Reducing 'AI Costs' Is the Trend! Meta to Develop 'Model Routing' to Replicate OpenRouter

Under the pressure of continuously expanding AI infrastructure spending, Meta is seeking ways to cut costs from within.

On July 21, tech media outlet The Information reported that Meta’s internal AI incubator, AAI Labs, is developing an AI model routing tool named "Switchboard." Its core logic is highly similar to OpenRouter’s Auto Router product—by scoring task difficulty, it routes simple requests to cheaper, smaller models, thereby reducing overall inference costs.

Switchboard is currently in its early stages, and its final implementation remains uncertain. However, internal documents obtained by The Information show that the Meta team has clearly outlined two potential paths: one is to deploy it within the company to compress costs, and the other is to publicly release it to external organizations running AI programming agents at scale.

Analysts point out that this means the tool is not just a cost-cutting measure but could also become Meta’s attempt to open up new revenue sources in the AI tools market.

Furthermore, reports indicate that the proposal of this project directly addresses a realistic pain point currently faced by the company: paying top-tier model prices for every programming request, including those simple tasks that could be easily handled by smaller models.

Addressing the Pain Point: Inference Cost Is the Biggest Barrier to Scalable Deployment

The report states that the rationale behind the Switchboard project was expressed quite bluntly in internal documents: "We pay top-tier model prices for every programming request, including simple ones."

The document further pointed out that most programming agent tasks can be fully handled by smaller models, with only a few tasks truly requiring the capabilities of frontier large models. However, the current situation is that "all requests are sent to the same model, leading to overspending on simple tasks or insufficient performance on complex tasks."

The document explicitly characterized inference costs as the "primary obstacle" to wider deployment of agents within the company, stating plainly: "Cost is the key factor limiting our ability to run agents at scale."

This statement corroborates a series of recent moves by Meta to control AI spending. As previously reported by The Information, Meta notified employees in June this year that it would begin setting caps on AI token usage, just weeks after encouraging broader adoption of AI tools across the company, while simultaneously building an internal platform to track AI spending and enforce token budgets.

Notably, the Switchboard project falls under Meta’s AAI Labs, an internal incubator established under the Meta Applied AI Engineering team, which was officially founded in March this year. According to internal documents reviewed by The Information, this mechanism allows employees to submit proposals for AI products and services. Once approved, small teams are responsible for building them, with the opportunity for public release.

As of July this year, AAI Labs had approved approximately 200 projects, covering three main directions: consumer products, developer tools, and internal infrastructure. Switchboard is one of them.

This mechanism reflects Meta CEO Mark Zuckerberg’s broader strategic vision—leveraging AI to enable small teams to build products quickly. In April this year, Zuckerberg told analysts that AI agents mean "small teams can make very rapid progress," predicting that this technology would drive "significant innovation." He also stated that Meta might create as many as 50 new applications.

The Model Routing Race: Beyond OpenRouter, Giants Are Entering the Fray

The model routing sector targeted by Switchboard is attracting increasing attention.

OpenRouter has gained considerable popularity among developers by helping them access various AI models at lower costs. According to a report by The Information last week, OpenRouter has entered into discussions regarding a potential acquisition with a larger tech company, a deal that could push its valuation up by billions of dollars—the company was valued at $1.3 billion in April this year.

The concept of model routing has garnered wider attention due to the built-in routing feature released by OpenAI with GPT-5, which automatically switches to cheaper models when user prompts are relatively simple. Since then, companies such as Databricks and Palantir have also developed their own routing tools to manage costs and improve efficiency.

Meta’s self-developed Switchboard is both a proactive response to its own cost pressures and a strategic choice to establish its own capabilities in this rapidly heating up sector.

Greater Ambitions: Seeking Diversified Monetization of AI Investments

Behind the Switchboard project lies Meta’s overall demand to transform its massive AI investments into new tools, new businesses, and new revenue sources.

Meta previously estimated that its spending on AI infrastructure and other equipment and facilities could reach as high as $145 billion this year, more than doubling the 2025 level. Meanwhile, Meta is reorganizing its engineering teams to strengthen its AI development capabilities.

Projects incubated by AAI Labs go beyond Switchboard. According to another internal document obtained by The Information, AAI Labs is also developing an AI guide application for drivers, which can run via Apple CarPlay and Android Auto, using AI to explain nearby landmarks and allowing drivers to ask questions.

The document positions this product as an extension of the Instagram Maps experience, which may eventually integrate location-based Reels content, travel recommendations, and even Meta Ray-Ban smart glasses.

The report states that these projects collectively outline Meta’s path: starting with employee ideas, rapidly prototyping AI products, and then launching them to the external market when appropriate—exploring incremental revenue space beyond the advertising business while controlling costs.