Neuralwatt launches energy-based pricing for AI inference, charging by kilowatt-hour instead of tokens to give customers transparent visibility into the true power cost of running AI models.

What happened: Neuralwatt Cloud is offering the first AI inference service priced by energy consumed rather than token count, at $5/kWh. The platform includes real-time per-request energy metrics, a dashboard showing usage trends, and model efficiency comparisons at no extra charge. It features an OpenAI-compatible API with median time to first token of less than 50ms and claims to be 40% more energy efficient than alternatives.

Why it matters: Traditional token-based pricing hides the actual resource consumption and cost drivers behind AI inference. By charging directly for kilowatt-hours consumed, Neuralwatt lets customers see exactly which models and prompts consume the most power, making it possible to optimize workloads and compare energy efficiency across different models—information that was opaque under conventional pricing schemes.

What to watch: Neuralwatt offers both a hosted Cloud service and Neuralwatt Deploy for on-premises use in customers' own data centers. New users can start with $5 in free credits and choose between per-kWh or per-token pricing. The platform supports multiple open-source LLMs through a single API.