$50M Series B Values Voice Infrastructure Provider Vapi at $500M

San Francisco-based voice infrastructure startup Vapi has closed a $50 million Series B round at a roughly $500 million post-money valuation. The funding, reported on May 12, follows Amazon’s decision to route 100% of its Ring inbound customer support calls through the platform. Peak XV Partners led the round, with participation from M12, Kleiner Perkins, and Bessemer Venture Partners, bringing Vapi’s total capital to $72 million.

The Amazon Ring Deployment

Ring evaluated over 40 AI voice vendors during the 2025 holiday surge before selecting Vapi. The transition from pilot to full production took two weeks. Rather than routing a fraction of calls to AI triage, Ring now pushes its entire inbound volume through Vapi’s infrastructure. Customer satisfaction scores improved following the rollout, proving that automated telephony can match or exceed human baselines when latency is tightly controlled.

Operating at Enterprise Scale

Vapi transitioned from a developer tool to a core telephony provider over the past year. The platform processes between one and five million calls daily and recently crossed one billion total calls. More than one million developers use the system, deploying 2.7 million unique agents. Enterprise annual recurring revenue grew 10x since early 2025, driven by clients like Intuit, New York Life, and ServiceTitan. If you need to evaluate and test AI agents at high volume, this scale demonstrates the viability of API-driven voice infrastructure.

Modular Architecture and Latency

Vapi operates as an API-native infrastructure layer rather than a bundled software product. Founders Jordan Dearsley and Nikhil Gupta originally built the underlying latency optimizations for a personal voice-based walking companion before pivoting to enterprise call centers.

Engineering teams can swap underlying large language models and voice engines without rebuilding their telephony stack. The platform optimizes for sub-500ms latency to maintain natural conversational pacing. This modular approach lets teams upgrade models independently while Vapi handles network transit, speech-to-text, and text-to-speech synchronization. If you build real-time voice agents, decoupling the AI model from the telephony transport layer prevents vendor lock-in.

Pay-As-You-Go Economics

Traditional contact center software relies on per-seat licenses. Vapi charges purely on consumption. The platform enforces a base fee of $0.05 per minute and passes third-party LLM and telephony costs directly to the customer.

Component	Typical Cost Per Minute
Vapi Platform Fee	$0.05
LLM & Telephony Pass-Through	$0.25 to $0.28
Total Estimated Cost	$0.30 to $0.33

This transparent structure makes it easier to reduce LLM API costs in production by swapping in smaller, cheaper models for simple routing tasks while reserving frontier models for complex support tickets.

Production Reliability Focus

Vapi plans to use the Series B capital to tighten uptime guarantees and enforce predictable latency under heavy load. The roadmap includes advanced operational guardrails to keep agents strictly within defined business logic. When designing voice support systems, your architecture must separate conversational intelligence from telephony transport to ensure rapid model upgrades without disrupting the customer experience.