Is "GPU-hour per user" a sane way to price LLM inference? (www.mirastacklabs.ai)

MIRASTACK LABS has introduced a novel pricing model for large language model (LLM) inference, shifting from the traditional token-based billing to a time-based framework. This approach allows users to purchase GPU capacity in time increments—from day passes for individual developers to comprehensive AI factories for enterprises—removing the constraints of token metering. By selling access to open-weight models on their GPUs in a manner similar to broadband services, developers can experiment and iterate without worrying about the costs stacking up with each token used. This model is significant for the AI/ML community as it enables more fluid and innovative use of AI resources, particularly for enterprises, research teams, and regulated sectors requiring stringent data controls. The predictable cost structure eliminates the anxiety around token consumption, encouraging teams to focus on creativity and experimentation. The offering resonates particularly with organizations that require on-premises solutions, such as defense, financial institutions, and critical infrastructure operators, looking to deploy AI safely within their own environments while maximizing computational efficiency.