AI Agents Command Megadeals as Infrastructure Bets Grow Larger

Valuation Velocity Is Becoming Its Own Signal

Factory's $200 million round, reported by The Terminal, is striking less for its size than for its speed. The AI coding-agent company reached a $5 billion valuation in April at roughly $1.5 billion, meaning investors tripled their mark in five months ^[1]. The syndicate includes Blackstone and Salesforce chief executive Marc Benioff as an angel, and the company has now collected more than $400 million across three large checks in under a year.

What this pace communicates to other investors is not simply that AI is hot — it is that a specific subset of AI companies, those with agents performing discrete, measurable professional tasks, can compress the typical repricing cycle dramatically. Founders raising today will encounter LPs and partners who have internalized this as a new baseline. That cuts both ways: expectations for traction milestones are rising alongside headline valuations.

The Infrastructure Layer Is Absorbing Enormous Capital

Two infrastructure rounds announced this week collectively approach $4.1 billion, which puts agent-layer rounds like Factory's in context. Crusoe announced the initial close of an anticipated $3.9 billion Series F at a $30.9 billion valuation, with Nvidia among its backers, as Channel Insider reported ^[2]. Separately, Cornelis raised $205 million to launch what it calls an Active Compute Fabric — programmable networking designed for large-scale AI and high-performance computing, according to Pulse2 ^[3].

These rounds matter to early-stage founders even if compute infrastructure is not their category. When Nvidia writes a check into a cloud infrastructure company, it is signaling where it expects GPU demand to concentrate. Founders building applications on top of AI infrastructure should understand that the compute cost curve and availability assumptions they are modeling today may shift as this new capacity comes online over the next 12 to 24 months.

The Cornelis raise is the less obvious one to watch. Its open networking architecture adds computation directly into the network fabric rather than routing everything through centralized accelerators. If that approach gains traction at hyperscaler or mid-market AI labs, it changes the economics of scale-out training and inference — and therefore the unit economics that application-layer founders can project to investors.

B2B Agents Need a Workflow Address, Not Just a Demo

Magentic's $18 million Series A, led by Felicis with participation from Sequoia and The Westly Group, funds AI agents that handle the full procurement cycle for large manufacturers — supplier selection, contract negotiation, order placement, and invoice clearance, as The Conveyor reported ^[4]. The detail worth noting is where the agents actually live: inside Microsoft Teams, email, and the customer's own systems.

That integration decision is almost certainly a positioning choice, not just a technical one. Enterprise buyers are skeptical of yet another SaaS dashboard. Magentic's framing — agents embedded inside the tools procurement teams already use — removes one of the most common objections at the pilot stage. Founders pitching B2B agents this quarter should be able to answer precisely the same question: where does your agent live in the buyer's existing day, and what does it replace rather than add?

Healthcare AI Is Consolidating Around Patient Workflows

DexCare's acquisition of Mila Health, announced September 21, extends its platform from scheduling into broader care coordination ^[5]. Mila's agents contact patients by call, text, and chat to move them through multi-step care journeys. DexCare says the combined system runs on a data model already deployed across more than 100 million patients, though the financial terms of the acquisition were not disclosed.

For founders in health technology, the consolidation signal here is specific: acquirers are targeting companies whose agents handle patient-facing communication workflows, not back-office analytics. If you are building in healthcare AI, the acquirer landscape is beginning to take shape around access and coordination rather than diagnostics or billing alone. That has implications for how you frame strategic value in a pitch now, even if an exit is years away.