Mirastack Labs shifts AI pricing from token counts to time-based capacity

Moving beyond the token

Most current AI services use a metered API model where users are billed for every single token generated. This means that every time an AI agent runs a loop or a developer conducts an experiment, it adds a specific cost to the bill. Mirastack Labs is challenging this standard by selling AI capacity by time instead of by unit.

As mirastacklabs.ai reported, the company hosts and runs open-weight models on its own hardware to provide this service. Their goal is to offer a model similar to how broadband internet is sold to the public. In this setup, you pay for a specific window of access rather than being charged for every piece of data processed.

This approach is designed to stop users from having to ration their thinking. When every single interaction has a price tag, it can limit the scope of what a system can do. By moving to a time-based model, the cost is established before the work begins, allowing for more fluid development.

Flexible service tiers

The company offers four distinct ways for different types of users to access their capacity. For individual developers and coding agents, they provide a flat time pass with unlimited tokens on open models. This is a self-serve option that can be set up in just a few minutes via their endlesstokens.ai platform.

For larger organizations with at least 60 seats, they offer a team-based model. This is billed per seat by time rather than by token and is structured as a quarterly or yearly agreement. This allows a whole workforce to use open models without worrying about fluctuating costs from high-volume usage.

More advanced needs are met through their infinitetokens.ai platform and their sovereign AI factory services. These options provide reserved inference and training clusters for specific endpoints or even full AI factories located in a customer's own data center. These services are billed for the platform and the term, ensuring that the price remains fixed regardless of the volume of tokens used.

Predictability for the enterprise

One of the primary drivers for this business model is cost predictability. In a token-based system, it is difficult to forecast exactly how much a complex AI project will cost as it scales. Mirastack Labs aims to solve this by having costs agreed upon upfront, whether that is based on seats or a specific term of service.

This is particularly relevant for teams that run AI for a living, such as research and model engineering teams. These groups often need to perform training, tuning, and evaluation at scale. When costs are fixed, these teams can run more intensive tests without the fear of a runaway bill from a high-volume experiment.

The company also targets regulated financial institutions and the defense or public sectors. These industries often require AI that stays inside strict audit and data boundaries. By providing sovereign AI that runs on-premises or in-country, they offer a way to maintain security while still benefiting from a predictable pricing structure.

Sovereign and private AI

For critical infrastructure operators and sovereign data center operators, the company offers a high-security 'AI Factory.' This service allows organizations to run AI on their own hardware or within their own domain so that nothing leaves their walls. This is a key requirement for many government and defense applications that require air-gapped systems.

The service is designed to be human-approved for systems that cannot afford to stop. By building these factories in-country and maintaining them, the company helps operators meet strict regulatory requirements. These setups are shaped specifically to the workload, including factors like location, capacity, and the specific term of the agreement.

This move toward 'sovereign' AI highlights a growing trend in the industry. As more organizations realize the importance of data privacy, they are looking for ways to run powerful models without sending sensitive information to a third-party cloud. Mirastack Labs positions its time-based capacity as a way to achieve this while keeping the budget stable.

The future of AI billing

The shift from tokens to time-based capacity represents a significant change in how AI resources are commoditized. While token pricing is easier for small, one-off tasks, it can become a hurdle for companies building complex, autonomous systems. Mirastack Labs is betting that the market will favor the stability of a 'window' of access over the volatility of per-unit billing.

This model may eventually become the standard for enterprise-level AI deployments. As companies move from experimenting with AI to integrating it into core workflows, the need for predictable overhead becomes paramount. By removing the cost decision from every single agent loop, Mirastack Labs is attempting to make AI more usable for production environments.

The success of this model will likely depend on how well it serves the diverse needs of their target audience. From a single developer needing a day pass to a government agency requiring a private factory, the goal is to provide a consistent experience. By focusing on capacity rather than units, they are trying to make the cost of intelligence more manageable for everyone.