Here is a fact about the global economy that is easy to forget while everyone argues about chatbots writing poetry: most of the machines that actually make things are old, expensive, and not going anywhere. A CNC mill on a factory floor can cost as much as a house and last for decades. Its owner is not going to throw it out because someone in Silicon Valley launched a new model. And so the interesting question in industrial AI is not "how do we build a smarter machine?" It is "how do we make the dumb machine we already own a little less dumb, without unplugging it?"
EdgeCross, a small company headquartered in Seoul and known in Korean as 엣지크로스, has built its entire existence around that second question. Its answer is admirably literal. You take an existing machine - a compressor, a metal detector, a robotic arm, a CNC machine - and you attach a device called MODLINK to it. MODLINK listens. It reads the machine's data, streams it to the cloud, and, in the ideal case, lets the machine start telling you things it previously kept to itself: that it is running hot, that a part is wearing, that it will probably fail on Thursday.
This is not the flashiest thing in AI. It is arguably one of the most useful. When a machine breaks on a factory floor, the clock is money, and the gap between "something is wrong" and "here is what to do about it" is where fortunes and deadlines quietly evaporate. EdgeCross sells a shorter gap. That is the whole business, and it is a good business to be in.
EdgeCross doesn't replace your machines. It teaches the ones you already own to talk.
The company was not born as EdgeCross. It started in 2015 under the name VITCON (빛컨), founded by Kim Min-gyou, a mechatronics engineer - which is to say, someone who genuinely understands both the mechanical and the electronic guts of industrial equipment. For its first years VITCON did the sort of thing you would expect: IoT systems in 2016, smart-factory solutions and an early version of MODLINK in 2017. In 2018 it raised money from Korea Development Bank, SparkLabs Ventures, and Walden SKT Venture - a respectable roster for a young hardware-adjacent startup.
Then, in 2022, it did something that companies do all the time but rarely do well: it renamed itself. VITCON became EdgeCross. Ordinarily a rebrand is a logo, a new font, a press release nobody reads. This one was a strategy. The name points at "edge" - edge computing, the practice of doing the smart work close to the machine rather than shipping everything to a distant data center - and it signaled a pivot from selling bespoke smart-factory projects to selling something more repeatable: AI-powered software that any machine owner could subscribe to. Your name, if you are lucky, points at where you are going rather than where you started. EdgeCross's does.
"EdgeCross leads the process of transforming existing machines into smart machines - easily, quickly, and without hassle, for both manufacturers and users."
The Corner Office
A conversational-AI guy walks onto the factory floor
The pivot came with a change at the top. In 2023, EdgeCross moved from a co-leadership arrangement to a single-CEO structure and installed Hoon Paek - previously the CTO of Minds Lab, one of Korea's better-known conversational-AI companies (now operating as Maum AI). Kim Min-gyou, the original founder, stepped back into an advisory role.
On paper this is a slightly odd hire. You have a company that makes hardware for grease-covered machines, and you put a conversational-AI executive in charge. But the logic is sound, and it explains most of what EdgeCross has done since. The hardest problems in AI are not really about language; they are about context - knowing enough about a situation to say something useful. Industrial machines are drowning in context. They just have not had anyone fluent enough to translate it. Paek's bet is that the translation layer, from raw machine data to a sentence a human can act on, is where the value lives. It is a defensible bet.
The sensor is easy. The meaning is hard.
The Product Line
Three ways to make a machine less quiet
Hardware · since 2017
MODLINK
The AIoT edge device that started it all. It bolts onto existing machinery and collects and controls real-time data - no new equipment, no rip-and-replace. This is the box that turns a silent machine into a data source.
SaaS · subscription
MachineManager
The software layer. A subscription platform where owners remotely monitor, control, and manage MODLINK-equipped machines - spotting anomalies and optimizing operations without a six-month integration project.
AI · LLM · 2024
MachineGPT
The conversational layer. Ask a question in plain language - "what's wrong with this machine?" - and get an answer with supporting documents. It speaks multiple languages, so a foreign worker can ask in their own.
MachineGPT is the piece that makes the strategy click. EdgeCross frames it, plausibly, as a fix for a specific and growing problem: the shortage of people who actually know how to run and repair industrial machines. A veteran machinist retiring takes decades of intuition with them. A new hire, or a worker who does not speak the local language, cannot absorb that overnight. MachineGPT is meant to stand in as an always-available expert - one that has read the manuals, watched the data, and will answer at 3 a.m. Early deployments, the company says, have cut fault-response time and reduced the cost of training new employees. This is AI as a colleague rather than a replacement, which is both the more honest framing and, on a factory floor, the more useful one.
The Flywheel
Why 3.2 gigabytes a day matters
As of April 2024, EdgeCross reported more than 8,000 machines connected to its devices, collectively producing roughly 3.2GB of operational data every day. Those numbers are worth sitting with, because they are the actual moat. In AIoT, thousands of companies can stream data off a machine. The differentiation is in what you do with the stream - and every additional machine connected makes the next prediction sharper, the next anomaly easier to catch, the next answer from MachineGPT a little more grounded. The software is the product you sell. The data is the thing that compounds.
The business model reflects the pivot. Rather than one-off, custom system-integration work - lucrative but non-repeatable - EdgeCross sells recurring, plug-and-play SaaS, aimed at mid-market enterprises and the small and mid-sized manufacturers that make up the backbone of Korean industry and rarely have the budget for a from-scratch smart factory. Lowering the barrier to entry, it turns out, can be worth more than raising the ceiling.
The Business at a Glance
CategoryIndustrial AIoT
ModelHardware + SaaS
HQSeoul, South Korea
Team size~12
BuyersManufacturers & SMEs
Legal nameEdgeCross Inc.
Funding Trail
2018 · EarlyKDB, SparkLabs, WSV
Series A~$2.1M reported
Series A dateNov 2023
BackersKDB, BSK, SparkLabs
Use of fundsMarketing + senior AI/IoT hires
Figures reported publicly; treat amounts as approximate.
The Bottom Line
The unglamorous frontier
There is a version of the AI story that is all about the frontier models and the trillion-dollar valuations, and there is another version - quieter, and arguably more consequential - about the metal detectors, the compressors, and the robot arms that keep the physical economy running. EdgeCross lives in the second version. It is not trying to build a smarter machine from scratch. It is trying to be the layer that makes the enormous installed base of existing machines legible: monitorable, self-diagnosing, and, thanks to a language model, conversational.
The company is small - about a dozen people - and it operates in a crowded field of industrial IoT and edge-AI vendors, from ADLINK and Axiomtek to the sprawling smart-factory suites sold by the industrial giants. Its edge is specific and narrow: retrofit the machines you already have, then put a chat interface on top. Whether that narrow edge widens into something durable depends on the flywheel - whether those 8,000 machines become 80,000, and whether the daily gigabytes of data keep making the product smarter faster than competitors can catch up. But the underlying instinct is right. The biggest opportunities in AI are frequently hiding in the least glamorous places, and a machine that can finally tell you what is wrong with it is worth more than a machine that can write you a sonnet.
A machine that can tell you what's wrong with it is worth more than one that can write you a sonnet.
