CES 2026 marks the arrival of "Physical AI"—machines that perceive, reason, and act in the real world. From NVIDIA's thinking self-driving cars to Samsung's AI fridges and humanoid home robots, here's everything you need to know about the biggest AI announcements at this year's show.
Back to work week is coming in rough this Tuesday. We were low-key enjoying not having a flood of AI news to sift through every day over the last two weeks (just kidding, it somehow still poured down with AI news over the holiday break), and then CES hit like a ton of bricks.
Official back to work week after the holidays is coming in rough this Tuesday. We were low-key enjoying not having a flood of AI news to sift through every day over the last two weeks (just kidding, it somehow still flooded AI news over the holiday break, just like… not as hard??), and then CES hit us like a ton of bricks yesterday.
The vibe this year? As best we can tell via armchair quarterbacking the show from afar, AI is going everywhere: your neck, your glasses, your pets, your cat's food bowl, even your haircut, heck, even a piece of wood!
Here's a few examples of what we mean:
But really, the first day of CES was all about NVIDIA. Jensen Huang took the stage in Las Vegas and said something that might define the next decade of AI: "The ChatGPT moment for physical AI is here—when machines begin to understand, reason, and act in the real world." Bold claim. Can he back it up?
Well, if CES 2024 was about generative AI and CES 2025 was about AI assistants, CES 2026 is about AI that actually does things. We're talking robots that fold your laundry, cars that explain why they're braking, fridges that track what you eat, and chips so powerful they need hot water cooling systems.
After parsing through keynotes, press releases, and demos, here's your complete guide to everything that matters from this year's show so far, and what it means for how you'll work, live, and interact with technology in the years ahead.
NVIDIA dominated CES 2026 with a staggering number of announcements. But the through-line was clear: the company is building the entire AI stack—from the silicon to the software to the simulation tools—that will power everything from your car to your hospital to your robot butler.
The biggest headline from NVIDIA's keynote was Alpamayo, which Jensen called "the world's first thinking, reasoning autonomous vehicle AI."
Here's the problem Alpamayo solves: Traditional self-driving systems separate perception (what the car sees) from planning (what the car does). This works fine for 99% of driving situations. But that remaining 1%—the weird edge cases, the unexpected scenarios—is where accidents happen.
Alpamayo 1 is a Vision Language Action (VLA) model that does something fundamentally different. Instead of just processing camera feeds and outputting steering commands, it reasons about what it sees. It can explain why it's making a decision, describe the scenario it's navigating, and think through novel situations step-by-step—kind of like how a human driver talks through a tricky merge.
The technical term is "chain-of-thought reasoning." The practical implication? Cars that can handle situations they've never seen before, because they're actually thinking through the problem rather than pattern-matching against training data.
At the 3:29 mark of the keynote, Huang explained how the system works: "Not only does it take sensor input and activates steering wheel, brakes and acceleration, it also reasons about what action it is about to take. It tells you what action it's going to take, the reasons by which it came about that action, and then of course the trajectory."
NVIDIA is also releasing AlpaSim, an open-source simulation framework that lets developers test these reasoning models in simulated environments before deploying them on real roads. Think of it as a flight simulator for self-driving cars—but one where you can create literally any scenario imaginable.
The first car to use this technology? The Mercedes-Benz CLA, which just received NCAP's safest car rating. It's hitting European roads in Q1 2026 and the US shortly after.
Major automotive partners are already on board. Lucid Motors, JLR, and Uber have all expressed interest in building on Alpamayo. Sarfraz Maredia, Uber's global head of autonomous mobility, called it "exciting new opportunities for the industry to accelerate physical AI."
Why this matters for you: If you've ever been nervous about trusting a self-driving car, Alpamayo represents a fundamental shift. These cars won't just drive—they'll be able to explain why they're driving that way. And when something goes wrong, there will be an actual reasoning trace to examine, not just a black box.
NVIDIA announced its next-generation AI computing platform: Vera Rubin. Named after the astronomer who discovered evidence for dark matter, it's designed to handle the skyrocketing computational demands of AI training and inference.
