Shipbuilders have a problem they cannot hire their way out of. The work is dangerous, the pay is contested, and the people who know how to do it are retiring faster than they can be replaced. So the industry is placing a different bet: a human-sized robot that learns to weld by being walked through the job.
The bet targets a real constraint, not a demo. U.S. shipbuilding trades face roughly 30% annual turnover, and more than 40% of the skilled workforce retires by 2030.
The robots are not ready. Prototypes land at the end of 2026 and commercial deployment begins in 2027. Today's units are teleoperated while they collect the data that will train autonomy.
The welder nobody can hire
The math is brutal. The U.S. shipbuilding and fabrication trades need an estimated 320,500 welders by 2029, and the pipeline is not close. The Navy reports about 30% turnover in skilled trades. By 2030, more than 40% of the experienced shipyard workforce will have retired, and fewer young workers are entering the trade.
US fabrication welding gap
Estimated shortfall across shipbuilding and heavy fabrication. · Persona AI / industry estimates, 2026
Shipyards are also hazardous. OSHA data puts the injury rate more than twice that of general industry, driven by fumes, falls, confined spaces, and heat. That is a direct return-on-investment target for any system that removes a person from the worst tasks.
Persona AI is not alone in seeing it. As we wrote in August, the defence robotics capital stack is splitting into buyers and builders, and the labor squeeze is exactly what pulls builders toward narrow, defensible niches instead of general-purpose machines.
A narrow bet, deliberately
Most humanoid programs lead with warehouse picking or home chores. Persona AI went the other way: a robot sized to a welder's envelope, built for the uneven floors, scaffolding, and tight access points of a live yard. The Gen 1 unit stands 175 centimetres, weighs 85 kilograms, and uses tendon-driven hands with about 12 degrees of freedom in the end effector so it can hold ordinary welding tools.
The founding team is the part worth watching. Nicolaus Radford ran NASA's Valkyrie humanoid program and later led Nauticus Robotics. Jerry Pratt was chief technology officer at Figure AI before joining as Persona's CTO. The $27 million pre-seed, led by Unity Growth and Tides Ventures with Mirae Asset and Embedded Ventures, is small for hardware but enough to reach a prototype.
Today the robot is guided by a human operator collecting real welding data. Autonomy is the destination, not the current state.
Two deals that matter
The first is HD Hyundai. A memorandum of understanding in May 2025 became a formal development agreement in March 2026, signed with HD Korea Shipbuilding & Offshore Engineering (HD KSOE), HD Hyundai Robotics, and the welding-tool firm Vazil. Persona AI builds the bipedal platform and the learning stack; HD KSOE supplies shipyard data and field validation. Prototypes are due by the end of 2026, with field testing and commercial deployment in 2027.
The second is the State of Louisiana. A January 2026 pilot at SSE Steel Fabrication in St. Bernard Parish trains humanoid welders in a working heavy-industry plant. The company has also partnered with the American Bureau of Shipping (ABS) on robotic data-certification standards, and with Under Armour on protective gear for the harsh yard environment. The certification work is the quiet part that matters: without a standards pathway, a pilot stays a pilot.
| Parameter | HD Hyundai deal | Louisiana SSE pilot |
|---|---|---|
| Goal | ✔ Shipyard welding at scale | ✔ Train humanoid welders in-plant |
| Timeline | ◐ Prototype 2026, deploy 2027 | ◐ Pilot running, data phase |
| Proof status | ✗ Signed, pre-production | ✗ Training, not production |
Both deals are real commercial validation for the niche, not evidence the robot already welds unsupervised at production quality.
Why shipyards, not warehouses
Shipyards are the largest real-world proving ground a humanoid can ask for, and the dual-use logic is hard to ignore. HD Hyundai builds naval vessels alongside commercial ships, and naval maintenance, repair, and overhaul is a defence-adjacent market with the same labor problem. The company frames its robots for "3D" work: dull, dirty, and dangerous.
It is not a sole play. Italy's Generative Bionics, an IIT spinout, is developing a welding humanoid with Fincantieri, with onsite testing expected by the end of 2026. Path Robotics has put a welding robotic dog named Rove into a Louisiana yard run by autonomous-vessel maker Saronic. NEURA Robotics paired with HD Hyundai Samho on legged shipbuilding robots. The field is crowding from both the humanoid and quadruped directions.
Every partner describes "smart shipyards where humans and robots work side by side." The current unit is teleoperated. The autonomy that justifies the valuation arrives only after the 2026 prototypes collect enough welding data to train it, and that timeline has slipped before in humanoid robotics.
When does a shipyard humanoid move from pilot to production?
Probability: 35%. The HD Hyundai timeline is explicit, but "commercial deployment" in a shipyard means a handful of supervised cells, not a fleet, and autonomy quality on real seams is unproven.
✅ Arguments for
A human-sized robot reuses existing scaffolding and tools, avoiding the retrofit cost that kills most industrial-robot deployments.
Confirmation criteria: HD Hyundai fields a working prototype in 2026 and reports unsupervised weld-quality parity on at least one seam type.
❌ Arguments against
Competitors with deeper pockets (HD Hyundai's own NEURA work, Figure, Apptronik) could own the platform layer and leave Persona as a niche welding app.
Disconfirmation criteria: 2027 passes with no unsupervised production weld, or HD Hyundai shifts to its in-house legged-robot program.
Development scenarios
🟢 Optimistic scenario (25%)
Implications: The startup becomes the reference welding-humanoid vendor and attracts a Series A at a step-up valuation.
🟡 Base-case scenario (50%)
Implications: A real but slow niche business; the "robot welder" stays a supervised tool, not a replacement, through the decade.
🔴 Pessimistic scenario (25%)
Implications: The niche proves real but the startup lacks the capital to scale past prototypes; an acquirer absorbs the team.
HD KSOE prototype demo by end of 2026 and whether it welds unsupervised
Louisiana SSE Steel pilot moving from training to a measured weld-accuracy number
ABS certification pathway for robotic welding data, the gate between pilot and production
Whether HD Hyundai's NEURA partnership stays separate or converges on Persona's niche