Summary
DeepSeek released R1 in January 2025. OpenAI raised concerns about AI companies training on another model’s answers. Public statements linked those concerns to DeepSeek, but they were allegations. This article looks at the training method and market response. It does not establish that DeepSeek used OpenAI’s answers.
Introduction
DeepSeek’s R1 launch raised questions about its training and price. OpenAI and U.S. officials questioned whether firms had trained models on U.S. AI answers. These were January 2025 claims. They do not prove DeepSeek used OpenAI’s system or answers.
OpenAI’s Claim About Others Training with Its AI Answers
AI distillation means using one model’s answers to train another model. A model is software that produces AI answers. OpenAI’s statement at the time said:
“We are aware that entities in the People’s Republic of China are actively trying to employ techniques… to duplicate advanced AI models from the U.S.,” OpenAI said in a statement.
The statement did not name DeepSeek or establish that its team used OpenAI’s answers. Reports discussed the statement alongside DeepSeek’s release, but the claim remained an allegation.
DeepSeek-R1: The AI System That Drew Attention
DeepSeek announced R1, an AI model designed to work through some problems step by step, on January 20, 2025. The release included model weights, or values the AI learned, and technical details. DeepSeek presented R1 as another model to compare with OpenAI’s o1. Read DeepSeek’s release.
At launch, DeepSeek’s API prices were lower than those of several rivals. Prices can change. Check current vendor pricing before choosing a model.
The market reaction
On January 27, 2025, technology stocks fell as DeepSeek’s rise drew attention to AI costs. Nvidia’s market value passed $3 trillion in 2024. Reuters said Nvidia’s market value fell by about $593 billion that day. At the time, it was the largest one-day loss for a U.S. company. This is a historical figure, not a current market measure. Read the Reuters report.
President Donald Trump called the moment a “wake-up call” for the U.S. tech industry. The drop raised questions about data-center costs. One trading day did not show whether demand for AI chips would last.
Public officials’ comments
David Sacks was an adviser on AI and cryptocurrency. At the time, he told Bloomberg Law that he saw “substantial evidence” against DeepSeek. That was his assessment, not a court finding. Read the Bloomberg Law report.
Possible Benefits for Business Owners
For business owners, the story is a reason to compare AI models on the tasks they use.
- Test the same work in more than one model and compare accuracy, time, and cost.
- Check each provider’s data policy before sharing business or customer information. Ask what the provider does with your data and how long it keeps it.
- Keep a fallback option for workflows that depend on one vendor.
OpenAI released the lower-cost o3-mini shortly after R1 launched. Buyers had another model to compare.
What to Remember
The 2025 story linked AI-training claims with a sharp market response. Those claims were not court findings. For copyright and AI training, see our overview of Meta’s AI copyright controversy. Our AI tools guide offers practical model-selection steps.