The numbers are staggering. A Rubin pod consists of 1,152 GPUs arranged across 16 racks. Each rack contains 72 Rubin chips, and each chip is actually two GPU dies connected together. The power consumption is twice that of the previous-generation Grace Blackwell system—but here's the engineering miracle: it uses roughly the same airflow and can be cooled with 45°C water. No chillers needed.
"We're basically cooling this supercomputer with hot water," Huang said during the keynote.
The platform includes several new components:
NVIDIA is positioning this as the infrastructure layer that every AI application will eventually run on—from training massive language models to running inference for millions of users simultaneously.
Why this matters: More computing power means more capable AI models. The GPT-4 that wowed everyone in 2023 was trained on the equivalent of thousands of GPUs running for months. The models trained on Vera Rubin infrastructure will make that look quaint.
Beyond the hardware announcements, NVIDIA released a massive collection of open-source models, data, and tools designed to accelerate AI development across industries. This is significant—NVIDIA could have kept these proprietary, but instead they're seeding the entire ecosystem.
Nemotron for Agentic AI:The Nemotron family got significant updates:
Bosch is using Nemotron Speech to let drivers talk naturally to their cars. ServiceNow is training on Nemotron datasets. CrowdStrike and Fortinet are building security applications on the safety models.
Cosmos for Physical AI:The Cosmos platform got major updates for robotics and autonomous systems:
These aren't just research projects. Salesforce, Uber, and Hitachi are already using Cosmos models for traffic and workplace AI agents. Franka Robotics and NEURA Robotics are using Isaac GR00T to train their robots.
Clara for Healthcare:NVIDIA also announced new models for drug discovery and medical research:
Plus, they're releasing 455,000 synthetic protein structures to help AI researchers build better models.
Why this matters: Open-source AI models accelerate innovation across the entire industry. Instead of every company building from scratch, developers can start with NVIDIA's models and customize for their specific needs. The more companies building on these foundations, the faster AI capabilities advance.
While NVIDIA had the most to share today, there was a lot more than that happening at CES this year. Let's dive into some of the other pieces.
Samsung's CES 2026 presentation centered on a single vision: "Companion to AI Living." The idea is that AI shouldn't just respond to commands—it should anticipate needs, understand context, and proactively help throughout the day.
Samsung's Family Hub refrigerator is getting a major AI upgrade with AI Vision built with Google Gemini. Internal cameras now track everything that goes into and out of the fridge, using Gemini's multimodal capabilities to identify food items even in tricky lighting or unusual packaging.
The practical applications are impressive:
Samsung also partnered with Hartford Steam Boiler (HSB) to offer insurance discounts for homes with connected SmartThings appliances. After a successful US pilot, the program is expanding globally. The logic: connected appliances can detect problems early, reducing the risk of expensive claims.
As of December 2025, SmartThings serves more than 430 million users—giving Samsung an unprecedented view into how people actually use their homes.
Samsung's new Vision AI Companion (VAC) transforms their TV lineup from passive displays into active assistants. Available across Micro LED, Micro RGB, OLED, Neo QLED, Mini LED, and UHD models, VAC uses contextual AI to enhance the viewing experience:
The display lineup itself got major upgrades. The 130-inch Micro RGB uses individual red, green, and blue light-emitting diodes—each microscopic—to produce what Samsung claims is the widest, most detailed color spectrum ever in a Samsung TV. The AI Engine Pro provides precise control over those colors in every scene.
For sports fans, AI Soccer Mode Pro uses AI to tune picture and sound to stadium quality, while AI Sound Controller Pro lets you adjust the volume of crowd noise, commentary, and background music independently.
Samsung's home appliance lineup is getting smarter across the board:
Why this matters: The "smart home" has been a promise for decades. What Samsung showed at CES 2026 feels different—less about connecting devices and more about those devices actually understanding what you need. The combination of Gemini's reasoning capabilities with Samsung's massive installed base could finally make the AI-powered home feel natural rather than novelty.
If Samsung's approach is embedding AI into existing appliances, LG's approach is building an entirely new category: the home robot.
LG CLOiD is designed around a "Zero Labor Home" vision—a future where household chores are handled by intelligent machines, freeing humans for more meaningful activities.
