DeepSeek V4 Flash drives China's lead in global AI token usage

New usage data cited by NDTV Profit puts DeepSeek V4 Flash at the top of global model consumption with 4.63 trillion tokens, a figure that anchors a broader narrative: China has taken the crown in global AI token usage, edging out US rivals on price and margins. The driver is cost — DeepSeek and peers like MiniMax benefit from cheaper domestic energy and aggressively efficient model architectures, letting them offer inference at prices US labs struggle to match.

The V4 Flash is the cost-optimized tier of DeepSeek's V4 family, tuned for high-throughput, low-cost serving rather than maximum capability. Its dominance in raw token volume reflects how much real-world workload is price-sensitive commodity inference rather than frontier reasoning.

The data lands the same week DeepSeek closed a ~$7.4B funding round at a $50B+ valuation and the US reportedly held off blacklisting the company. For Western incumbents, the signal is uncomfortable: OpenAI is reportedly weighing 'drastic price cuts' anticipating a price war, and the token-share figures suggest the low end of the market is already tilting toward Chinese open and hosted models. Caveats apply — token counts measure volume, not revenue or value, and methodology behind cross-vendor token tallies is rarely transparent.