinary companies and startups are the big winners
Generative AI is becoming a standardized commodity, meaning companies that want to build AI solutions get access to powerful technology for a fraction of the cost. Ironically, it's a Chinese company that's now democratizing AI development.