At CES 2026, LG demonstrated CLOiD performing realistic household tasks:
These might sound simple, but they represent genuine robotics challenges. Manipulating soft materials like clothing, navigating varied home environments, coordinating with appliances—each of these has been an unsolved problem for home robotics.
CLOiD consists of three main components:
The Arms: Two articulated arms with seven degrees of freedom each—matching human arm mobility. The shoulder, elbow, and wrist allow motion in all directions, while each hand includes five independently controlled fingers for fine manipulation.
The Base: A wheeled platform using autonomous navigation technology developed from LG's robot vacuum line. LG chose wheels over legs for stability and safety—a low center of gravity means the robot won't tip if a child or pet bumps into it.
The Head: Functions as a mobile AI home hub, equipped with a main processing chip, display, speaker, cameras, sensors, and voice-based generative AI. It can communicate through speech and "facial expressions," learn household patterns, and control connected appliances.
CLOiD runs on what LG calls "Physical AI"—a combination of two model types:
These models have been trained on tens of thousands of hours of household task data, enabling CLOiD to recognize appliances, understand user intent, and execute appropriate actions.
CLOiD integrates with LG's ThinQ smart home ecosystem and the ThinQ ON hub, allowing it to coordinate across all connected LG appliances.
Alongside CLOiD, LG announced LG Actuator AXIUM, a new brand of robotic actuators (the joints that make robots move). Actuators are one of the most expensive and critical components in any robot—think of them as the muscles that translate motor power into movement.
LG's decades of experience building motors for appliances gives them a natural advantage here. The AXIUM line promises lightweight, compact, high-efficiency actuators that could make advanced robots more affordable.
Why this matters: Home robots have been "just around the corner" for twenty years. What's different about CLOiD is that it's built by a company that already has products in millions of homes, understands how people actually live, and has the manufacturing scale to make these devices affordable. The "Zero Labor Home" might still be years away, but LG just demonstrated that the underlying technology works.
If LG is approaching home robotics from the premium end, SwitchBot is coming from the accessible end. The company that made its name with affordable smart switches unveiled an ambitious vision for "Smart Home 2.0" powered by AI robotics.
The star of SwitchBot's CES booth is Onero H1, billed as "the most accessible AI household robot." It's not as capable as LG's CLOiD, but it's designed to be affordable enough for mainstream adoption.
Key specifications:
Onero H1 can handle contact-intensive tasks like grasping, pushing, opening, and organizing. The robotic arms (called A1) will be available for pre-order on SwitchBot's website soon.
SwitchBot's Lock Vision Series is the world's first deadbolt smart lock with 3D structured-light facial recognition. Using over 2,000 infrared projection points, it creates precise 3D facial maps for millimeter-level biometric accuracy and near-instant unlocking.
The system works even with hats, glasses, or makeup, and includes 3D liveness detection to prevent spoofing with photos or videos. All biometric data is stored locally—nothing goes to the cloud.
The Lock Vision Pro adds palm-vein recognition as an alternative, using near-infrared sensing to read internal vascular patterns. This works even with slightly wet or dirty hands.
Both models support Matter-over-Wi-Fi for hub-free integration with Apple Home, and feature dual-battery backup systems for reliability.
SwitchBot AI MindClip is an 18-gram wearable voice recorder and "personal knowledge engine." It continuously captures meetings and conversations, then transforms them into:
Think of it as a second brain that remembers everything you hear and can answer questions about past discussions. It supports over 100 languages.
Why this matters: SwitchBot represents the democratization of smart home technology. Their products are typically a fraction of the price of competitors, making AI-powered home automation accessible to far more people. If Onero H1 ships at an affordable price point, it could bring humanoid robotics to mainstream consumers years earlier than expected.
Remember when Boston Dynamics robots were just doing backflips on YouTube? Those days are over. If CES 2026 had a "hold my beer" moment, it might probably be this: Boston Dynamics finally taking Atlas out of the lab and onto a public stage for the first time ever.
For years, Boston Dynamics has been the company that makes viral robot videos—backflips, parkour, dancing to "Do You Love Me." Impressive, sure, but always with the unspoken question: when will these actually do something useful?
