NVIDIA’S AUTO LEADER NAVIGATES THE FUTURE OF AUTOMATION

Today, I’m diving into a conversation with Xinzhou Wu, the head of automotive at Nvidia. With the AI boom making headlines, Nvidia has become one of the most valuable companies globally, largely due to the insatiable demand for its GPUs. However, Nvidia’s influence extends into the automotive sector, where it has been a key player for years, providing chips that power various vehicles. Xinzhou has played a pivotal role in developing a comprehensive autonomous driving system that automakers can easily integrate, already featured in newer Mercedes EVs.

I wanted to hear his thoughts on how the automotive industry is navigating the significant shift towards self-driving electric vehicles. While every carmaker and supplier touts this future, it feels more distant than ever as we approach 2026. The EV adoption rate in the U.S. is lagging, self-driving technology seems stuck on the final 20% of challenges, and vehicle prices continue to rise, putting pressure on consumers amid inflation and increasing energy costs.

Xinzhou highlights the remarkable progress in redefining the car itself, a concept known as the “software-defined vehicle.” This approach centralizes control into a few powerful computers rather than relying on numerous independent electronic control units (ECUs). If you’ve been following the automotive space, you’ve likely heard many manufacturers express the need to move away from ECUs, and Xinzhou believes that moment is finally upon us.

We also discussed the Chinese automotive industry, which has gained a competitive edge by building on EV architectures from the ground up, avoiding the transition challenges faced by legacy gas-powered manufacturers. Xinzhou’s background at a Chinese original equipment manufacturer (OEM) gives him unique insights into this dynamic.

Our conversation also touched on life at Nvidia, a company known for its unique culture and leadership under Jensen Huang. Xinzhou shared that his three years there have been a whirlwind of learning, especially as he competes for resources against the booming AI sector. His insights into what drives these resource allocation decisions, particularly when working with cost-sensitive automakers, were particularly intriguing.

Naturally, we had to discuss AI and how Nvidia’s approach to autonomy merges traditional methods with advanced reasoning models. The idea of an AI model communicating with itself to navigate driving scenarios is both fascinating and amusing.

And, of course, no discussion about electric vehicles or autonomy in the U.S. would be complete without mentioning Tesla. I asked Xinzhou directly whether Tesla’s full self-driving capabilities can deliver on Elon Musk’s promises without using lidar. His response is sure to spark debate.

Now, let’s jump into the conversation with Xinzhou Wu, head of automotive at Nvidia.

—

Xinzhou Wu, welcome to the show!

Thanks for having me.

I’m eager to discuss the current state of the automotive industry. It seems like we’re in a period of significant transformation, with the future of cars being redefined. The challenges of transitioning to electric vehicles and the complexities of the U.S.-China trade situation have made things messier than ever. Many automakers are pulling back, and your position at Nvidia gives you a broad perspective on these changes since you supply many major manufacturers worldwide.

So, where do you see the automotive industry on this winding road toward autonomy and electrification?

That’s a great question. I’ve been in the automotive sector for about 15 years, starting with Qualcomm, where I led their automotive team. The term “software-defined vehicle” has been around for a while, but with advancements in AI, we’re entering what I call the “AI-defined vehicle” era.

The auto industry has undergone rapid changes over the last decade. My experience at a Chinese OEM, where I led their autonomous driving team, has given me a front-row seat to this evolution. Cars have transitioned from primarily mechanical and electrical systems to platforms that can be upgraded through over-the-air software updates. Now, with generative AI, we’re using AI to rewrite much of the vehicle software, accelerating development and redefining what a vehicle is.

Let’s clarify some terms. “Software-defined vehicle” is a bit vague. The idea is to eliminate the numerous ECUs currently controlling various systems and centralize them into one or two powerful computing units. Tesla is well-known for this, and Rivian has made significant investments in this direction. Other legacy automakers have attempted similar approaches, but many have struggled. Do you think the industry will ultimately transition to software-defined vehicles, or will legacy automakers remain stuck in their ways?

