Malcolm Reeves: Hey — before we get into anything, I need to name something that I think is genuinely the strangest detail in a week full of strange details.
Malcolm Reeves: NVIDIA's academic reference humanoid — the physical robot they use to demonstrate Isaac GR00T, their open humanoid foundation model — runs on a Unitree H2 Plus chassis. Unitree is a Chinese robotics company. And on August thirteenth and fourteenth of twenty-twenty-six, Jensen Huang and LG Chairman Koo Kwang-mo announced a partnership to build a next-generation bipedal humanoid on that same platform, explicitly framed as a US–South Korean strategic alliance.
Cole Brennan: Hold on — the alliance is built on a Chinese chassis.
Malcolm Reeves: The blueprint, yes. The thing NVIDIA hands developers and says 'this is the reference.' And yet the story being told is allied nations versus the Chinese robotics ecosystem. Those two things are both true simultaneously, and I don't think the announcement wants you to hold them at the same time.
Cole Brennan: That's — huh. Because Jensen Huang is standing next to Koo Kwang-mo, they're both quoted in the same release, Q1 2027 public unveiling, this grand technological handshake. And underneath it, Unitree. I mean — that's not a footnote, that's the structure.
Malcolm Reeves: And so what we're really trying to work out today is whether the gap between that narrative and that infrastructure tells us something true about what this announcement actually is.
Cole Brennan: Yeah — and I want to know who it's for. Like, is this about the robot, or is this about proving something else entirely?
Malcolm Reeves: And the 'who it's for' question — that's actually where the unveiling detail cracks open. Think of Isaac GR00T as the operating system and LG's robot as the first name-brand laptop built on it. The OS isn't finished. But the laptop announcement makes the OS feel real.
Cole Brennan: Wait — so Q1 2027 isn't really a robot launch.
Malcolm Reeves: It's a platform checkpoint. Now, Project GR00T was unveiled at NVIDIA GTC in March 2024. Over two years of public development. Then the Isaac GR00T Reference Humanoid demo at Computex, June first of this year. And now LG commits to Q1 2027. The checkpoints keep advancing — but there's no at-scale deployment anywhere in that sequence.
Cole Brennan: Right — but the word they used is 'unveiling.' Not launch. Not production. Unveiling. A robot that walks across a stage and picks up an object — that counts. That clears the bar.
Malcolm Reeves: It does. And the audience watching isn't consumers — it's every appliance manufacturer, every industrial supplier asking: do I build my own stack or license NVIDIA's?
Cole Brennan: Which is — okay, actually that reframes the Tennessee thing for me. LG's CLOiD wheel robots validating at their washing machine manufacturing line there — I kept reading that as a footnote, but it's, wait, no — that's the proof-of-concept run before the humanoid even gets on stage.
Malcolm Reeves: You see, the CLOiD line is the dress rehearsal. The humanoid is the premiere.
Cole Brennan: And then there's the Cheonan facility — eighty megawatts, targeted for 2028. Nobody leads with that number. That's not robot infrastructure, that's platform infrastructure. LG is building the server farm that runs the thing before the thing has proven it works.
Malcolm Reeves: Which means the risk isn't the robot failing on stage. The risk is they've already committed eighty megawatts and a Vera Rubin reference site to a platform that hasn't demonstrated scale. The unveiling just has to be convincing enough to justify what's already been spent.
Cole Brennan: And that's the layer under the layer, right — because every single piece of intelligence on this robot is NVIDIA. Like, not mostly. Every piece. Jetson Thor for onboard compute, Isaac GR00T as the foundation model, Halos for Robotics handling safety, Isaac Lab doing the simulation training — LG is, I mean, LG is assembling a chassis around someone else's entire nervous system.
Malcolm Reeves: And Isaac Lab is the part people skip over. Published November 2025, successor to Isaac Gym — it's where the robot learns, at GPU scale, before it ever touches a physical floor. Every hour of sim training happens inside NVIDIA's environment. LG doesn't own that either.
Cole Brennan: Plus Cosmos for sim-to-real transfer.
