Runpod Report Shows Developers Adapting to Higher GPU Costs

Runpod said in a press release that rising GPU prices and limited data center capacity are pushing developers to optimize existing infrastructure. Its Fall 2026 State of AI Compute report uses anonymized platform data from 1.2 million users across 186 countries, collected from April 1, 2025 through August 31, 2026.

Quantized models accounted for 40.7% of Pods running models with more than 70 billion parameters. The share was 32.5% for models with 8 billion to 70 billion parameters and 11.6% for models below 8 billion parameters.

Four in five Pods that referenced a frontier model API also ran local weights or a local serving stack. Among Pods using agent frameworks, 63% combined external frontier APIs with locally hosted open models. Qwen appeared on 74.2% of text endpoints, while Qwen3.x usage increased from 24.3% in November 2025 to 50.4%.

Resources created autonomously by agents generated 24% of Runpod revenue, up from 12% one month earlier. Runpod also reported a 31% increase in its B200 list price between January and August 2026. The company cautioned that its findings reflect activity on its own platform and do not represent all AI workloads.