Gartner forecasts that inference costs for a 1-trillion-parameter model will fall more than 90% by 2030 versus 2025, driven by hardware improvements, model efficiency, edge inference, and specialized chips, making models under 100 billion parameters relatively commoditized. However, Gartner argues enterprises will not capture these savings directly: frontier labs will retain margin to offset current losses, and the shift toward agentic AI tasks consumes 5 to 30 times more tokens per query than standard generative AI, pushing enterprise spend upward even as per-token rates decline. For financial institutions, this structural dynamic suggests that enterprise AI spend may not compress even as unit costs fall, complicating ROI models that assume cost reduction as a primary benefit.
353 other tracked developments touching the same claims.