The biggest news out of CES wasn’t a car, but a brain. Jensen Huang, CEO of NVIDIA, took to the stage to announce Alpamayo, a new autonomous vehicle platform launching later this year.
The first vehicle to use NVIDIA’s new autonomous platform will be the 2026 Mercedes-Benz CLA. Huang described it as a breakthrough in artificial intelligence. It isn’t a system that merely reacts to events around it - but one that thinks ahead and reasons.
"It's trained end-to-end. Literally from camera in to actuation out," Huang told the audience at CES. "It reasons what action it is about to take, the reason, and the trajectory."
Here’s a look at NVIDIA’s autonomous system in action:
The promise of Alpamayo is simple: a Vision-Language-Action (VLA) model that can explain its decisions. It offers reasoning traces (similar to the thought traces produced by reasoning LLMs), showing its work and explaining why it took a certain action. NVIDIA also claims that this transparency helps address the black-box problem that has plagued autonomous driving systems.
But while the tech world applauded, the team at Tesla, which has been running end-to-end neural networks on millions of cars for years, offered a stark reality check.
I’m not losing any sleep about this.
— Elon Musk (@elonmusk) January 6, 2026
And I genuinely hope they succeed.
Exactly What Tesla is Doing
Elon Musk’s reaction to the announcement was a mix of validation and warning.
He went on to mention that NVIDIA’s revolutionary approach to end-to-end models is exactly what Tesla began doing in 2023 with FSD V12.
The shift to end-to-end learning, where the car learns to drive by watching videos rather than following hard-coded rules, is a transition Tesla completed years ago. For the Tesla team, NVIDIA’s arrival at this architecture isn’t a leapfrog moment; it’s an admission that Tesla’s path was the correct one all along.
Well that’s just exactly what Tesla is doing 😂
— Elon Musk (@elonmusk) January 6, 2026
What they will find is that it’s easy to get to 99% and then super hard to solve the long tail of the distribution.
The March of 9’s
The core of the commentary isn’t on how AI works, but rather how hard it is to perfect.
Elon noted that NVIDIA and other OEMs that will rely on the Alpamayo stack will learn a harsh lesson quickly. Getting 99% of the way there is easy, but the last 1% is the hardest portion to solve. Ashok Elluswamy, Tesla’s Director of AI, also commented here, explaining the Long Tail Trap.
The final 1% is the bizarre, one-in-a-million scenarios that take years to solve. It’s the person in a chicken suit crossing the road; a roundabout where traffic flows in three different directions; or a sign partially covered in snow saying 5 instead of 50.
Ashok’s comment hints at the brutal grind that Tesla has endured over the last four years. Millions of miles of data are discarded every day to curate the rare edge cases needed to train neural networks to handle the real world. NVIDIA, by contrast (and similarity), is attempting to solve this gap with simulation tools and a much smaller dataset from partners.
The Hardware Lag
Beyond the software challenges, there’s one additional structural disadvantage for NVIDIA and the companies that will rely on it.
NVIDIA doesn’t build cars; it sells chips and software platforms to companies like Mercedes-Benz. Elon pointed out that even if Alpamayo is perfect, getting it onto the road at scale would be a logistical nightmare.
You’re right.
— Elon Musk (@elonmusk) January 6, 2026
The actual time from when FSD sort of works to where it is much safer than a human is several years.
The legacy car companies won’t design the cameras and AI computers into their cars at scale until several years after that.
So this is maybe a competitive…
While NVIDIA launches Alpamayo on the Mercedes CLA later this year, it will likely be a low-volume rollout. Tesla, meanwhile, has millions of vehicles on the road today collecting the long tail data necessary to train that elusive last 1%.
Welcome to the Grind
NVIDIA’s announcement is a massive validation of Tesla’s end-to-end neural network approach. The fact that the world’s leading AI chipmaker has adopted the same architecture as Tesla proves the industry is beginning to consolidate around a single solution.
But as Elon and Ashok point out, the architecture is just the starting line. The race will be won in the data, finding the rare edge cases that cause reasoning to fail. NVIDIA has built a brilliant engine; now they just have to drive it through the same chaotic and messy world that Tesla has been proving itself in for a decade.
Mutual Respect
Of course, Jensen Huang took it in stride, praising Tesla’s approach and tech stack as the most advanced available today.
Nvidia CEO Jensen Huang in new interview today on @Tesla’s FSD:
— Sawyer Merritt (@SawyerMerritt) January 6, 2026
“I think the Tesla stack is the most advanced autonomous vehicle stack in the world. I’m fairly certain they were already using end-to-end AI. Whether their AI did reasoning or not in somewhat secondary to that… pic.twitter.com/pYuPYdwW3h