Malcolm Reeves: Cosmos, yes. World foundation models — that's the bridge between simulation and actual factory floor behavior. And now consider: in July 2026, NVIDIA announced NEC, Hitachi, OMRON, SoftBank, Preferred Networks — Japanese manufacturers — all building on Cosmos and the same physical AI stack. LG isn't pioneering this. LG is joining a roster.
Cole Brennan: Wait — so the Q1 2027 unveiling isn't LG's moment, it's NVIDIA's proof point. LG just happens to be the one on stage.
Malcolm Reeves: That's — yes. And here's who's actually watching that stage. Not consumers. Picture a factory floor supervisor in Osaka — call her Daria — and she is not watching the robot walk. She's watching whether her board will greenlight licensing NVIDIA's stack instead of building their own. That's the audience NVIDIA needs.
Cole Brennan: And LG actually has something real to show her, because — this is the part that's genuinely interesting — LG Innotek on actuators and sensors, LG Energy Solution on batteries. That vertical integration across hardware is unusual. Most players don't have that. But the actual intelligence running those actuators? NVIDIA's. So Daria sees a polished robot and the question she's asking is about software licensing, not about whether LG built something.
Malcolm Reeves: And nobody's independently validated that sim-to-real transfer actually works at factory scale. Isaac Lab's multi-modal learning is — it's published, it's not proven in deployment. That gap is where this whole announcement could unravel.
Cole Brennan: And the Unitree thread makes it stranger — we'll get there, but that detail does something to the geopolitical framing that I don't think the announcement survives intact.
Malcolm Reeves: And the framing doesn't survive the supply chain — that's exactly right. But now consider NVIDIA's actual incentive here. Because if Unitree runs Isaac GR00T, and LG runs Isaac GR00T, and Hitachi runs Isaac GR00T — NVIDIA collects on the software stack regardless of which nation's manufacturer wins the hardware race. The geopolitical narrative is real to LG. It's irrelevant to NVIDIA.
Cole Brennan: Wait — NVIDIA wins whether Unitree wins or LG wins.
Malcolm Reeves: By design. The platform is agnostic to national origin. That's not a bug in the strategy.
Cole Brennan: Which is actually — I mean, Figure AI saw exactly this and walked. They exited the OpenAI partnership in February 2025, started training Helix fully in-house. Like, they looked at the dependency math and said no. And Figure is one of the serious players.
Malcolm Reeves: Is that right — fully in-house from February 2025?
Cole Brennan: Fully. Their Helix VLA model — built without an external foundation model platform. No licensing dependency. And LG is moving in the exact opposite direction at the exact same moment in the market.
Malcolm Reeves: And that's where the Isaac GR00T 'open' language gets uncomfortable. Because nobody's actually defined what open means here — open-source, open-access, open only to partners running Jetson Thor? The distinction determines whether LG built on a public good or signed a licensing arrangement they'll have to renew.
Cole Brennan: And there's no independent validation on Cosmos or GR00T benchmarks either — NVIDIA's own numbers, no third-party verification. So the robot that LG unveils in Q1 2027 is demonstrating a stack that, actually wait — nobody outside NVIDIA has confirmed actually performs the way the benchmarks say.
Malcolm Reeves: And Unitree is building on that same unvalidated stack. The allied-nations framing is describing a press release. The supply chain is describing something considerably more tangled.
Cole Brennan: And the Cheonan facility is the part that won't leave me alone. Eighty megawatts, 2028 — that's not contingent on Q1 2027 going well. That commitment already happened. LG is restructuring as an NVIDIA-powered industrial AI company whether the robot impresses anyone or not.
Malcolm Reeves: And that's the honest place this lands, I think. The Q1 2027 stage moment — whatever LG shows — the proof of concept isn't LG's and it isn't NVIDIA's separately. It belongs to the question every major manufacturer watching has to answer. Do you build your own stack or do you license NVIDIA's? Jensen Huang and Koo Kwang-mo weren't just announcing a robot. They were setting a table that other people have to sit at.
Cole Brennan: Yeah. And the people making that call — they weren't in the room in Seoul.
Malcolm Reeves: No. They weren't. And that's — I think that's where I have to leave it. Genuinely unresolved. The eighty megawatts is already committed. The robot might work. The stack might not scale. And NVIDIA collects either way.
Cole Brennan: Good conversation. Heavy one.
