Foundry Raises $100 Million for Autonomous Factories

Large companies have learned over the past couple years that they can get more mileage out of AI agents by rethinking their processes from scratch, rather than merely automating the existing ways of doing things. Manufacturers are beginning to learn the same lesson as they adopt robots powered by physical AI models.

One such manufacturer is Foundry, a startup that was founded less than a year ago, which has raised a new round totaling $100 million to build automated factories, the company told The Information exclusively. Marlinspike Partners led the Series A funding round, and Khosla Ventures, which led Foundry’s seed round, participated.

Foundry is designing its factories around AI-powered robots, rather than slotting robots into existing production lines. Those robots are organized into work cells that focus on skills such as cabling, fastening and precise placement, rather than making a particular product.

Due to that flexibility, Foundry can produce a variety of products for its customers within a single factory, following a manufacturing-as-a-service business model. It currently has a handful of customers in industries including military, energy, autonomous vehicles and telecommunications.

“The value of autonomous facilities is that you can build many things under one roof,” said co-founder and CEO Adarsh Kulkarni.

“The same model that has really learned fastening can work on satellites and drones and robot dogs and autonomous submarines,” he said. “We build drones on Monday, batteries on Tuesday, satellites on Thursday.”

That flexibility comes at the cost of some inefficiency, Kulkarni acknowledged, but he said versatility that enables fast iteration is more valuable to customers. “One mass producing automatic facility is highly efficient, but it's highly efficient in only one axis, which is building a million of exactly the same thing,” he said. “You have basically no design flexibility.”

Specialized factories will still make more sense for some mass manufacturing, such as producing iPhones or cars, at least for the next several years, Khosla Ventures founder Vinod Khosla said in an interview. But Foundry’s approach allows it to change the products it makes “rapidly based on market demand in very short cycles,” which opens an opportunity for factories in the U.S. to compete against cheaper labor in China, he argued.

“American factories have to be done differently: much more capex efficient, much more product flexible, much more demand responsive, and closer to the demand. You can't have 60-day shipping times,” he said.

Kulkarni previously ran the robotics division at Scale AI, the data-labeling startup part-owned by Meta, where he led work on third-party robotics data. His co-founder, Emmanuel Lamacchia, previously ran the Model Y program at Tesla.

Foundry recently moved to its first factory in Emeryville, Calif. The facility is about 130,000 square feet. Foundry is still setting it up, but it is already shipping products to paying customers.

The factory will eventually hold thousands of robots: primarily industrial arms wielding tools that Foundry develops, in addition to a fleet of vehicles to shuttle partially assembled products from one work cell to another.

One type of robot is conspicuously absent at the facility: humanoid robots that walk on two legs. The potential for such machines to automate manufacturing tasks has lifted the valuations of humanoid companies to astronomical heights in recent years, but Foundry’s aversion to humanoids indicates that the market for humanoids could be smaller than some proponents have suggested.

“I‘ve done my time with legged robotics, and I love it, and it’s amazing, but humanoids are very inefficient in a lot of ways,” said Kulkarni. “Deploying a humanoid in an existing manual facility might make sense, but if you’re building the building from the ground up, there is no reason you need to depend on that.”

(Khosla, who expects highly capable humanoids in the next two years, said he foresees humanoids becoming a “key component” of Foundry’s operations in the future.)

Foundry’s robots have their own limitations. Even after being trained for specific manufacturing jobs, AI-powered robots still struggle with some tasks, such as those requiring “micron-level precision,” Kulkarni said. For those tasks, Foundry employs hundreds of human technicians. Data collected from their work will feed into and improve its AI models.

The ability to hold onto data from making a wide variety of products is one advantage that Foundry has over companies that sell robots to factories and warehouses, some of which are reluctant to allow an outside company to collect data from their facilities.

“It‘s pretty easy for us to negotiate that we own the data,” Kulkarni said. “All of this training data is a byproduct of making our customers’ manufacturing lives easier.”

Khosla said that the ability for Foundry’s robots to learn more over time is key. “These are learning factories, so if they learn on assembling defense gizmos, they will apply it to assembling alarm clocks too.”

Here’s what else is going on…

Big Number

Meta Platforms is slightly increasing the size of its planned stock sales to fund AI investments to a total of $97.5 billion from $95 billion, the company said.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Physical Intelligence, which builds AI software for robots, is in talks to raise at least $1 billion at a valuation of roughly $20 billion led by existing backers Avenir and Lux Capital, Bloomberg reported.

Approaching.AI, which provides software and hardware for running AI models, raised more than RMB 1.3 billion (roughly $183 million) in a Series B funding round led by Zhejiang Finance Innovation, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital and Qianjiang Century City.

Hex Technologies, which provides software to track and analyze AI agents, raised $100 million in funding at a $1.2 billion valuation led by Peak XV Partners.

Infisical, which secures AI agents’ access to company credentials, raised $40 million in a Series B funding round at a $600 million valuation led by Elad Gil.

Machenta, which creates robotic AI models that can learn new tasks from one demonstration, raised $40 million in a seed funding round led by Creandum and Union Square Ventures.

Hyperbolic, which operates a GPU marketplace, raised $30 million in a Series A1 funding round led by Walden Catalyst.

Avante, which makes an AI benefits-management platform, raised $30 million in a Series A funding round led by Frist Cressey Ventures.

22one, which builds AI tools for law enforcement, has hit a valuation of nearly $400 million after raising funds from Sequoia, General Catalyst and Lux Capital.

Vera Health, which runs an AI clinical-evidence search engine, raised $28.5 million in a Series A funding round led by Threshold Ventures.

ClarityCare, which builds AI agents for healthcare payers, raised $16.3 million in a Series A funding round led by Headline.

Live2.ai, which helps AI to help brands manage their social media campaigns, raised ₹130 crore (about $14 million) in a pre-Series B funding round led by IvyCap Ventures.

Xplore AI, which builds reinforcement learning environments for physical engineering, raised a $11 million seed round co-led by Lightspeed Venture Partners and Base10 Partners. The company also released a new benchmark for physical engineering tasks called Real World Engineering Bench.

OpenAI plans to buy Google’s AI chips, The Information reported today, citing a person familiar with the matter. Google and Oracle also announced a partnership that will make Google’s tensor processing units available to Oracle cloud customers, the first time the AI chips would be made available on a hyperscale cloud platform outside of Google’s.

As part of OpenAI’s three science breakthroughs yesterday, its models disproved a foundational theory within cryptography, the AI company claimed on Monday.

Anthropic said it acquired Mithril Medicine, a one-person startup developing treatments for two rare forms of muscular dystrophy, a sign of the AI developer’s growing ambitions in drug discovery.

Theori, an AI cybersecurity firm, said it found a vulnerability that reveals the chains of thought from OpenAI models, rather than just summaries of their reasoning.

Justin Sun, founder of the blockchain startup Tron, launched a pilot program for the Justin Sun Prize on Monday. The prize will give awards for up to $1 million for discoveries in mathematics and aims to help integrate AI into math findings while encouraging continued rigor.

Axiom, which uses AI to formally verify math and computer code, announced that it had verified a math proof of the “Enormous Theorem” using AI.

This week on AI Deep Dive

Check out the most recent episode of AI Deep Dive. An up-close look into how frontier AI actually gets built — the models, the money, and the people behind them.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.