Factory, a startup that sells autonomous “Droid” agents for enterprise software engineering, just tripled its valuation to $5 billion in five months. That alone is a notable data point in a crowded market. What makes it worth a second look is how it happened: not through a bigger feature set or a flashier demo, but through the same investor logic normally reserved for cloud infrastructure: utilization, lock-in, expansion revenue. Coding agents are no longer being sold, or priced, like developer tools. They are being priced like infrastructure.
The fastest re-rating in enterprise software
Factory raised $200 million at a $5 billion valuation on September 16, 2026, up from the $1.5 billion valuation it held just five months earlier, in April. Total funding now tops $400 million, backed by Blackstone, Sequoia Capital, Khosla Ventures and Insight Partners, among others. Its customer list (Nvidia, Adobe, Morgan Stanley and Ernst & Young) reads like an enterprise infrastructure sales deck, not a dev-tool one, according to TechCrunch’s coverage of the company.
Factory is not the outlier. Rival Cognition, maker of the Devin agent, raised $2 billion at a $48 billion valuation on September 8, 2026, up from $26 billion just over three months prior. In that window, Cognition’s run-rate revenue grew from $492 million to roughly $900 million, per TechCrunch’s reporting on the round. Two competitors, two re-ratings within weeks of each other: investors appear to be underwriting this category on infrastructure-scale growth assumptions, not standard SaaS multiples.
From seat licenses to metered compute
The valuations track a pricing shift that is already visible in procurement. Vendors across the coding-agent stack are dropping flat seat licenses in favor of usage-based billing modeled directly on cloud infrastructure: metered by compute, tokens or agent-runtime minutes rather than by user count. The effect on budgets is dramatic: per-engineer AI coding bills have jumped from the $20–$100 a month typical of a copilot subscription to $2,000–$5,000 a month in some agentic deployments, with outlier cases reaching $20,000 in token charges, according to The Register.
That is precisely the cost profile of infrastructure, not software: unpredictable, usage-linked, and scaling with adoption rather than headcount. CIO.com reports that at some enterprises, AI coding token costs are now on track to rival human payroll. A line item that used to sit under “software licenses” is migrating toward the same budget category as cloud compute.
What enterprises are actually buying
The buyer is changing along with the pricing model. Factory’s and Cognition’s enterprise deployments span incident response, code refactoring, security review and automated triage, organization-wide platform contracts rather than individual developer subscriptions. Devin’s feature set now includes Auto-Triage for incident investigation and a “Security Swarm” for vulnerability scanning, deployed at companies including Nvidia, GE Aerospace and Citi, per TechCrunch’s profile of the round.
That procurement pattern matters more than any single valuation number. When enterprises evaluate infrastructure, price per unit is rarely the deciding factor: integration ease (41%) and total cost of ownership (35%) drive the purchase decision, while cost-per-token decides only 8% of deals, according to a VentureBeat analysis of enterprise AI buying behavior. Coding-agent vendors that can plug into the existing SDLC (GitHub, Slack, Linear, CI pipelines) are being valued the way infrastructure incumbents are: on how hard they are to rip out once installed.
The gap between valuation and verifiable ROI
The infrastructure framing cuts both ways. Consumption pricing is great for vendor revenue predictability, but it pushes forecasting risk onto the buyer, and enterprises are struggling with exactly that shift. VentureBeat’s compute-gap research found that 78% of IT leaders have been hit with unexpected charges from consumption-based AI pricing, and 90% of CIOs now name cost forecasting as their single biggest AI deployment challenge.
Finance teams are responding accordingly. Even as agentic workflow tooling remains the fastest-growing line in enterprise AI budgets, CFOs are tightening controls and demanding measurable returns rather than open-ended experimentation. For engineering leaders, the practical implication is straightforward:
- Treat coding-agent spend as a metered infrastructure cost with its own forecasting and alerting, not as a software subscription line item.
- Evaluate vendors on integration depth and total cost of ownership, not headline per-seat price. That is already how the market is pricing them.
- Instrument usage before scaling rollout; the enterprises getting burned by “surprise” bills are largely the ones that skipped this step.
The takeaway
Factory’s $5 billion valuation is not really a story about one company’s product. It is confirmation that the market has stopped treating autonomous coding agents as developer tools and started treating them as infrastructure: priced on consumption, sold on integration depth, valued on how deeply embedded they become in the software delivery pipeline. For engineering and finance leaders, the question worth asking this quarter is not “which coding agent has the best demo,” but “do we have the same cost governance over our agent spend that we already have over our cloud bill?” The vendors have already answered that question with their pricing models. Most buyers have not.