Temporal Raised $550M Because Agents Keep Crashing

On September 14, Temporal announced a $550 million Series E at a $12.55 billion valuation, co-led by Lightspeed with Wellington Management, Growth Equity at Goldman Sachs Alternatives, and Tiger Global. T. Rowe Price and SV Angel joined; a16z, Sequoia, Index, GIC, Sapphire and Amplify returned.

Temporal does not make a model, an agent, or anything a consumer will ever see. It makes workflows that survive the machine they are running on dying. That is the entire product, and it is currently one of the fastest-compounding businesses in AI infrastructure.

The numbers underneath the round

The growth figures are unusually specific for a private company, and they are the argument for the valuation.

Annualized revenue run rate is up more than 200% year over year. Net dollar retention has stayed above 200% since February 2026 — meaning existing customers alone more than doubled their spend, before any new logo. The platform processed 1.9 trillion billable actions in August, up more than 350% year over year. Open-source installs passed 43 million, up 134% since December 2025. Paying customers number more than 4,300, up 139% year over year. Headcount doubled to 570 in a year.

Note the shape of those numbers. Actions grew 350% while revenue grew 200%, which means per-customer consumption is climbing faster than the price. Net dollar retention above 200% sustained for seven months is not a land-and-expand motion — it is a workload that grows on its own once installed. Something is generating dramatically more durable executions inside existing accounts than it was a year ago.

That something is agents.

Why agents break the old assumptions

A traditional backend request is short. It runs for milliseconds, it either succeeds or it fails, and if it fails you retry it. The whole architecture of web services is built around that assumption: statelessness is cheap because state is brief.

An agent run breaks every part of that. It runs for minutes, hours or days. It calls external tools that rate-limit, time out, and return garbage. It accumulates context that is expensive to rebuild. It spawns subagents. And critically, its steps are usually not idempotent — the agent already sent the email, already charged the card, already filed the ticket. You cannot fix a crash by running the whole thing again.

So the failure mode is not "request failed, retry." It is "the process died 40 minutes in, after three irreversible side effects, holding context that cost real money to build." Every team that ships an agent to production discovers this in roughly the same week.

Durable execution is the answer that already existed. Temporal persists the execution state of a workflow, so that when a worker crashes, the workflow resumes at the exact step it reached, with its history intact, without re-running completed side effects. It was built for payments, order fulfillment and infrastructure provisioning — domains where "did the transfer happen or not" has to have one answer.

Agents turned out to be the same problem with worse odds.

The layer that got repriced

This is the second time in a month that the market has put a price on agent plumbing. OpenAI shipped a managed Agents API in public beta on September 10, handling session orchestration, context compaction, sub-agent coordination, lazy tool loading and crash recovery — with no fee beyond tokens, tools and sandbox compute, and nine sandbox partners including Cloudflare, Modal, E2B and Vercel.

Read those two events together. OpenAI is giving orchestration away to sell tokens. Temporal just raised half a billion dollars selling orchestration as a product. Both are correct, and the difference is scope.

The managed API is the right default for an agent that lives entirely inside one vendor's models and tools. Temporal is what you reach for when the workflow spans your own systems — when step three calls a mainframe, step five writes to a ledger you are audited on, and step seven has to be provably not-double-executed. That is why net dollar retention is where it is. Once a durable workflow is the backbone of a regulated process, it does not churn; it grows.

The company says the money goes toward global operations, deeper investment in the core primitives beneath the platform, and the reliability and security work enterprise customers asked for. That is a boring roadmap, and it is the correct one for a business whose entire value proposition is that it does not lose your state.

What to take from it

The agent narrative has been dominated by capability: which model reasons better, which harness plans further, which benchmark moved. Temporal's round is a reminder that the binding constraint on agents in production has not been intelligence for a while. It has been execution guarantees — whether the thing that ran for six hours can be trusted to have run exactly once.

1.9 trillion billable actions in a single month, growing 350% a year, is what that constraint looks like when it is priced. The models get the headlines. The retry logic got the $12.55 billion valuation.