Nvidia CES 2026 keynote LIVE — All the biggest announcements from Jensen Huang and what to expect from GeForce On

Welcome to the live blog!

I'm here in the front row after a chaotically huge queue to get into what is easily the biggest event of CES 2026 — the Nvidia keynote.

Let me take you through what you can expect to see!

PREDICTION: The main event will be all about AI and robotics

Nvidia is a $5T company, and a lot of its huge value comes from being the picks and shovels company in the AI gold race. So I anticipate the party continues in being told about the future of AI and Robotics — maybe more company partnerships and new data center deployments.

PREDICTION: Self-driving takes another step forward

Nvidia Cosmos has been shown a lot of times over the past Nvidia GTC events, so I anticipate you'll see much more about this tech — maybe some car company updates?

Gaming will probably be left out of this keynote

Kinda unsurprising (given RTX 5090 got all of about 12 minutes of last year's show). Don't expect Jensen to talk about RTX PC gaming during this keynote. AI and robotics will probably be the key focus here.

Plus, GeForce is getting its own update at 9pm PT! So check back in with this live blog and I'll show you everything PC gaming-wise!

GeForce On Community Update | CES 2026 - YouTube

Watch On

The pre-show is over, and final checks are being made! It's almost time to start.

It's a packed house here in Vegas! Even all the standing room has been taken at the back... It must be all the hype to see how shiny Jensen Huang's leather jacket is this year.

Everybody is sat and we're ready to roll! Jensen's opting for a fashionably late entrance here.

Happy to be here, but disappointed in the audience's football knowledge

Coming from Nottingham, you best believe there's a particular team I shout for. Not only do people not know Forest, they didn't even think it wasa football shirt. Ball knowledge is woeful here — so I'm going to bore my neighbours with a full story of the team's legendary Premier League promotion in 2022.

Stock price check!

Things are smooth sailing for Nvidia right now. Normally there's a bit of a dip before Team Green events, but things are remaining steady. There must be a lot of excitement for what's to come

Annd we're off!

Starting with gameplay footage from Virtua Fighter and more (swear someone from Nvidia threw that in because I said don't expect gaming), we're seeing impressive use cases.

Fit check - the jacket is shiny!

Shiny jacket = a happy Jensen with a lot to announced. But on a serious note, we're diving headfirst into the success of Nvidia in scaling AI, making it Agentic, learning laws of nature and more.

AI models have expanded beyond simple chatbots with thinking models and agentic AI into systems that can be present everywhere you are.

"Wherever the universe has information, we can teach a large language model to understand its representation and turn it into AI," Huang commented.

But most important in Nvidia's mind is the drive for open models reaching the frontier — about six months behind but catching up ever since DeepSeek R1 shocked the world last year.

(If you're curious, I'm sat down over on the bottom right)

Nvidia's also hitting frontier AI models too

This comes off the back of Team Green building billions of dollars of AI supercomputers (DGX Cloud), this has led to Nvidia building a huge model of open ecosystems for every purpose across biomedical, Agentic (Nemotron) Cosmos physical AI, Gr00t for Robotics and for autonomous vehicles to boot.

As you can see, Nvidia leads the way in number of open models.

Leading the way on a lot of AI leaderboards

Nvidia's open models are catching up to the frontier and topping leaderboards in OCR recognition, PDF reading, natual language searching and more. On top of that, its agents are multi-model, multi-model and works hybridly across cloud compute models.

Simply put, Nvidia's made AI agents that are plug-and-play into other applications.

Nvidia Blueprints: the way to make and train your own AI model

Running locally on your own devices or on cloud compute models, Nvidia blueprints can be used to easily create AI applications. While Nvidia is clearly aiming for B2B sales with this (Crowdstrike, Netapp, etc), this model using Hugging Face to make a robot multi-modally aware and have a memory is an adorable application for the home!

Time for physical AI!

Full self-driving and robotics is something that Nvidia has aimed for over 8 years, and with Cosmos Ai training on the physical world, the Thor inference chip for Gr00t robotics, and RTX Pro simulation training AlpaMayo, this triple threat in the omniverse is the key to physical AI.

"The ChatGPT moment for Physical AI is almost here"

The key to it is synthetic data — pretrained on data to create 3D simulations and continue learning. This synthetic data can be created from images, text descriptions or telemetry video.

Rather than relying on very slowly training physical AI in the real world, the computational power of a GPU can render hundreds of thousands of scenarios to train robots and self-driving cars so much faster.

BREAKING: Nvidia just announced an open reasoning AI model for autonomous vehicles

Introducing Alpamayo — this is a breakthrough open source model for self-driving cars. Instead of being trained on data and driving reactively, this model is able to use reasoning at lighning quick times to be proactive.

As you may know, reasoning models usually generate the best answer for being more well thought-out — this coming to self-driving is significant, and it was just shown off with a driving demo across San Francisco.

And it's coming to the 2025 Mecedes Benz CLA

It's been rated the world's safest car by the automotive bodies. It's heading to manufacturing, and launching in the US in Q1! The Nvidia car is real after eight long years!!

Time for Robotics!

We have company! BDX droids from Star Wars have just enter from stage left, and they're entirely autonomous — trained on Nvidia Cosmos and following everything Jensen has been saying on stage.

He's having a rather fun conversation with the Droids!

Nvidia is joining up with Siemens and their manufacturing business is going to get a serious upgrade with Nvidia's physical AI — training robotics via synthetic data from digital twins of factories.

Vera Rubin is going to power future AI data centers, but can it run Crisis? Probably

The Rubin GPU is here for data centers of the future, and it's packing up to 5X increased performance over Blackwell, while only having 1.6x more transistors. And it doubles with extreme co-design with two separate GPU chips on the die.

And you know what happened to Blackwell? Went to a data center first, then consumer GPUs around a year or two after, so could this new architecture be in the RTX 60 series?

Fun big numbers about the Vera Rubin rack. It's 2.5 tonnes, has over 2 miles of copper cabling, can be specced with up 54TB of LPDDR5X RAM, and packs 220 trillion transistors.

In terms of how much data it can move, about 2x of the word's internet data at high speeds in Jensen's own words.

WTF is Silicon Photonics?

Jensen is talking about the new Spectrum-X chip to reinvent networking for these data center racks. Now, we've seen the end result of it with silicon photonics. Rather than using more traditional connectivity, silicon photonics uses lasers to transmit data at insanely fast speeds.

This is being made at TSMC using this newmanufacturing method. Really fascinating!

Context memory — an SSD for your SSD

To stop bottlenecks of context data, there is now context memory storage — allowing for these server racks to quickly find regularly used data. This may sound like RAM, but it's not! This is "Context Memory"

And the end result? Faster and more power efficient (lighter on your wallet too)

Takes a shorter amount of time to train an AI model on a 1/4th fewer GPUs, and costs less per AI token generation — dramatically less.

And that's a wrap!

Recap coming soon! And we'll start the countdown to the GeForce ON news soon

All eyes on GeForce ON

GeForce On Community Update | CES 2026 - YouTube

Watch On

Like Jensen said himself, GeForce and AI come hand-in-hand. We've heard all about the data center efforts. Now its time to figure out the gaming side of it. At 9pm PT, Nvidia is set to make some announcements.

Brace yourself for no new GeForce RTX GPUs

Looks like we shouldn't expect any new GeForce GPUs after all, as the official Nvidia Twitter account just told us so.