Jensen Huang told reporters on Monday that Nvidia is partnering with South Korea's LG Group on humanoid robots and data centers. LG Electronics had already surged 30 percent - hitting its daily trading limit - for a second straight session the prior week on news that LG Group Chair Koo Kwang-mo would meet with Huang. The stock is up more than 300 percent this year. The market heard "humanoid robots" and went to work pricing a $40 trillion future.
That's the wrong number to focus on. The real story is in the other half of that sentence: data centers. Specifically, what Nvidia needs LG to do - and what that reveals about where Nvidia's growth is going next.
The $81.6 billion reality check
Nvidia's latest quarter tells the baseline story. Q1 fiscal 2027 revenue hit $81.6 billion, up 85 percent from a year ago. Data Center revenue alone was $75.2 billion, up 92 percent. The stock trades near a $4.6 trillion market cap. These are numbers that make most TAM projections look like rounding errors.
Huang has repeatedly stated that humanoid robots represent a $40 trillion total addressable market. Put plainly, that is a 489 multiple on Nvidia's current annualized revenue. It is not a revenue forecast. It is a market-structure claim - the assertion that an entirely new category of physical AI demand will emerge and that Nvidia will sit at its foundation layer, the same way CUDA sits beneath AI training today.
That claim is plausible in direction, but irrelevant for anyone deciding what to do with Nvidia stock this year or next. What matters is the architecture play, not the TAM number.
What Nvidia is actually building
Last week at Computex, Huang unveiled the Isaac GR00T humanoid reference design - an open platform built on Unitree's H2 Plus hardware chassis, powered by Nvidia's Jetson Thor edge compute, with tactile hands from partner Sharpa. The system combines open software, edge AI inference, and synthetic data generation into one stack. Developers can build, train, simulate, and deploy humanoid robots without designing their own compute pipeline from scratch.
This is not a product launch. It is an architecture bet. Nvidia is doing with humanoid robots exactly what it did with AI data centers: build the reference design, open the software platform, and make every other player in the ecosystem dependent on its stack. The CUDA moat migrated from a training advantage to an inference and developer-platform lock-in. Isaac GR00T is that same playbook, transposed to physical AI.
But here's what makes the LG deal structurally different from a simple robotics partnership: LG already has a proven data center infrastructure business. LG signed a multi-billion-dollar annual cooling agreement with Microsoft for global AI data centers earlier this year - supplying cold plates, coolant distribution units, and chillers for liquid cooling systems. LG also showcased AI-driven data center cooling solutions with real-time optimization at Data Center World 2026.
Nvidia doesn't just need robot partners. It needs the companies that build the physical infrastructure - power, cooling, deployment - to adopt its architecture. That is where the Vera Rubin DSX comes in.
The DSX layer: Nvidia's play for the whole factory
In March 2026, Nvidia released the Vera Rubin DSX AI Factory reference design - a codesigned infrastructure blueprint that integrates compute, networking, power, and cooling into a single framework optimized for maximum tokens per watt. Then in June, Nvidia made the Omniverse DSX Blueprint generally available: a digital twin platform for designing and operating gigawatt-scale AI data centers from simulation through deployment.
What this means is straightforward. Nvidia is no longer just selling chips to data center operators. It is selling the architecture of the data center itself - and the simulation tools to plan it. Siemens, Vertiv, and Switch have already integrated DSX into their reference designs. LG is being pulled into the same orbit.
This is the transition I've been tracking: from Nvidia as a chip supplier to Nvidia as the infrastructure architect. Hardware sets the ceiling; the architecture sets the multiple. When Nvidia controls the reference design, the cooling vendor, the digital twin, and the software layer, it becomes much harder for customers to substitute out.
The LG partnership crystallizes this shift. LG is a cooling infrastructure company that wants into robotics. Nvidia is a compute company that wants into the whole factory. Their overlap - physical AI and next-generation data centers - is where both companies actually need each other.
Where the risk sits
Demand is not the issue. The issue is timing and contribution. Nvidia's robotics revenue is, for all intents and purposes, zero. The global physical AI market - encompassing industrial and humanoid robot procurement - was valued at $81.4 billion in 2025, with a projected 33.5 percent compound annual growth rate. That is fast growth from a small base. Goldman Sachs projects the humanoid robot market at $38 billion by 2035. Against Nvidia's $81.6 billion quarterly revenue, these numbers are not the next earnings driver. They are the option Nvidia is writing on a five-to-ten-year horizon.
However, the data center infrastructure layer is material now. If the DSX reference architecture takes hold and LG and its peers standardize around it, Nvidia captures value beyond chip sales - through software licensing, simulation subscriptions, and architecture lock-in. That is the hardware-to-software value migration in motion. It was hardware that drove Nvidia to a $4.6 trillion market cap. Software and architecture control will determine whether it extends further.
So what for capital allocation?
The debate is not whether Nvidia stays important in the AI trade. It is whether the return profile from here is still as compelling as the alternatives.
I still believe Nvidia has a pathway to $20 trillion by 2030, but much of that return is likely back-half weighted - concentrated in 2028-2030 as inference economics mature, software monetization scales, and physical AI moves from reference design to deployment. The LG partnership is a signal that this architecture play is accelerating, not that robotics revenue is about to appear on the income statement.
For existing Nvidia holders, this doesn't change the long-term thesis. It does change how you think about the return curve. If you entered at $1 trillion market cap, the asymmetry is still on your side. If you are evaluating Nvidia today at $4.6 trillion, the question is whether the next step of that climb is front-loaded or back-loaded. The data suggests the latter.
The LG deal matters because it shows Nvidia is not waiting for robotics to happen - it is building the architecture layer that makes robotics possible, and embedding that same architecture into the data center infrastructure that already generates $75 billion per quarter. That is the move. Not the $40 trillion number. The architecture.
What would change my view: if DSX adoption fails to gain traction among infrastructure vendors, if the Isaac GR00T platform does not attract developer mindshare in the next 12 to 18 months, or if Nvidia's data center revenue growth decelerates meaningfully while the physical AI story remains a narrative without revenue. Until then, the long-term position holds - but the timing of the payoff is the variable that determines whether you hold, trim, or rotate.
Demand remains robust. The infrastructure architecture play is the real story. The humanoid robots are the option, not the engine. Know the difference before you allocate.