Vapi 完成 5000 万美元 B 轮融资，语音 AI 智能体已处理超 10 亿通电话

Vapicon 2026 is coming

The platform that makes every business feel human.

Watch the Series B announcement video

By the time you die, you will have spent 43 days of your life on hold.

Forty-three days of “your call is important to us.” Forty-three days of menu trees and transfer hand-offs and re-entering your account number into the keypad and re-explaining yourself to a fourth person. Multiply that across the country, and Americans spend 900 million hours a year on hold. Multiply it across the world, and you get to three trillion minutes a year of business-to-customer phone calls. This is the channel that companies continue to pay billions of dollars to build, staff, maintain, and apologize for.

We started Vapi to end it.

Today we're announcing our $50M Series B, led by Peak XV, with participation from YC, M12 (Microsoft's Venture Fund), Kleiner Perkins, and our earlier investors. It brings our total funding to $72M. The round closed on the back of a year in which our enterprise revenue grew 10x, and our agents crossed 1,000,000,000 calls handled. But this post isn't about the round. It's about the bet behind it.

The spark

In the summer of 2023, I was depressed and obsessed with a problem.

I'd been working with speech models for years and could see three curves bending at the same time: latency dropping, models getting smarter, cost falling, all at once, across the entire voice stack. Stitch them together carefully, and you could build something that felt like a person on the other end of the line.

So for my daily walks, I built an AI therapist named Harmon. I printed business cards. I rode the bus around SF and handed them out to strangers. Call this number. Get free therapy.

Nobody wanted AI therapy. But the first time I had a real, fluid, back-and-forth conversation with a model on the phone, I knew. Not in the abstract, in my gut. This is how everything is going to work. Voice is the default way humans have always connected. We were about to start talking to technology, and our technology was going to talk back.

The gap

It’s two years later, and that moment is about to land in everyone's pocket. Think of where Siri, ChatGPT voice, and Gemini Live are heading: an assistant that listens, remembers, and talks back like a friend. That's the device every consumer will have within eighteen months - voice-native, context-aware, with you all day. People will talk to their software more than they tap on it. The bar for what a conversation can feel like is about to reset, and it will reset fast.

With all this technology readily available, why should those same people call their insurance company and get an IVR from 1998?

That gap - between the conversation a consumer has at home and the one an enterprise serves them on a customer service line - is about to become the most embarrassing thing about every Fortune 500 company in the world. Customers will start punishing it. Brands will start losing because of it.

We are building the platform that closes the gap. One that makes every business feel human.

The bet

Most of the voice AI market right now is selling cost reduction. Replace the human, save the money, and get more profitable. That's a fine starting wedge, but it’s not the end game.

The end game is taking the call center - the largest cost center in most enterprises - and turning it into a revenue center. A place customers actually like to call into.

Think about that for a second. The contact center at AT&T, JP Morgan, State Farm - historically a budget line measured in billions, optimized to be as cheap as possible to make the angry customer go away. Now, picture an agent that doesn't just resolve the call but resolves it better than any human ever has. Closes more. Retain more. Sells more. Listens better. Knows more about your customer. Every attribute of your best human agents is scaled to your entire customer base

That requires voice agents that get measurably better with every call. A platform where every conversation is a training signal, every resolution is a labeled example, and every miss becomes the next iteration's learning. The agent your customers talk to next month will be obviously better than the one they talked to today.

Self-improving voice agents, at production scale, with a feedback loop tight enough that the line between AI and the best person you have ever talked to stops being just an idea.

That's what we're building.

The unglamorous half

You can't put agents like these in front of millions of customers without the boring half of the work: compliance, observability, hard guardrails, predictable latency under load, and clean escalation to humans when the situation calls for one. Every call is a production workload. Every call has the potential to embarrass the company on the other end of it, if things do go wrong.

We've spent years on this part. It's why Amazon Ring went from evaluation to running 100% of inbound on Vapi in two weeks. It's why we win in financial services, healthcare, and insurance - categories where one bad call can cost more than a year of savings and reputational damage.