That question just got answered. At CES 2026, Boston Dynamics finally took Atlas out of the lab and onto a stage; its first-ever public appearance. And the message was clear: this isn't a research project anymore. It's finally (20 years and counting...) an actual product.
"We've been working on humanoids for more than a decade," the team explained during the keynote. "The rapid advancements in AI over the past few years are the piece that we needed. That moment is finally here."
In addition to the live stage demonstration of the prototype, because this is Boston Dynamics we're talking about, the keynote also featured a K-pop dance number featuring their famous Spot robots.
The specs are wild. The production Atlas is genuinely impressive hardware:
Safety features include human detection and fenceless guarding—critical for robots working alongside people. The system integrates with existing industrial infrastructure via barcode scanners, RFID, and Boston Dynamics' Orbit software platform, which connects to MES, WMS, and other enterprise systems.
"Our new Atlas is the most production friendly robot we've ever designed," said Zack Jackowski, GM of Atlas. "This generation significantly reduces the amount of unique parts, and every component has been designed for compatibility with automotive supply chains."
This part is cool too: Once one Atlas learns a skill, it shares that knowledge with every other Atlas through their Orbit platform. Robot hivemind, basically. What could go wrong?
Perhaps the biggest news buried in the announcement: Boston Dynamics is partnering with Google DeepMind to integrate cutting-edge foundation models into Atlas.
This isn't just about better navigation or object recognition. Foundation models give Atlas genuine cognitive capabilities—the ability to understand instructions, reason about novel situations, and adapt to environments it's never seen before. It's the same AI revolution that's transforming language models and image generators, now applied to physical robots.
The partnership makes strategic sense for both sides. Boston Dynamics has world-class hardware and decades of robotics experience. Google DeepMind has some of the most advanced AI models on the planet. Together, they're building robots that can think as well as they move.
Hyundai Motor Group, Boston Dynamics' majority shareholder, isn't just buying Atlas robots—they're building them. The company announced a $26 billion investment in U.S. operations, including a new robotics factory capable of producing 30,000 Atlas robots per year starting in 2028.
First deployment: the Hyundai Motor Group Metaplant in Savannah, Georgia. Atlas will start with parts sequencing tasks in 2028, then graduate to repetitive motions, heavy loads, and complex operations by 2030.
Hyundai is also building what they call the Robotics Metaplant Application Center (RMAC)—essentially a data factory for training humanoid skills. Every task Atlas learns at the RMAC can be instantly replicated across the entire fleet through the Orbit platform. One robot learns; all robots know.
The supply chain is also coming together. Hyundai Mobis will manufacture the actuators—the joints that make robots move—creating a vertically integrated robotics production system.
The entire 2026 supply is already allocated to Hyundai and Google DeepMind, with additional customers coming in early 2027.
Atlas enters a suddenly crowded humanoid market. Tesla's Optimus has been grabbing headlines. Figure AI raised massive funding. Chinese companies like Unitree are shipping robots right now. And NVIDIA just released Isaac GR00T foundation models for humanoid development.
But Boston Dynamics has advantages the newcomers don't:
Why this matters: Tesla's Optimus has been grabbing headlines, but Boston Dynamics just flexed decades of robotics expertise combined with modern AI. The production version they showed isn't a demo—it's going into actual car factories. The "data factory" approach is smart. Hyundai isn't just deploying robots—they're building the world's most complete dataset for training humanoid skills in manufacturing. That flywheel could be hard to catch.
No CES is complete without new processors, and both Intel and AMD delivered major AI-focused announcements.
Intel's latest laptop chips are built on their 18A (essentially 2nm) process technology—a major achievement for the company that's been playing catch-up in manufacturing.
Key highlights:
The focus is on enabling AI PCs that can run sophisticated AI models locally, without relying on cloud connectivity. This matters for privacy-sensitive applications and situations where internet access is limited.
AMD's latest processors emphasize AI-driven tasks for everyday use:
AMD is also expected to announce FSR Redstone, their new AI-powered upscaling technology for gaming.
