Token Actual Cost Drops to $0.99, SemiAnalysis Reveals Big Die Consumption Accounts for 30% of Cost

According to Perceive Beating monitoring, AI-based Intelligent Entity widely adopted in the enterprise sector is disrupting the unit economics of the professional services industry. Research firm SemiAnalysis revealed that internal Big Model Token expenditure has accounted for 30% of total employee compensation, with an average individual monthly consumption of nearly 5 billion Tokens and core contributors consuming over 1 trillion Tokens per month. Tasks that originally took analysts hours to complete, such as Excel model conversions and financial report chart creation, can now be done in minutes at a cost of a few dollars in tokens.

The drastic reduction in actual usage costs is key to reshaping the unit economics of the professional services industry. Although Opus 4.7 has an official price tag of $5 per million Tokens for input and $25 for output, the Intelligent Entity's task efficiency of up to 300:1 input/output ratio and over 90% hit rate on prompts has brought the actual blended Token cost down to just $0.99 per million.

The combined acceleration of software and hardware is further reducing production costs. Running DeepSeek R1 on B300, the throughput of a single GPU has increased from the baseline of 1000 tokens/second to 14000 tokens/second through software optimizations like wideEP, disagg, and MTP, realizing a 14x pure software throughput improvement. At the hardware level, the optimized configuration of GB300 NVL72 achieves a throughput 17 times that of H100 (reaching 32x under FP4), providing a structural guarantee for the profit margin growth of large model developers and predicting that the Token price in 2027 will be significantly lower than current levels.