Arda Raises $232M to Make Factories Reconfigure at Software Speed

Bob McGrew's last job was chief research officer at OpenAI. Before that he spent a decade building products at Palantir. His new company, Arda, starts from a less glamorous observation: a factory can spend three to six months and close to $1 million automating a single assembly station, then watch the product change before the line is ready.

A human worker absorbs a revised instruction in a day. The machine around that worker takes months.

The bottleneck sits with the manufacturing engineer

Arda's agent reads CAD files, PDFs, PLM data and product requirements. It builds one model of the product and its parts, then maps that model into a manufacturing plan: operations, tools, quality checks, fixtures, stations and visual work instructions.

Engineers still review and refine each step. The software removes the blank-page problem and the manual translation between design and factory floor.

At Anduril, the results arrived before Arda was incorporated. The two companies began working together during the startup's stealth period. Anduril now runs Arda inside ArsenalOS, the stack it uses to plan production.

Basic work instructions that once took 10 hours now take 15 minutes, according to Anduril. More complex engineering workflows that ran up to 30 hours now take about 2.5 hours.

Our manufacturing engineers are doing two hard things at once. They're keeping up with design teams that iterate constantly, and they're scaling production to meet demand. Arda helps them get design changes onto the floor faster and catch potential issues sooner. We would be behind on our production goals without it.— Matt Grimm, co-founder and COO, Anduril Industries

The comparison that matters is not a chatbot versus a spreadsheet. It is a design revision that reaches the floor in an afternoon versus one that waits for a manufacturing engineer to rebuild a process from scratch.

Sources: Anduril and Arda statements, October 2026

What $232 million buys

The company emerged from stealth on October 20 with $232 million in total funding. The most recent slice is a $160 million Series A at a $1.5 billion valuation, co-led by Altimeter, Accel and Khosla Ventures. XYZ Venture Capital, 8VC, SVAngel and Felicis participated.

The earlier round was a $72 million seed led by Accel and Founders Fund, with Khosla Ventures and XYZ Venture Capital also taking part. A seed round of that size is unusual outside frontier AI, and it sets the expectation for what comes next.

Arda says the money goes to engineering and research hiring, deployments with manufacturers in electronics and the data-center supply chain, and the robotics models needed for automated assembly.

McGrew's co-founders come from Palantir and Adept. The team includes manufacturing engineers from Tesla and AI researchers from MIT, according to the company. That mix is the pitch: people who have shipped software at scale, paired with people who have stood on factory floors.

Apoorv Agrawal of Altimeter, who led the round, told CNBC that manufacturing is a $3 trillion industry in the United States alone. "Yes, the price is steep today, but the long-term opportunity is very high," he said.

The harder problem is reprogramming the robot

Manual assembly lines are the first market. The second is automation itself.

Conventional automation requires custom equipment, lengthy engineering work and careful calibration for each assembly step. Arda is prototyping AI agents that program robots to carry out its assembly plans, station by station.

The goal is an automated line that can be updated in days after a design change. That would put robots close to the flexibility of a human line, where a supervisor explains the new step and work resumes.

The prize is not cheaper robots. It is a factory that treats a design change as a software update.

McGrew frames the end state in geographic terms. "Imagine a fully automated version of Shenzhen in West Texas — that's what we want to make possible," he said. The line is memorable. The operating metric behind it is time-to-reconfigure.

Why the timing looks right

Three signals sit around Arda's launch.

In June, Jeff Bezos said his company Prometheus had raised $12 billion at a $41 billion valuation to pursue an "artificial general engineer." The ambition overlaps directly with assembly planning.

On October 20, AGCO and Agility Robotics signed a commercial agreement to pilot the Digit humanoid at AGCO's Hesston, Kansas plant starting in 2027, initially for parts sorting and stacking.

On October 12, the U.S. Army announced its first Manufacturing Rodeo, a competition designed to stress the industrial base and expose where production breaks.

Each item points at the same constraint. Design tools improved. Factory engineering did not keep pace. The gap is now large enough to attract frontier-AI money and government attention at the same time.

That is also the risk. A crowded field makes hiring, customer access and integration speed the scarce resources. Arda's early Anduril deployment gives it a reference case that most competitors cannot buy.

What would break the thesis

The Anduril figures measure individual engineering tasks. They do not prove that factory output, unit cost or delivery time improved. Those outcomes depend on review time, error rates and how often a plan has to be redone.

Arda has not disclosed pricing, the number of active deployments or a timetable for updating an automated line in days. The robotics prototype is a prototype.

A $1.5 billion valuation also raises the bar for every quarter after launch. The software has to spread from one anchor customer to many, across industries with different tolerances and different definitions of a good assembly plan.

As we wrote in October, in hardware the factory has become the constraint. The production line now decides which designs reach the field. Arda is selling the tool that shortens that decision.

What would make Arda's bet pay off?