Ari Jacoby's Concentrate AI enters the AI routing fight as token bills bite

Why it matters

AI routing is becoming the cost-control layer for companies that adopted multiple models before they built model governance. The category's value depends on whether developers treat routing as independent infrastructure or let cloud platforms absorb it.

Ari Jacoby's Concentrate AI emerged from stealth Wednesday with more than $5 million in funding, Business Insider reported, entering a market that OpenRouter has already turned into one of the clearest infrastructure bets of the multi-model AI cycle.

Jacoby's pitch is not that companies need another model. It is that they need a control plane for the models they are already using. "The model landscape is so fragmented, it's so hard to track," Jacoby told Business Insider. "And with us, it's all under one roof."

That is the bet behind AI routing: sit between developers and model providers, then send each request to a model based on price, latency, availability, capability and data policy. The category exists because AI adoption has moved faster than AI cost discipline. OpenAI and Anthropic still command attention at the frontier, but the bill arrives in tokens, and a growing number of engineering teams are learning that model choice is now an operating expense decision.

The new gateway tax

OpenRouter has become the reference point for the category. The company says its platform offers access to 400+ models from 60+ providers through an OpenAI-compatible API, with automatic failover, edge routing and custom data policies. Its homepage also claims 100T monthly tokens and 8M+ global users, figures that are company-reported and not audited in the materials reviewed.

The investor interest followed the usage curve. In late May, OpenRouter raised $113 million at a $1.3 billion valuation, according to Business Insider. RuntimeWire reported in May that the financing was a Series B led by CapitalG, with NVentures, ServiceNow Ventures, MongoDB Ventures, Snowflake Ventures and Databricks Ventures among the backers. Business Insider also quotes OpenRouter strategy head Adam Swick saying developer demand has "exploded" over the last six months.

Concentrate AI is arriving after that capital signal, not before it. Its funding disclosure leaves out the cap table: named investors, round type and valuation are not part of the public debut. That matters because routing is a scale business. The company that wins trust with developers can sit in the payment path, observe model demand in real time and become the switching layer when prices or performance change.

DeepSeek made the pitch easier

The cost argument became concrete this spring. Business Insider reported that DeepSeek (@deepseek) released V4 models in late April and quickly gained share on routing platforms. By mid-May, more tokens were passing through OpenRouter for DeepSeek models than for Claude models, according to Business Insider.

The price spread explains why. Business Insider cited Claude Haiku at $1 per million input tokens and $5 per million output tokens. On OpenRouter, the priciest DeepSeek V4 version cited in the report ran at $0.43 per million input tokens and $0.87 per million output tokens. That does not make DeepSeek the right choice for every workload, but it makes the routing layer economically legible: if a cheaper model is good enough for a task, the gateway can keep the premium model out of the hot path.

Zach Moskow, identified by Business Insider as a founding team member at Concentrate AI, framed one of the tradeoffs: some customers worry about the security of Chinese models, but those models are often hosted on AWS in the U.S. That is exactly the sort of nuance routing companies want to own. The decision is not simply DeepSeek versus Claude. It is which model, hosted where, under what policy, for which prompt, at what cost.

The clouds are already there

The routing market is not empty white space. Amazon Web Services, Microsoft and Google Cloud each offer tools for routing AI tasks to appropriate models, according to Business Insider. Those companies have procurement gravity and infrastructure depth that startups cannot match.

The counterposition from OpenRouter and Concentrate AI is choice. OpenRouter's catalog spans hundreds of models, including providers and model families such as OpenAI, Anthropic, Google, Meta, Microsoft, Amazon Bedrock, Mistral, Cohere, DeepSeek and Qwen, according to the company's site. That breadth is harder for a single cloud vendor to replicate without weakening its own platform incentives.

Vercel has also moved into the market. Business Insider reported that Vercel, valued at $9.3 billion, built its routing product after using it internally. Harpreet Arora, Vercel's head of AI infrastructure and the product's lead, described it as a "centralized hub" for managing cost swings, outages and the expanding model menu.

Lanai is attacking the adjacent spend-management problem. Business Insider reported that the AI observability company launched Token Tuner, a product meant to diagnose the effectiveness of AI spending. Lanai chief product officer Mohit Mehta told Business Insider: "You're going to manage your AI spend to value, just like you manage your workforce to value."

That line captures why Jacoby's timing is rational. The first wave of AI infrastructure rewarded access to the best model. The next wave is rewarding judgment about when not to use it. Concentrate AI's challenge is to prove that judgment can be productized before OpenRouter, Vercel or the cloud platforms make routing feel like a default feature rather than a standalone company.