Anthropic's IPO is just the start — the market for AI model routing is booming

AI model company listings accelerate, exposing limits of the 'one best model' strategy

IDC, Gartner see multi-model routing as the low-cost, high-efficiency path forward

OpenRouter's estimated acquisition price jumps fivefold in three months as AI ecosystem expands

As AI model companies move toward initial public offerings, capital is also flowing into the routing business — technology that connects multiple AI models and selects the right one for each task. Anthropic is targeting a listing as early as late this month or early October, while OpenAI has filed confidentially for an IPO of its own. Competition in the AI market is expanding beyond model development to the question of how to deploy a growing roster of models efficiently.

Anthropic's listing is drawing attention because it will be the first time a major AI model company's value is assessed directly in the public market. The company submitted a draft registration statement to the US Securities and Exchange Commission (SEC) in June, with a late-September to early-October listing now under discussion. OpenAI recently filed its own IPO paperwork, making back-to-back public listings by large AI model companies an increasingly concrete prospect.

As the number of AI models has multiplied, simply picking "the single best model" is no longer enough to balance cost and performance. ChatGPT, Claude, Gemini and a widening field of alternatives now differ meaningfully in capability, price and response speed. Straightforward tasks such as document summarization or classification can be handled by relatively inexpensive models, while complex coding or reasoning calls for high-performance ones. Matching the model to the task is more efficient than running every workload through a premium model.

The technology drawing attention in this context is routing — a system that evaluates an incoming request and directs it to the most suitable AI model. Much like a traffic-management system that routes vehicles along the optimal path based on destination and road conditions, AI routing selects and connects the appropriate model based on the nature and complexity of each request. Simple translation or document classification goes to a cheaper model; complex coding or reasoning is sent to a high-performance one. Because the selection weighs price and response speed alongside raw capability, routing gives enterprises a practical tool for managing AI costs.

As AI model selection shifts toward combining multiple models, the foundation for routing market growth is widening. IDC forecasts that by 2028, 70 percent of enterprises that actively use AI will adopt a multi-model approach — selecting and connecting whichever model best fits each task. The expectation is that services combining several models will become commercially mainstream, moving away from the single-model-handles-everything paradigm.

The core of routing is assigning the right level of model to each task — cutting model costs while securing the performance each job requires. Gartner has analyzed that "inference tiering" — using inexpensive models for simple tasks and high-performance models for those requiring stronger judgment — will give AI model users a low-cost, high-efficiency strategy.

The importance of routing grows as AI agents become more widespread. Unlike a simple question-and-answer exchange, an AI agent handling a single task makes multiple calls to AI models — planning, searching and re-evaluating along the way. Even if the cost per individual model call falls, the total cost becomes harder to manage as the number of calls required for one task increases.

Gartner projects that the inference cost for an AI agent to complete a single task will increase more than fivefold by 2028. Simply subscribing to a basic rather than a premium AI model plan will not be enough to contain that cost growth. That is the backdrop against which routing — selecting the right model for each task based on its difficulty and purpose — is becoming central to cost management.

At least one routing-focused company has seen its valuation surge in a matter of months. OpenRouter, which provides a routing service connecting more than 400 AI models for developers and enterprises, was valued at around $1.3 billion in a Series B round in May. Payments company Stripe then acquired it in August for between $7 billion and $8 billion — a more than fivefold jump in under three months, according to foreign media reports.

As the center of gravity in AI competition shifts from model development toward real-world deployment, the strategic value of companies that connect multiple models or secure the touchpoints where models are actually used is rising. Acquiring a routing provider gives a company visibility into developer demand across AI models. Combining that with an existing business creates a path to uncovering new demand and mapping out directions for expansion.

Nvidia's recently reported pursuit of Hugging Face reflects the same dynamic. Hugging Face is a platform where AI developers publish open-weight models that other developers and enterprises can download and use. Nvidia agreed to acquire Hugging Face for $12.9 billion, according to foreign media reports, though the possibility of the deal falling through has since been raised. The episode highlights the strategic value of the deployment layer: Nvidia's interest was understood to center on gaining insight into which AI models are actually being used and on what hardware.

The fact that even a chipmaker that does not develop AI models itself is moving to secure a foothold in the deployment layer shows that the center of gravity in AI competition is broadening from model development to model use. Routing — connecting and selecting among multiple models — is seen by the market as a major pillar of that deployment layer.

Kim Il-hyuk, a researcher at KB Securities, said that as services automatically selecting the right model for each user request become widespread, "users will be able to get results without worrying about whether they are using GPT or Claude." He added that routing providers "not only earn revenue from connecting models but also gain insight into which models are seeing growing developer demand — so it is worth paying close attention to the shift in leadership from the model layer to routing."

This article is part of HeralDeep, a members-only content service of The Herald Business. Subscribers can access the full list of articles in the series through their membership.

kacew@heraldcorp.com