Amir Frenkel's Noetive raises $41M to build world models for factories

Why it matters

Noetive is funding models, enterprise deployments and sensing hardware in one seed round, testing whether industrial AI can become infrastructure rather than another software copilot.

Amir Frenkel and Dan Barak launched Noetive on September 16th with a $41 million seed round and an expensive premise: industrial businesses need AI that can observe a factory floor, learn how the operation works and make decisions across the software and machinery already in place.

Eclipse led the financing after helping Frenkel create Noetive from its earliest stages. Craft Ventures, The Westly Group, Swish Ventures, Factory, Incite Ventures, Gigascale Capital, Operator Partners and Liquid 2 Ventures also participated, Noetive said in its launch announcement.

The individual backers make the round look partly like an extension of Frenkel's professional network. Meta CTO Andrew Bosworth, Airbnb CTO and former Meta generative AI executive Ahmad Al-Dahle, former Meta CTO Mike Schroepfer, Nest co-founder Matt Rogers and Meta Reality Labs Pittsburgh founding director Yaser Sheikh invested. So did Bedrock Robotics CEO Boris Sofman and the three co-founders of Decart AI.

Noetive calls itself an "AI NeoLab," a label that gives the founders room to operate as both a research group and a product organization. Its first commercial target is more concrete: production planning and operational coordination inside manufacturing, logistics and construction businesses.

Frenkel takes AI back into the physical world

Frenkel spent nearly a decade at Meta, most recently as vice president of generative AI, after earlier leadership work at Google and Amazon. At Google, he worked on Google Glass and wearable computing. His career has repeatedly moved between machine perception, consumer hardware and the software needed to make physical devices useful.

In April, Frenkel joined Eclipse as its first chief AI officer and wrote that he would build a new business with the venture firm. Five months later, that project has become Noetive.

The arrangement makes Eclipse closer to a co-builder than a conventional seed investor. Eclipse says it worked with Frenkel to form the thesis, recruit the founding group and move Noetive from an internal concept into an independent operation. That structure also explains the size of the seed round: Noetive is funding model research, enterprise deployments and proprietary sensing hardware at the same time.

Barak brings a different operating history. He began as an engineer, joined face.com early and led product work before Facebook acquired it. He later led growth product teams at Lyft and co-founded Stackbit, a visual web-editing platform that Netlify acquired in June 2023. At Noetive, Barak is co-founder and chief product officer, responsible for turning the research agenda into software industrial operators can use.

That pairing is central to the bet. Frenkel has spent years developing AI and perception systems inside large technology platforms. Barak has built products, closed an earlier startup and taken another through an acquisition. Noetive now has to compress those experiences into systems that can survive far less controlled environments than a web application or research demonstration.

A brain, plus eyes and ears

Noetive describes its product as an "intelligence of record" for physical operations. The phrase is a deliberate challenge to the systems of record that factories and logistics operators already use to track orders, inventory, equipment, staffing and schedules.

Noetive's proposed layer consists of a self-improving AI system that learns how an operation functions and agents that work across existing software. A multimodal sensing pod supplies what Noetive calls the system's "eyes and ears," collecting real-world information that may never reach an enterprise database.

The sensing hardware is the technically consequential part of Noetive's pitch. Software agents can read an order-management system or revise a production schedule. They cannot infer that material is sitting in the wrong place, machinery is moving more slowly than planned or a construction site has diverged from its drawings unless someone or something captures those conditions.

Noetive wants to combine those observations with digital records, giving its models a continuous view of plans and physical execution. The initial use cases include production planning, consolidating shipping manifests, planning pallets and comparing construction schedules with conditions on a job site.

Noetive says it is working with a small group of design partners in manufacturing, logistics, energy and data centers. The public product material also emphasizes construction, suggesting the founders are still testing where a shared world-model architecture can produce the clearest economic return.

Steuben Foods is the named early customer. CEO Menachem Katz said in Noetive's announcement that a production-planning process that previously ran monthly and required a week can now run daily and finish in minutes. That result is a customer account published by Noetive, rather than an independently measured benchmark, but it identifies the standard Noetive will face: fewer planning hours, faster adjustments and greater output from existing equipment.

The $41 million pays for fieldwork

Industrial AI carries costs that a browser-based agent can avoid. Noetive must gather data from fragmented enterprise systems, install sensing equipment, learn the operating constraints of each site and prove that model-driven decisions will hold up around workers, machinery and physical inventory.

The seed round gives Frenkel and Barak enough capital to pursue that work before forcing Noetive into a standardized software package. Noetive says it will use the financing for research into self-improving AI, field deployments and hiring AI researchers and full-stack engineers.

That freedom also creates the central product challenge. A research lab embedded with a handful of industrial customers can solve valuable problems through close collaboration. A venture-scale software business eventually needs deployments that repeat across facilities without recreating the research project for every customer.

Noetive's sensing pod may help establish that common foundation. Consistent hardware can give its models a standard source of physical data even when customers use different planning, inventory and equipment systems. It can also make deployments slower and more capital-intensive than software alone.

Frenkel's wager is that the complexity is defensible. Industrial operations contain years of unwritten knowledge, local workarounds and dependencies that conventional enterprise software records poorly. A system that can capture those details and improve as conditions change would sit close to the decisions that determine capacity, delivery times and operating margins.

Noetive has raised enough money to test that thesis in the field. The next measure is whether Frenkel and Barak can turn bespoke access to factory floors into a product that learns faster each time it enters a new one.