Governance and outperformance are not separable. Without governance, no enterprise should put an agent live. Without agents that outperform humans, governance is just paperwork. We have built a foundation and are continuing to invest in both at the same time. That's what a platform is.

What it looks like when it works

Amazon Ring evaluated dozens of vendors but decided on Vapi and went to production in two weeks. Today, every inbound call to one of the most demanding consumer support operations in the world runs through Vapi. Even more impressive, CSAT went up. Jason Mitura, VP of Software Development, said it simply: “A lot of AI tools promise great outcomes. Vapi has delivered on them.”

Kavak, the largest used-car marketplace in Latin America, made a company-wide bet on agents. Their CPO, Alejandro, went to the CEO and pitched a rebuild of every workflow - sales, operations, support - around voice AI, with the entire company either building or tuning agents. They started with one use case. They now run dozens. They built memory and personalization on top of our platform themselves, so a returning customer picks up exactly where they left off. We didn't ship that. They did; that is the power of our platform. Their call center became a revenue center. They doubled revenue in their Mexico market.

Instawork moved candidate interviews from humans to agents. Supply doubled. The agent never has a bad day, so every candidate gets the same clean interview, whether at 2 pm or 2 am. Some candidates stay on for thirty minutes just to talk about their lives.

A million developers have built on Vapi. Millions of agents have been created. We've handled more than a billion calls. We did almost no marketing to get there. The product is just that good, and the word spread.

In every one of these cases, the same thing happens: customers stop noticing. The business, on the other end of the line, starts to feel like a human.

What this round funds

With this round, we are focusing our investment in three places. Distribution, so the enterprises that should be evaluating us are evaluating us. Delivery, so customers go from "we want to do this" to "we are doing this" in weeks instead of quarters. And the engineering investment that matters most - core infrastructure, governance, and the self-improvement loop that turns every call into the agent's next training signal.

We are continuing to focus on building the parts that actually let enterprises ship with confidence. With tools like Composer, Monitoring, and Simulations - launched over the past few months - customers can close the loop between building an agent, testing it against reality, and adapting it when reality pushes back. You can watch every call. You can catch the failure before your customer does.

The end state is a platform where the agent that picks up the phone tomorrow is reliably better than the one that picked up the phone today. We are building toward that, every day, at every layer.

The big claim

There is a generational company to be built in voice.

Three trillion minutes a year of business-to-customer voice. A category Twilio served for fifteen years and barely scratched. The biggest cost center in the Fortune 500, sitting there, waiting to become a revenue driver, and almost nobody is in the room.

But we are. We are building the platform that makes every business human.

If you want to spend the next few years on the most consequential thing you can do in AI right now, come build it with us.

A billion calls in, and we’re just getting started.

How an in-house review bot cut our PR merge time in half

Fast decisions alongside voice AI: early experiments with Jev

Build vs. buy: what you're really deciding with voice agents

What makes a great conversational AI experience?

Why model flexibility matters in voice

Questions from the Anthropic and Vapi Webinar, answered.

Built for the Ear: Designing Conversations for Voice

How we Bootstrapped the Voice Agents on the Vapi Homepage

Introducing Vapi Monitoring

Vibe code voice agents

Your Voice Agents Need Tests. Now They Have Them.

GPT-5.1 Just Fixed the Thing That's Been Bugging Me for Years

Introducing Squads: Teams of Assistants

GPT-5 Now Live in Vapi

How we solved latency at Vapi

What Is Signal Processing? Voice AI Definition Guide

How to Use Grok 3 in a Voice Agent

Unpacking LLM Temperature

How to Build a GPT-4.1 Voice Agent

What Is Gemma 3? Google's Open-Weight AI Model

Introducing Vapi Workflows

What Are IoT Devices? A Developer's Guide to Connected Hardware

Building a GPT-4.1 Mini Phone Agent with Vapi

What Is GPT? Understanding A Core Technology for Voice AI

What is a Phoneme? An In-Depth Look for Technologists

Launching the Vapi for Creators Program

Introducing Vapi Voices

Vapi x Cartesia: Ultra-Realistic Voice AI with Sonic 2.0

Start building