Vapi Had 18 Incidents in 90 Days. They Just Raised $50M to Fix It for Amazon Ring.

On May 28, 2026, Vapi's status page logged another incident: "Increased Call Failure Rates," six user reports, active outage. This was incident number 18 in the last 90 days, including 5 major outages with a median duration of 101 hours and 40 minutes, according to IsDown's status monitoring data. That is the reliability record of the platform that just closed a $50M Series B at a $500M valuation.

This is not an attack on Vapi. Their infrastructure genuinely powers 1 billion calls and 1 million developers. The round was led by Peak XV, Microsoft M12, Kleiner Perkins, and Bessemer. Amazon Ring evaluated more than 40 AI voice vendors and chose them. That is a legitimate technical achievement.

The problem is not whether Vapi's infrastructure is good. The problem is whose SLA the $50M is going to improve, and whether your agency is in that conversation.

Why this matters for AI voice agencies

When your agent fleet goes down, your clients' phones stop getting answered. If you are running inbound reception for a dental practice, a law firm, or a home services company, a missed call at 9am costs that client a real appointment. Three missed calls in a month costs you the contract.

Eighteen incidents in 90 days is roughly one incident every five days. Five of those were major. The most recent was four days ago. During that same window, Vapi was closing a $50M round and announcing Amazon Ring as their flagship client.

Vapi's own post-raise statement explains what the capital is for: "governance, predictability, tighter uptime SLAs, call-level monitoring for enterprise operators." That language is accurate. It is also telling. Enterprise operators means Amazon Ring, New York Life, Intuit, and ServiceTitan. Those are the named clients in their press materials. The agency running 300 to 2,000 minutes a month is not a named client in that framing.

When two SLA tiers exist, the larger account gets prioritized in the engineering queue. That is not a policy failure. It is just how a company with a $500M valuation allocates its infrastructure investment. The question for every agency on Vapi's platform today is: which tier are you actually in?

The Retell parallel

Retell had its own database outage on May 25, four days before Vapi's May 28 incident. Call history, analytics, and the QA dashboard went completely offline, documented in Retell's own community forum. Both dominant infrastructure providers had significant outages in the same week.

This is the structural problem with building your agency on API infrastructure that was not designed specifically for agency operations. Vapi and Retell are excellent voice AI engines. They are not agency operating platforms. When they go down, your whole client fleet goes down with them, because the infrastructure is shared and the fallback layer, if it exists at all, is something you had to build yourself.

Retell put it well in their own comparison content: "A 20-minute outage by a voice provider becomes a 20-minute outage for the whole agent fleet on platforms without fallback." The irony of that statement appearing in Retell's own materials during their own outage week is not lost on anyone watching the space.

What we're doing at Hermes about it

Hermes controls the upstream voice infrastructure relationship directly. That means we are not reliant on passing Vapi's or Retell's outage timeline to our operators as-is. When a provider has an incident, we have architectural options that a wrapper or direct API customer does not.

More importantly, Hermes is not building for Amazon Ring. We are not raising a $50M round to win a Fortune 500 RFP. The platform is built specifically for the operator running 3 to 20 clients, charging $1,500 to $3,000 per client per month, who needs every call handled correctly because each one directly affects a client contract worth $1,500 to $3,000 per month.

Starter at $149 per month. Business at $399. Agency at $699. Overage at $0.24 per minute against a $0.18 landed cost. The margin is locked because we run the upstream relationship, which means we do not inherit pricing changes from a VC-backed infrastructure provider's investor call. We also do not inherit their SLA priorities.

By builders, for builders. That is not a tagline. It is a decision about whose problem we are solving.

Action steps for agencies on Vapi this week

1. Check your uptime exposure right now. Go to isdown.app/status/vapi and look at the incident history for the last 90 days. Count how many fell during your clients' business hours. If you have a client in a timezone where a major outage hit at 9am their time, that client felt it whether or not you did.
2. Calculate what one outage costs you in client trust. A client paying $1,500 per month is paying $50 per day. An inbound receptionist agent that went dark for four hours on a Tuesday morning costs that client real missed appointments. Add up what one 4-hour outage during business hours costs across your active client base. That is the actual risk number you are carrying.
3. Ask your current platform what their fallback is. Specifically: if the primary voice provider has an incident, what happens to calls in progress and queued calls? Is there automatic routing to a backup? Is there a status page you can show clients? Is there a documented SLA? If the answer is a shrug or a support ticket, you do not have a fallback. You have hope.
4. Separate the engine from the operating platform. Vapi is an engine. A great one. But an engine is not an agency operating platform. You still need the CRM, the white-label portal, the campaign builder, the billing surface, and the client-facing layer that keeps your brand in front of clients instead of Vapi's. If you are managing all of that separately, you are paying for and maintaining 5 to 7 tools, and every one of them can fail independently.
5. Tell your clients about reliability before an outage does it for you. If your clients know you run on Vapi, an incident at Vapi is now their incident too. The white-label layer exists so this conversation happens on your terms, not on Vapi's status page timeline. If you do not have a white-label layer between your clients and your infrastructure provider, that is the first thing to fix.

Frequently asked questions

The bottom line

Vapi raised $50M to solve a real problem: enterprise-grade reliability for Fortune 500 clients. Amazon Ring needs that. New York Life needs that. Those are real customers with real requirements that justify the investment.

The agency with 8 clients running 1,200 minutes a month needs something different: a platform that was designed around their P&L from the start, with infrastructure decisions made for their volume, their clients, and their margin requirements. Not a platform that will get around to solving their problems after it finishes solving Amazon Ring's.

One platform. Your brand. Your margins. From $149 per month. No VC round changing whose SLA we prioritize.

Sources: IsDown, Vapi status history (18 incidents, 90 days) · TechCrunch, Vapi $50M Series B at $500M valuation · GlobeNewswire, Vapi Series B announcement · Retell community, May 25 database outage · Retell, on fallback architecture and outage propagation