DeepSeek’s $6 million R1 cost-efficient model training might be a ruse — the Chinese startup reportedly spent $1.6 billion and bought 50,000 NVIDIA GPUs while its top researchers earned $1.3 million