OpenAI’s work in the tech industry led to the current situation where companies are competing with each other to put the most powerful AI in the palm of our hands. Now, DeepSeek could be replicating a similar situation, but in terms of lowering the costs of developing powerful AI models. However, OpenAI suspects that its AI models were illegally used to train DeepSeek’s.
DeepSeek’s impact on the current AI industry and big companies’ shares
The emergence of DeepSeek shook the foundations of the artificial intelligence industry as we know it. The Chinese firm presented models capable of matching—or even surpassing—many other well-known and established ones. That said, DeepSeek’s real impact stems from the cost-performance ratio they achieved. The company claims that they only spent around $6 million training their AI models. This is just a small fraction of the millions/billions that other big names have invested so far.
According to the team behind DeepSeek, the low cost is due to the fact that they used thousands of older NVIDIA H800 chips instead of more modern and expensive hardware to train the R1 and V3 models. The development resulted in a loss of more than 17% of NVIDIA’s stock value in just one day. This is equivalent to around $600 billion in capitalization.
But what if DeepSeek resorted to certain “shortcuts” that it did not talk about when launching its models? That is the suspicion currently held by OpenAI, ChatGPT’s parent company. OpenAI and Microsoft are currently investigating whether DeepSeek used the OpenAI API to integrate GPT models with their own.
Microsoft and OpenAI reportedly have evidence of DeepSeek using GPT to train its models
Microsoft security researchers told Bloomberg that they detected a large-scale data exfiltration from OpenAI developer accounts in late 2024. The Redmond giant suspects that these developer accounts were affiliated with DeepSeek. On the other hand, OpenAI told the Financial Times that it found evidence of DeepSeek using data distillation. Distillation is a technique that allows large datasets to be “compressed” into smaller ones for use in training AI models.
In this case, OpenAI and Microsoft suspect that DeepSeek used distillation to integrate GPT models into their own. This would have saved them all the millions that OpenAI has invested in artificial intelligence training. Although OpenAI allows third-party developments on its platform through its API, the usage policies strictly prohibit data distillation.
“We know PRC (China) based companies — and others — are constantly trying to distill the models of leading US AI companies,” OpenAI’s statement reads. “As the leading builder of AI, we engage in countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models, and believe as we go forward that it is critically important that we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology,” adds the company.
Suspicions have already reached Trump’s cabinet
David Sacks, the one chosen by the new administration in Washington to handle AI and crypto in the country, told Fox News that “there’s substantial evidence that what DeepSeek did here is they distilled knowledge out of OpenAI models and I don’t think OpenAI is very happy about this.”
