The AI player that no one saw coming and can keep it all
While everyone was wondering which of the big techs would be the winner in the race for artificial intelligence (AI) — OpenAI, Google, Meta, Anthropic — an unexpected player entered the arena and kicked the board without anyone noticing… at first… Hello DeepSeek. What really happened?
The rise of DeepSeek
DeepSeek, a Chinese AI lab, did the unthinkable: it launched DeepSeek V3, a world-class language model (LLM), designed to compete with ChatGPT and the rest of the players. The result was impressive in performance, but that wasn’t what shook the world; what was truly shocking was its ridiculously low cost.
They trained it for less than $6 million. To put that into perspective, Meta spent ten times that amount on its AI model, Llama. DeepSeek accomplished the same thing at a tenth of the cost and has basically the same level of power. At the time, DeepSeek had taken a giant leap into the AI economy, and, surprisingly, no one noticed. No one talked about it.
DeepSeek strikes the 2nd blow: its reasoning model
If you thought the story ended there, wait until you see what happened a month later.
DeepSeek did it again. They released DeepSeek R1, a reasoning model capable of breaking down complex questions step by step, analyzing every angle until it comes up with a final answer. Its performance was such that it could easily take on OpenAI’s O1. This means competing head-to-head with the best of OpenAI and its GPTs.
By then, it was clear that DeepSeek hadn’t just gotten lucky. It was real and was already executing a pattern.
The markets ignored the signal until a week later, when the truth hit like a bullet train. The biggest casualty? NVIDIA, which is the world’s leading company in AI computing, for providing the high-end chips that the models need.
If you’re selling bottles of water in the desert and someone suddenly discovers a river where you can scoop up water with your bare hands, your business is in trouble. That’s what happened with NVIDIA: DeepSeek didn’t need the premium chips that Nvidia sells at 90% margins to American companies. They used the limited version, the one they could barely get because of US sanctions… and still managed to produce a high-caliber model.
Who would want to pay a premium today when it has been proven that lighter products can serve just as well? This brutally caused NVIDIA to collapse in the market.
The news spread like wildfire and DeepSeek is one of the most downloaded apps in the last week.
The future of AI is uncertain
The question is no longer whether DeepSeek is real. It is.
They’ve shown that you can train world-class AI models. You can do it with open source, and you can do it on an absurdly low budget.
If they can do it a third, fourth or fifth time, the AI industry has changed hands forever.
Are we witnessing the beginning of the end of OpenAI and the West’s dominance in AI? Is this the start of a revolution where those big techs that we all expected to take over abruptly lose control of the game?
The only thing that is certain is that DeepSeek is not finished. And now, the whole world is watching.