Qualcomm continues pushing Windows on ARM with the Snapdragon X2 Elite, featuring 80 TOPS of AI performance. The company also announced the Dragonwing IQ10 humanoid robotics platform—a full-stack AI architecture powering partnerships with Figure, Kuka, and others.
Why this matters: The amount of AI processing available in consumer devices has increased roughly 10x in the past two years. This enables a new category of AI applications that work entirely on your device—no cloud required, no latency, no privacy concerns about data leaving your computer.
The through-line connecting every major announcement at CES 2026 is the transition from AI that generates to AI that acts.
For the past few years, AI has been primarily about producing content—text, images, code, music. ChatGPT writes your emails. Midjourney creates your images. GitHub Copilot suggests your code.
CES 2026 showcased a different future: AI that perceives the physical world, reasons about what it sees, and takes action. Cars that explain their decisions. Robots that fold laundry. Refrigerators that track your eating habits. Locks that recognize your face.
This is a fundamental shift in what AI systems can do:
The companies that dominated CES 2026 share a common understanding: the next wave of AI won't just be about better models—it'll be about those models doing useful things in the real world.
Several key developments to watch over the coming year:
The AI revolution has a bottleneck most people don't think about: memory bandwidth. AI chips can only process data as fast as they can access it, and traditional memory designs weren't built for the parallel processing that AI requires.
SK hynix unveiled the world's first 16-layer HBM4 (High Bandwidth Memory) at CES 2026, featuring 48GB of capacity. High Bandwidth Memory stacks memory chips vertically and connects them with thousands of tiny wires, enabling data transfer rates that would be impossible with traditional memory layouts.
This isn't just incremental improvement—it's foundational infrastructure. Every major AI model, every robotics system, every autonomous vehicle relies on fast memory access. HBM4 removes a constraint that was beginning to limit what AI systems could do.
In one of CES's more unexpected announcements, LEGO revealed SMART Bricks—a system designed to bring AI-powered interactivity to physical play without screens.
The system includes:
It's essentially augmented reality without the glasses or phone—the intelligence is embedded directly in the toys. Imagine building a castle where the drawbridge actually recognizes when an "enemy" minifigure approaches and raises itself, or a spaceship that makes different sounds based on which pilot you place in the cockpit.
LEGO is betting that the next generation of kids will expect their physical toys to be as responsive as their digital games. SMART Bricks is their answer.
The Withings Body Scan 2 represents how AI is transforming personal health monitoring. This isn't just a bathroom scale—it's a comprehensive health assessment device that tracks over 60 biomarkers.
New capabilities include:
The AI component analyzes trends over time, flagging concerning patterns before they become serious health issues. Priced at $600 with a Q2 2026 launch (pending FDA clearance), it's positioned for health-conscious consumers who want clinical-grade monitoring at home.
The SwitchBot AI MindClip tackles a problem knowledge workers face daily: remembering everything from meetings and conversations.
This wearable ring records and transcribes meetings in over 100 languages, then uses AI to generate highlights, action items, and searchable transcripts. A button on the ring lets you mark important moments in real-time—tap it when someone commits to a deadline, and that moment gets flagged in the transcript.
Privacy concerns are addressed with local processing for sensitive audio and clear visual indicators when recording is active. But the value proposition is compelling: never again lose a crucial detail from a meeting because you were too busy taking notes to actually listen.
Lighting company Govee announced AI Lighting Bot 2.0 and DaySync, technologies that bring adaptive intelligence to home lighting.
AI Lighting Bot 2.0 uses natural language understanding to adjust lighting based on conversational commands. Instead of fiddling with apps and sliders, you can say "make it feel like a cozy evening" and the system interprets that into specific brightness, color temperature, and dynamic effects.
DaySync automatically adjusts lighting throughout the day to match natural light patterns—bright and cool in the morning to promote alertness, warm and dim in the evening to support sleep. The AI learns your preferences over time, building a personalized lighting profile.
New hardware includes the Floor Lamp 3, Ceiling Light Ultra, and Sky Ceiling Light—all designed to integrate with the AI system for whole-room atmospheric control.
