AIToday
Large Language ModelsAI Business & IndustrySemafor TechPublished: Aug 6, 2026, 04:01 JST3 min read

Sapiom raises $35M to slash AI token costs for startups

Sapiom raises $35M to slash AI token costs for startups

Key takeaway

  • Sapiom, a San Francisco startup, raised $35 million to help businesses reduce their spending on AI model tokens by automatically routing requests to the most cost-efficient model.

  • The company reduced one client's token bill tenfold, from $1.2 million to $100,000 per month, by operating its own server infrastructure and charging customers only for compute rather than adding a premium.

3 Key Points

  1. What happened

    Sapiom, a San Francisco startup, raised $35 million in a Series A led by Dragonfly to help businesses reduce their spending on AI model tokens. The company reduced one client's monthly token bill from $1.2 million to $100,000 by routing requests to the most cost-efficient AI model, and is launching a new API that automatically selects the right model for each task.

  2. Why it matters

    Startups using frontier AI models from providers like OpenAI and Anthropic face unsustainable token costs that can prevent them from deploying their products even when demand exists. Sapiom's routing service and own server infrastructure offer a way to cut those costs dramatically — potentially opening up AI product deployment for cost-constrained teams.

  3. What to watch

    Sapiom competes directly with OpenRouter, the largest existing routing firm. Sapiom's competitive edge is its own server infrastructure in San Jose and its willingness to charge only for compute rather than adding a markup, unlike most competitors who act as middlemen.

In Depth

Read the full story

Polsia, a San Francisco startup that runs swarms of AI agents to help operate other businesses, faced a crisis of scale this year. The company's projected annual revenue skyrocketed from $100,000 to $10 million, but its token spending on Anthropic's models kept pace, reaching $1.2 million per month — a cost structure that made deployment impossible despite strong demand. Founder of Polsia turned to Sapiom, a routing startup that optimizes AI costs, to evaluate and restructure the spending. After running tests, Sapiom reduced Polsia's monthly token bill tenfold, to approximately $100,000 per month. "It's just unsustainable," said Ilan Zerbib, Sapiom's founder and CEO, noting that startups cannot deploy their products profitably at the rates charged by frontier-model providers like OpenAI and Anthropic. Sapiom's solution centers on automatic routing: a single API key that intelligently directs each request to the most cost-efficient AI model for that specific task. The company is now launching this model router, putting it in direct competition with OpenRouter, the largest name in the routing space. Sapiom serves a few thousand customers and operates differently from most competitors: it hosts open weight models (AI models publicly released by creators) on its own server racks in a San Jose data center. Most routing firms do not own infrastructure and instead act as middlemen, inserting themselves between customers and other AI providers. Critically, Sapiom does not charge a markup on the models it provides; instead, it charges customers directly for the compute resources it consumes. This cost pass-through model gives customers savings that competitors cannot match. To fuel expansion, Sapiom raised $35 million in a Series A led venture fund Dragonfly, with Dragonfly partner Haseeb Qureshi leading the round. The company had previously raised $15 million in a seed round led by Accel earlier this year.

Context & Analysis

Sapiom's funding and product launch reflect a growing pain point in the AI economy: frontier model providers charge per token in a way that can quickly become prohibitively expensive for startups trying to scale. Polsia's trajectory — projecting revenue from $100,000 to $10 million annually — illustrates the mismatch: explosive business growth paired with equally explosive token costs that made the product economically unviable until Sapiom intervened. The startup's own infrastructure in San Jose differentiates it from established routers; by hosting open weight models directly rather than acting as a middleman, Sapiom can pass through lower costs to customers and avoid the margin squeeze that affects competitors who resell capacity from other providers. This model suggests a structural shift in how AI software businesses might operate: rather than accepting the token prices set by OpenAI and Anthropic, startups now have an option to route workloads to cheaper alternatives without sacrificing performance for every task.

FAQ

How much did Sapiom reduce one client's token costs?
Sapiom reduced the monthly token bill from $1.2 million to about $100,000 per month — a tenfold reduction — for Polsia, a San Francisco startup that deploys AI agents for other businesses.
How is Sapiom different from competitors like OpenRouter?
Sapiom operates its own server racks in a San Jose data center and serves open weight models from that infrastructure, whereas most competitors act as middlemen between customers and other providers. Sapiom also charges customers only for compute rather than adding a premium.
What funding did Sapiom secure?
Sapiom raised $35 million in a Series A led by venture fund Dragonfly, and previously raised $15 million in a seed round led by Accel earlier this year.

Get the latest Large Language Models news every morning

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · takes 30 seconds · unsubscribe anytime

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleMoove raises $250M for autonomous vehicle infrastructure

The AI news that matters, in one minute each morning.

Sign up free