Research exposes DeepSeek’s AI training cost is not $6M, it’s a staggering $1.3B
SMRTR summary
DeepSeek's AI training costs are much higher than previously thought, reaching $1.3 billion instead of $6 million. The company operates 50,000 mixed-model GPUs and employs innovative Multi-Head Latent Attention technology, reducing inference costs by 93.3%, but faces challenges from U.S. export controls and competition from more economical alternatives like Google's Gemini Flash 2.0.
SMRTR provides this summary for quick context. The original article belongs to Interesting Engineering.
Read the original article