Product-led, not services-led
We build one product for many fabricators, not bespoke automation for one. It is a harder business and a better one.
About
An independentcompany, built forone hard process.
In brief
Weldeon is building original AI-native technology for welding and metal fabrication. Not a division, not a consultancy, not a reseller, not a systems integrator — a product company aimed squarely at the arc.
Next step
Thesis
Physical AI arrives where the constraint is physical
The last decade of AI investment went to processes made of text and pixels. The processes that make the physical world — cutting, forming, fitting, welding, inspecting — were left with dashboards.
Welding is the sharpest version of the opportunity. It is safety-critical, code-bound, craft-dependent, hazardous, rate-constrained and desperately short of people. It is also, unusually, a process where the sensing problem and the control problem are both tractable now: the sensors exist, the robots exist, and the physics is well understood.
What has been missing is the layer that closes the loop between them, holds a qualified envelope, and leaves a record an auditor will accept. That is Weldeon.
Beliefs
Five things we hold to
They show up in the product more than in the marketing.
- 01
Autonomy must be auditable
In a code-bound process, an action nobody can explain is worse than no action at all.
- 02
The envelope is sacred
The qualified WPS range is a hard boundary in the controller adapter. No agent, no support ticket, no override.
- 03
Sensing before control
Most welding automation fails because it assumes the joint. We measure it, every time.
- 04
Prove it on one cell
Big autonomy claims should be settled by a small, honest scorecard on the shop's worst station.
- 05
Say what is not true yet
Traction, certifications and customer results are marked plainly. Fabrication buyers reward that and punish the alternative.
Market
The opportunity, sized
$20B
TAM — welding and fabrication automation, inspection, weld data and factory software
$5B
SAM — autonomous welding, parameter control, weld-defect sensing and twins
$300M
SOM — three-year obtainable share
~12%
annual growth in the underlying market
Market figures are Weldeon's own estimates drawn from public industry sources, and are stated as estimates rather than audited data.
Timing
Why this, why now
Three curves crossed at once. Infrastructure and energy-transition demand put record volumes of welded steel into the order book — bridges, wind towers, pipelines, ships, rail. Reshoring added domestic fabrication capacity that has to be staffed. And the welder shortage moved from a hiring problem to a structural one.
At the same moment, the technical inputs matured: cheap high-rate sensing, GPU-capable industrial edge hardware, physics-informed models that run fast enough to plan a weld sequence overnight, and robot controllers that expose real interfaces.
A fabricator today can buy more welding robots than they can staff. The binding constraint has moved from capital to autonomy.
Moat
What compounds
Five sources of advantage, in the order they become durable.
- 01
Weld data
Arc signatures paired with adjudicated NDT outcomes are rare, expensive and enormously predictive. Every controlled weld adds to it.
- 02
Envelope and code models
Modelling AWS D1.1, ASME BPVC, ISO 3834 and EN 1090 as machine-enforceable constraints is slow, unglamorous work that does not need doing twice.
- 03
Adapter surface
Every power source, controller, tracker and NDT system integrated is a fixed cost paid once and reused across every customer.
- 04
Twin calibration
Twins fitted to real materials, fixtures and measured distortion get better with each site and each measured assembly.
- 05
System of record
Once weld genealogy runs on Weldeon, the platform is inside the quality system — the hardest place in a fabricator to displace.
“A fabricator can buy more welding robots than they can staff. The constraint has moved from capital to autonomy.”
Company
How we operate
Small, technical, and on the shop floor more than in the office.
Engineers on the floor
Every engineer visits shops. Models built without the smell of a fab shop are models that assume the joint.
Welding expertise in-house
Welding and quality engineers sit inside product, not in a customer-success function bolted on later.
Honest disclosure
Status that is unknown or aspirational is labelled. It costs us a little in a first meeting and saves us everything in the second.
Status
Where the company actually is
Stated plainly, because this is the section most companies fudge.
- Stage
- Early. Recruiting three to five design partners for paid pilots on high-ROI cells and lines.
- Customers and revenue
- [UNKNOWN] publicly. Disclosed under NDA to prospective customers, partners and investors.
- Incorporation and company age
- [ASPIRATIONAL] as stated publicly; formal corporate details are shared in diligence.
- NVIDIA Inception
- [ASPIRATIONAL]. Weldeon meets the profile of an independent, pre-IPO, AI-native startup; application status is not yet complete.
- Certifications
- SOC 2 Type II and ISO 27001 are on the roadmap and are not claimed until independently audited.
Writing
How we think, in longer form
The journal is where the engineering arguments live.
The welder shortage is a rate problem
What happens to takt when the constraint station cannot be staffed at all.
Why the envelope belongs in the adapter
Where to enforce a qualified range, and why every other location is wrong.
Fit-up is the weld decision nobody logs
The largest upstream cause of lack of fusion, and what measuring it changes.
Questions
About Weldeon
-
Founder and team detail is shared directly in conversation and in diligence materials rather than published here while the company is early.
-
Yes — welding and process engineers, perception and controls engineers, and deployment engineers who like factories. See careers.
-
Fundraising status is not published. Investors who want the data room should contact us directly.
Begin
Bring autonomy to thearc, the fit-up andthe inspection bay.
Book a pilot on one cell or line. We agree the scorecard before we start — first-pass yield, rework hours, defect escapes, takt — and we report against it in the open.