$300 million. A $2.45 billion valuation. $610 million raised in total. A $600 million production agreement with the largest US military shipbuilder, a 17-fold revenue increase in 18 months, and a first-pass weld yield above 97%. Path Robotics announced that package on 10 November, and every number in it describes the same humble act: joining two pieces of metal.
That is the paradox. The most valuable young physical-AI company in the American industrial base this month is not selling humanoids, self-driving trucks, or a general-purpose robot brain. It is selling a welder that never gets tired, into a country that cannot find enough people willing to weld.
We have read a lot of factory-automation decks. Most of them sell a demo. This one sells a throughput figure to a Navy that is behind schedule on ships. That is why Georgian led the round at roughly four times the prior valuation, and why the real question is not whether $2.45 billion is too much. It is whether welding is a wedge into the rest of the factory or a very well-funded cul-de-sac.
The trade with 80,000 empty chairs
The American welding shortage is not a venture-capital talking point. About 80,000 welding jobs need to be filled every year through 2029, and a large share of the current workforce is nearing retirement. In the 2026 Reshoring Initiative survey, US manufacturers ranked the supply of skilled workers above cost as the main obstacle to bringing production home. A shipyard can buy steel. It cannot buy welders who do not exist.
Conventional welding robots do not close that gap. They replay a taught path, joint by joint. Heat warps metal, parts arrive slightly out of tolerance, and a fixed program welds into the gap or misses the seam. Those economics work for long runs of identical parts, which is the opposite of what a submarine or a destroyer programme looks like.
CHRONICLE: Path Robotics
────────────────────────────
2018 · 🏠 Lonsberry brothers found the company in Columbus, Ohio
2026 · 🚢 $600M performance agreement with HII under HYPR
2026 · 💰 $300M Series E at $2.45B; $610M raised in total
◉ NOW · 🐕 Rove quadruped carries the welding model to the part
🔥 NEXT · 🏗️ Shipyard deployments in 2027; 50 Rove preorders
Chronology: Path Robotics release, Georgian investor memo and WorkBoat, November 2026.
From a family garage to a dry dock
Andy and Alex Lonsberry learned to weld with their father in the family garage, then built off-road vehicles at his factory. They founded Path in 2018 with a narrow thesis: pick the task that manufacturers most want to automate and least want to do. Welding qualified on both counts.
Their claim is that the company has collected tens of millions of welded inches of real production data over eight years. That corpus is the asset. It is also the part of the story a competitor cannot copy with a better model architecture, which is what Georgian is really buying.
The commercial path ran from small and mid-sized job shops toward larger, more complex customers. Path now operates a 180,000-square-foot plant with more than 30,000 square feet of warehouse, grew headcount 97% year over year, and expects to reach 400 employees by the end of 2026. Customers named in the release include HII, Lockheed Martin and Saronic, plus mid-sized manufacturers such as EK Machine, Millerbernd Manufacturing, TransAmerican Power Products and TYCROP.
The anchor deal is the $600 million performance-based production agreement with HII, signed earlier in 2026 under the High-Yield Production Robotics programme. Georgian describes a seven-year horizon and a scope covering both crewed and uncrewed naval platforms. HII frames it as throughput. Eric Chewning, the shipbuilder's executive vice president for maritime systems, said the partnership is meant to grow the maritime industrial base and advance adaptive automation in fabrication.
We started with welding because it is one of the hardest jobs on the factory floor to automate, and we've proven that our technology can succeed in real production environments.— Andy Lonsberry, CEO and co-founder, Path Robotics
What Obsidian actually decides
Path separates the problem into two pieces. The Weld World Model is a neural simulator trained on high-fidelity welding data; it encodes what a weld is. Obsidian is a reinforcement-learning agent that runs inside the deployed robot, reads the sensor stack and decides, seam by seam, how to execute. Four lasers on the torch and two cameras capture the geometry before, during and after each pass.
The company reports first-pass yield above 97%, against 70% to 85% for human welders. Treat that comparison as a company figure until an independent shipyard audit appears. The mechanism behind it is more persuasive than the percentage: the system adjusts while the metal moves, which is the exact failure mode of a taught path.
Share of welds that pass on the first attempt
Company-reported figure for Path's cells; the human range is the comparison the release supplies. · Path Robotics, November 2026
Rove is the second form factor, and the more interesting one commercially. Unveiled in April on a Boston Dynamics quadruped, it puts the same model on a walking body so the robot travels to the part instead of the part travelling to a cell. Path is working through 50 preorders, with Saronic among the first adopters, and says Rove will start alongside human welders in shipyards, construction sites and heavy industrial sites in 2027.
That sentence is a promise, not a deployment. Welding is dirty, hot and electrically hostile; a quadruped that survives a shipyard for a full production cycle would be a genuine engineering result. It is also the natural next step: if the model is body-agnostic, the same software can address tasks before and after welding, which is exactly what the new capital is meant to fund.
The numbers the release left out
The 17-fold annual recurring revenue growth is a multiple without a base. Path does not disclose the dollar figure, so the growth rate cannot be sized against the $2.45 billion valuation. A performance-based $600 million agreement is a ceiling tied to throughput milestones, not an order book. Both facts are normal for a private company. Both matter for anyone trying to underwrite the round from the outside.
| Company | Round | Wedge | Named customers |
|---|---|---|---|
| Path Robotics | $300M Series E at $2.45B | Welding cells and the Rove quadruped | HII, Lockheed Martin, Saronic |
| Nidus | $30M pre-seed and seed | Wire-harness assembly software on existing arms | None disclosed |
| Foundry Robotics | $100M Series A | Autonomous assembly factory as a service | Defense and industrial, unnamed |
Funding and customer disclosures from company releases and Dealroom, October–November 2026.
The competitive field is real. Fanuc, ABB and Lincoln Electric already sell welding automation into the same shipyards, and they have service networks Path does not. Machina Labs raised $124 million in February, with Lockheed Martin Ventures participating, to attack adjacent metal-forming work. Nidus raised $30 million on 10 November to put embodied AI on wire harnesses inside defense factories. As we wrote in November, Foundry Robotics took $100 million for autonomous assembly plants.
What separates Path from that group is not the model. It is the shipyard contract. A named prime customer with a performance-based agreement, references from two other primes, and a seven-year horizon is the rarest asset in industrial automation. Foundry and Nidus are still refusing to name a single customer.
A disclosed ARR figure that makes 17x legible against the $2.45B mark.
Evidence that the 97% first-pass yield holds on naval work, not just on Path's own cell data.
A Rove unit completing a full shipyard production cycle, which is a 2027 question.
What the next weld has to prove
The round is best read as a bet on the bottleneck, not on the robot. American shipbuilding has orders, political will and no welders. A company that converts that shortage into throughput gets paid whether or not its model is the most elegant in the field.
The $600M HII agreement is a throughput ceiling, not revenue already booked, and the 17x growth has no disclosed base.
The next real proof point is a shipyard running Rove through a production cycle, not another funding announcement.
The first weld these robots made was a demonstration. The next one has to hold a destroyer together, on schedule, with a Navy inspector watching. That is a harder audience than a venture board, and a better one.