Test report DSG-7707 · Rev F · tested November 17, 2026

AI Devices & SystemsDevice under test

Skild AI Courts Korean Hardware Makers for Robot Push

Skild AI, valued above $14 billion, is courting Korean robot hardware makers. CEO Deepak Pathak expects factory robot adoption within one to two years.

Read
4 min
Words
825
Node
3nm
Operator
Priya Raman

Spec summary

  1. Skild AI is valued at more than $14 billion, with investment from SoftBank, Nvidia's NVentures, Bezos Expeditions, Samsung and LG Group.
  2. CEO Deepak Pathak says motors account for half the cost of a robot; Korea holds key hardware manufacturing strength.
  3. Since June, Skild AI's system has run on a Blackwell GPU assembly line at Nvidia's Houston plant, deployed in under a week.
  4. S1 robots learn tasks from a single demonstration video, replacing training on hundreds of hours of data.
  5. Pathak predicts factory robot adoption spreads within one to two years, before any ChatGPT moment for physical AI.
Skild AI Founder Seeks Deeper Ties With Korean Robot Hardware Makers
Fig. ASkild AI Founder Seeks Deeper Ties With Korean Robot Hardware Makers — marcin049 / pixabay

By Priya Raman

Skild AI is valued at more than $14 billion, has backing from SoftBank, Nvidia's NVentures, Bezos Expeditions, Samsung and LG Group — and runs its robot software on exactly three publicly named production sites. That ratio frames the pitch the Pittsburgh startup brought to Seoul this month.

Co-founder and CEO Deepak Pathak visited Korea to deepen ties with the country's hardware manufacturers. He outlined his case in an interview with The Seoul Economic Daily on the 10th at the Grand Hyatt Seoul in Yongsan District. He attended Deep Dive Into 2026, a technology forum hosted by the Ministry of SMEs and Startups and organized by the Korea Institute of Startup and Entrepreneurship Development. He is also expected to meet officials from LG Group and other major Korean companies during his stay.

Why Korea, and why now?

The argument rests on component economics. "Korea has outstanding manufacturing capability in hardware, including the design of motors, which account for half the cost of a robot," Pathak said. Motors, actuators and other robot components are where Korean suppliers hold an edge, in his view. "Outside China, there aren't many countries with real strength in robot hardware manufacturing," he said.

One counterweight to China's dominance — that is the strategic bet behind the outreach.

Founded in Pittsburgh in 2023, Skild AI develops robot foundation models, or RFMs — the software "brain" of a robot. Its differentiator: the model runs regardless of robot maker or machine type, whether humanoid, quadruped or robotic arm.

What does the technology actually do?

Pathak grounds the claim in in-context learning, or ICL — showing a model a few examples so it produces answers without further training. "ChatGPT became as popular as it is because of in-context learning," he said. "We applied that to robots and built a general-purpose model that works regardless of the hardware."

The company chose the brain over the body because hardware remains the binding constraint. "Take robot hands. There is still no hardware that can run for more than 100 hours straight and guarantee high-quality work," Pathak said. "With a powerful brain, a gripper is good enough to get the job done."

The flagship products are S1, the RFM, and Skild Brain, a robot intelligence system built on it. A robot running S1 watches a single video of a human performing a task and executes it immediately, with no additional training. Previously, teaching one task required hundreds of hours of collected training data.

"You have to teach a one-year-old everything, but an eight-year-old can figure things out on their own," Pathak said. "S1 is like an eight-year-old. That's why a robotic arm that was cooking can also flip a fried egg, even though that wasn't in the demonstration video."

The egg anecdote is a demo claim, not an audited benchmark. Buyers should note the distinction.

Where is it deployed today?

Since June, the system has run on a line assembling Blackwell graphics processing units at Nvidia's plant in Houston. The figures the company discloses:

  • About three hours of data collection
  • Two days of robot training
  • Three days of factory deployment setup

That totals under one week to automate the line, without changing existing factory equipment.

"Our solution has also been deployed at an LG Group battery plant," Pathak said. "Robots handle the chemical-handling tasks in battery manufacturing that are difficult for people to do, and workers control the robots from outside the area." A third deployment automates food packaging at a Market Kurly distribution center.

Three sites. Each is real, and each is narrow.

When does the "ChatGPT moment" arrive?

Not yet, by Pathak's own account. He said robots will spread inside factories within one to two years, before any broader "ChatGPT moment" for physical AI. Factories come first because tasks there are repetitive and standardized.

He conceded open problems. "In many cases you can't use the cloud on site, so you need robot models that conserve computing resources on-device and offer stronger security," he said. "We're also interested in combining digital intelligence with physical AI so that when a robot hits a problem mid-task, it can write code on its own to solve it."

Longer term, he sees near-unbounded demand. "Claude and ChatGPT are advancing math and science, but bottlenecks appear at the stage of physical experiments, because of the people involved," he said. "In that sense, the role robots can play in fields such as science and space will only grow."

In April, LG Group Chairman Koo Kwang-mo met Pathak and co-founder Abhinav Gupta in Silicon Valley to discuss cooperation. The Korean conglomerates are courting Skild AI as much as the reverse. Whether the $14 billion valuation survives contact with mass factory deployment — within the one-to-two-year window Pathak himself set — is the question the next twelve months will answer.

via en.sedaily.com (Source)

Filed under

  • skild-ai
  • robot-foundation-models
  • industrial-robots
  • humanoid-robotics
  • south-korea
Share this article:

More from Priya Raman

Priya Raman

Show full bio

Correspondent covering money, funding and market shifts around hardware layer of artificial intelligence at Die Signal.

661 articles

Same lot · LOT-362D

« Previous articleNext article »