Black day for NVIDIA: Chinese AI company DeepSeek is to blame

NVIDIA faces a black day on the stock exchange, recording losses double digit, and the trigger is DeepSeek, a Chinese firm Artificial Intelligence which, in addition to being an NVIDIA client, has burst into the market with groundbreaking claims. Likewise, its application has become the most downloaded of the iStore, surpassing even Chat GPTBut what exactly has this company done and what statements has it made to cause such a stir globally?

What triggered the NVIDIA stock sell-off?

In a nutshell, the cost of DeepSeek is significantly inferior to that of OpenAI and its most recent model, the 01. For example, the cost of 01 is $15 for a million input tokens y $60 for a million output tokens. In contrast, the model launched by DeepSeek, based on the R1 of the company, cost 0.55 dollars for a million input tokens y $2.19 for a million output tokens.

Esa abysmal difference The cost hike has caused a stir, as DeepSeek claims to have trained its model with an investment well below that required by other large companies such as OpenAI, Google o GOAL to perfect their solutions IA.

How did this affect NVIDIA?

Until now, the market and public opinion assumed that training AI models was extremely expensive, requiring billions of dollars invested in data centers with NVIDIA GPUs. That scenario had been beneficial for the company. Verde, since most of their current revenues are related to artificial intelligence and their stock value is supported by this demand. Not in vain are they considered pioneers in IA, and almost everyone uses CUDA.

AI Benchmark: DeepSeek R1/V3 vs OpenAI 01

What DeepSeek is proposing debunks the perception that training AI requires enormous resources, as they claim to have done so at a fraction of the cost of traditional US corporations.

Cost of training DeepSeek-V3

For market analysts, if this is true, the demand for AI-focused GPUs could decrease, since training artificial intelligence models It wouldn't be so expensive as expected. Therefore, NVIDIA could see its sales volumes reduced in the future, harming its income.

Distrust towards DeepSeek

As a Chinese company, there is no shortage of critics who question the claims about training costs. They point out that the figures provided may not reflect the actual expenditure DeepSeek incurred to achieve a competitive model against the OpenAI 01, with comparable results in benchmarks. Beyond the controversy, the fact is that the company charges much less than OpenAI currently offering their services at a much lower price.

Some experts think otherwise.

There is another sector of AI specialists who believe that, rather than decreasing NVIDIA's revenue, the reduction in costs in training models could have the opposite effect. If it is more cheap develop AI, more organizations could be interested in entering the market, increasing the demand for hardware specialized and, therefore, increasing GPU sales.

And with the U.S. government’s interest — led by President Donald Trump — in investing in AI infrastructure, the scale could become even larger. (Information about exact policies or government investments is subject to future change, so it’s worth staying updated.)

Source: Marketwatch