One of the underappreciated stories from CES 2026 is how much the AI ecosystem has matured through partnerships. A few years ago, every company was building their own AI stack from scratch. Now, strategic partnerships are accelerating development across the industry.
NVIDIA announced deep integration with Siemens, embedding CUDA X, physical AI, agentic AI, Nemo, and Neotron into Siemens' industrial software platforms.
The implications are significant. Siemens software is used to design and run manufacturing plants around the world. By integrating NVIDIA's AI capabilities, those plants essentially become "gigantic robots"—with AI optimizing every aspect of production from scheduling to quality control to predictive maintenance.
Huang described it as "agentic chip designers and system designers working with us"—AI that doesn't just assist human engineers but actively participates in the design process.
The NVIDIA-Mercedes partnership produced the first commercial vehicle with chain-of-thought reasoning AI. But the partnership goes deeper than just hardware and software.
Mercedes contributed its safety-certified development processes—the German automaker knows how to build systems that regulators trust. NVIDIA contributed the AI models and simulation tools. The result is a vehicle that passed NCAP safety certification while incorporating cutting-edge AI capabilities.
This model of partnership—pairing AI expertise with domain expertise—is likely to define how AI reaches other safety-critical applications in healthcare, aviation, and infrastructure.
Academic partnerships matter too. Berkeley DeepDrive, the autonomous driving research center at UC Berkeley, praised NVIDIA's decision to open-source Alpamayo:
"The launch of the Alpamayo portfolio represents a major leap forward for the research community. NVIDIA's decision to make this openly available is transformative as its access and capabilities will enable us to train at unprecedented scale—giving us the flexibility and resources needed to push autonomous driving into the mainstream."
Open-source AI accelerates academic research, which feeds back into commercial products. This virtuous cycle is one reason AI is advancing so rapidly.
With all the excitement about Physical AI, CES 2026 also raised serious questions about safety that the industry will need to address.
The Mercedes-Benz CLA demonstrates one approach to AI safety: redundancy. The vehicle runs two complete autonomous driving stacks simultaneously:
A "policy and safety evaluator" constantly monitors both systems, deciding when to trust Alpamayo's reasoning and when to fall back to the more predictable classical stack. It's like having two pilots—one creative problem-solver and one by-the-book proceduralist.
"All safety systems should have diversity and redundancy," Huang emphasized during the keynote.
Google's announcement of Deep Think for the Gemini app came with an unusual disclosure: the company's safety evaluations found that the model "might have reached a 'critical capability level' for helping bad actors with dangerous biological, chemical, or nuclear information."
Google deployed the model anyway, with extra safety measures and usage monitoring. But the fact that they disclosed this concern publicly signals a maturation in how the industry thinks about AI safety. Transparency about risks, even when deploying products, builds trust more than pretending risks don't exist.
LG specifically addressed safety concerns with CLOiD's design. The wheeled base was chosen over legs partly for stability—a robot that can't fall over is inherently safer around children and pets. The low center of gravity prevents tipping if someone bumps into it.
But harder questions remain. What happens when a home robot misinterprets a command? What safeguards prevent a malfunctioning robot from causing injury? How do we ensure robots respect privacy in our homes?
These aren't hypothetical concerns—they're design challenges the industry is actively working on. The answers will determine whether home robots become as trusted as other home appliances.
If CES 2026 was the debut of Physical AI, what comes next?
CES 2026 was the coming-out party for Physical AI; technology that bridges the gap between silicon intelligence and real-world action.
For consumers, this means smarter cars, helpful home robots, and AI-powered appliances that actually understand how you live. For businesses, it means new opportunities to build on open-source foundations and integrate AI into physical products.
For everyone, it means the AI revolution isn't just about chatbots and image generators anymore. It's about machines that can see, think, and do.
The transformation won't happen overnight. Self-driving cars still face regulatory hurdles. Home robots need to prove their value proposition. Privacy concerns need to be addressed. But the direction is clear, the technology is ready, and the major players are committed.
Welcome to the Physical AI era. It's going to be quite a ride...
Grant Harvey is the Lead Writer of The Neuron, where he continues to lead the publication's daily coverage of AI news, tools, and trends.
Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.
Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved
Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.
