'AI routing' startups are betting they can help companies grappling with token costs

Concentrate AI's team. Courtesy of Concentrate AI Venture funds are backing startups that manage the landscape of AI models. OpenRouter raised $113M million; Concentrate AI emerged from stealth with $5 million to compete. Cheaper AI models like DeepSeek's V4 have gained traction due to cost efficiency. The backlash against rising AI costs is growing, and now investors are pouring cash into startups that help save companies from that sticker shock. These startups, known as AI-routing companies, help developers direct their tasks to different AI models. They also monitor for overspend and quickly resolve outages — and they're seeing a gigantic surge in demand. In late May, OpenRouter announced it raised $113 million, valuing it at $1.3 billion. And on Wednesday, the new competitor Concentrate AI emerged from stealth with more than $5 million in funding, Business Insider is the first to report. "The model landscape is so fragmented, it's so hard to track," Concentrate AI's CEO Ari Jacoby told Business Insider. "And with us, it's all under one roof." Both companies are growing at the right time. AI coding tools have kicked off a rabid demand for tokens, the units of AI's input and output. Anthropic and OpenAI charge many customers based on how many tokens they use, so their powerful models are giving some companies sticker shock. To help customers figure out the most cost-effective AI tool to get their tasks done, routing startups provide access to frontier models from those labs, as well as a slew of cheaper models from providers such as Google, DeepSeek, MiniMax, and Xiaomi. Big Tech companies also offer AI routing tools OpenRouter and Concentrate AI have competitors with a lot more firepower. Amazon Web Services , Microsoft, and Google Cloud each have tools that route tasks to the appropriate types of AI models. The startups say they focus on developers and smaller teams, and they give them the option to choose from more model types than cloud providers do. OpenRouter offers more than 400 models, strategy head Adam Swick told Business Insider, adding that developers' "demand has exploded" over the last six months. Vercel, a cloud application startup valued at $9.3 billion, also built its own AI routing product after finding it useful for internal use. Harpreet Arora, Vercel's head of AI infrastructure and the tool's lead, told Business Insider that his team immediately realized the tool could become popular with customers when people started highlighting its benefits. He sees it as a "centralized hub" — a place to pick and choose when costs fluctuate, models go offline, and more models enter the market. Developers turn to cheaper models like DeepSeek AI routing startups are seeing shifting demand for AI models as cheaper options emerge. Both OpenRouter and Vercel saw the Chinese lab DeepSeek outstrip its competitors this spring. In late April, the lab released its new V4 models. They impressed developers on capability benchmarks and cost. Claude's cheapest model, Haiku, costs $1 per million input tokens and $5 per million output tokens. On OpenRouter, the priciest DeepSeek V4 version ran at a fraction of that cost, at 43 cents per million input tokens and 87 cents per million output tokens. DeepSeek's share of overall token usage, both on Vercel and OpenRouter, quickly increased, indicating that more developers were turning to the lab's models. By mid-May, more tokens were passing through OpenRouter for DeepSeek models than Claude ones. The same trend appeared with Vercel. Zach Moskow, a founding team member at Concentrate AI, said that some customers are worried about the security of Chinese models, but they're often hosted on AWS, right in the US. Jacoby, Concentrate's CEO, said that when companies realize many high-quality models are available at low prices, budgeting becomes simpler. Other startups are also pouncing on the panic over AI token costs. Lanai, an AI observability startup, recently launched a tool called Token Tuner that diagnoses the effectiveness of customers' AI spending. Chief product officer Mohit Mehta told Business Insider that he's hearing complaints about the unpredictable costs of top-end AI and expects the trend toward cheaper, more basic models to grow. "You're going to manage your AI spend to value, just like you manage your workforce to value," Mehta said. Have a tip? Contact this reporter via email at scouncil@businessinsider.com , or over text, Signal, Telegram, or WhatsApp at 415-757-8198. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely . Read the original article on Business Insider