Temporal talks $12B six months after $5B: durable execution is now agent infrastructure

Temporal Technologies is in talks to raise around $500 million at a pre-money valuation of at least $12 billion, Bloomberg reported on August 18th ^[1]. The company announced a $300 million Series D at $5 billion on February 17th ^[4], which makes this one of the fastest repricings an infrastructure buyer is likely to see in a vendor already sitting in production.

The round has not closed, and the amount and terms could change ^[2]. At the reported floor, $500 million would buy roughly 4% of the company after the financing ^[3], implying a post-money of at least $12.5 billion ^[1]. That is about seven times the $1.72 billion post-money mark set by the $146 million Series C in March 2025 ^[14][2], and the proposed round is roughly 1.7 times the size of a Series D announced about six months earlier ^[3][4].

What changed is not the software. Samar Abbas and Maxim Fateev worked together on Amazon Simple Workflow Service; Abbas went on to lead Microsoft's Durable Task Framework, and Fateev had led SWF's architecture and the storage backend for Amazon Simple Queue Service ^[6]. They reunited at Uber, co-created the open-source workflow engine Cadence, and founded Temporal in 2019 to build a commercial platform on the same durable-execution model ^[7]. The product records execution history and application state so a multi-step workflow can retry or resume from its last recorded point instead of starting over ^[8].

The reclassification is on the demand side. Agents make model calls, use tools and wait on external systems or human approval, running for minutes, hours or days, and each added step is another place where state can disappear, an API can fail, or an unsafe retry can repeat an action ^[10]. Abbas's framing in the February announcement was that agents fail "because the systems around them can't handle real-world execution" ^[11]. Temporal says customers use the platform to maintain agent state, trace failures and keep long-running tasks moving through interruptions ^[12]. It does not train models; it is betting companies will pay for a common execution layer beneath them, as they pay for databases and observability ^[13].

For anyone with Temporal in the critical path, the numbers to negotiate against are the vendor's own. In March 2025 it reported more than 2,500 Cloud customers, 183,000 weekly active open-source developers, 7 million unique cluster deployments, 4.4 times revenue growth over 18 months and net dollar retention of 184% ^[16]. By February 2026 it reported revenue growth above 380% year over year, monthly installations past 20 million and 9.1 trillion lifetime Cloud actions ^[17]. Those figures are Temporal's, and they mix commercial and usage measures: percentage growth hides the starting base, and installs and action counts do not convert into recurring revenue ^[18]. The 184% net dollar retention is the line that reads differently at renewal than in a press release ^[16], because expansion inside existing accounts is what a $12 billion mark has to be underwritten by.

The hedge is in the licence. The core platform ships under MIT and Temporal sells Temporal Cloud as the managed service ^[9], so an exit from Cloud pricing is an operations project rather than a contract fight - which is a different negotiating position than a closed-source dependency, and worth costing out before the next term sheet resets list prices.

Watch whether the round closes at the reported floor or above it ^[1][2], and whether disclosed primary funding of roughly $1.15 billion ^[15] ever comes with absolute revenue rather than multiples ^[18]. If your renewal lands after the close, ask for term length and action-volume commitments now.