Absolutely. I’ve witnessed the transformation in China from 2018 to 2023, where both new and legacy automakers have had to adapt to a centralized computing architecture to stay competitive. Globally, we’re collaborating with partners like Mercedes to implement this essential computer-based architecture across their vehicles. While some companies will be slower than others, I’m confident the industry is moving in this direction.

I’m curious about your experience at XPeng, a Chinese automaker. It seems they had a unique advantage by starting fresh with EV architectures, unlike legacy manufacturers that had to transition from gas vehicles. Was that your experience?

That’s part of it. Chinese automakers have fewer legacy burdens, which is an advantage. However, even established global players in China have had to adapt quickly to keep pace. The software-defined vehicle concept has been around for a while, and Tesla has taken it to full production. I believe all OEMs will eventually follow suit because it’s essential for survival. Autonomy will become a necessity, and the only way to achieve that is through the architecture I mentioned, which allows for software upgrades without relying on numerous ECUs.

The path to this future seems bumpier than expected. Factors like changing political climates, fluctuating EV sales, and the competitive landscape with Chinese automakers complicate matters. From your perspective as a supplier, what challenges have made this transition harder?

You’ve touched on several key points. The auto industry is complex, involving extensive supply chains and a large workforce. Changes in architecture require long-term commitments, as vehicles need support for 10-15 years. Nvidia, as a supplier, also commits to supporting our technology for that duration, which can slow down progress.

Additionally, the rapid pace of technological change means companies need to adapt their talent pools to keep up. Nvidia can help bridge this gap by providing the necessary technology and support for autonomous vehicles. While the industry may not move at the same speed, my role is to help everyone reach the goal of autonomy as quickly as possible.

Let’s talk about your role at Nvidia. The company has been on a remarkable run with AI, and every GPU produced is in high demand. How large is the automotive team at Nvidia?

We have a sizable automotive team, numbering in the thousands. We’re working on a comprehensive platform that includes hardware, software, models, and infrastructure. Nvidia also benefits from collaboration with other teams, leveraging their work on foundational models.

How is your team structured? Is it organized around hardware, software, and models?

Yes, we have distinct teams for product, strategy, and engineering. We also have a critical mapping team that focuses on high-level autonomy paths and data infrastructure.

Is your team primarily based in the U.S., or is it global?

Most of our team is in the U.S., but we also have a presence in China and Europe to support our global product and platform.

You mentioned relying on foundational models developed by Nvidia. How does your team fit into the broader AI strategy? Are you integrated or more siloed?

Great question. Nvidia has centralized hardware and software teams responsible for the overall roadmap. The automotive team operates as a separate organization focused on building the automotive platform while leveraging the work of the hardware and software teams. We also have a model team that collaborates across various projects.

Given the high demand for GPUs, do you find yourself competing for resources against the booming AI business?

Yes, that’s a reality. Even Nvidia has limited GPU supply. We prioritize our internal resources and often collaborate with colleagues to allocate compute power for different projects. Sometimes, we need Jensen’s input to make those decisions.

What does that resource allocation debate look like? Is it based on ROI, market size, or other factors?

It’s a combination of all those factors. Revenue is crucial, but Nvidia also seeks strategic opportunities that could lead to significant future growth. We balance immediate revenue needs with long-term potential.

Nvidia operates uniquely under Jensen’s leadership. What’s it like working in that environment?

It’s a unique experience. Jensen engages with different groups for technical strategy and product reviews, which provides valuable insights into his strategic thinking. Learning from his technical depth has been inspiring.

As you discuss the potential for autonomy, what does the revenue model look like? Are you selling chips and software to automakers, or is there a subscription model for consumers?

We believe that everything that moves will eventually be autonomous. Currently, we drive 13 trillion miles annually, with autonomous miles being a tiny fraction. Nvidia aims to provide foundational technology, from chips to operating systems, to help the ecosystem develop. We envision a revenue model where we earn a percentage from every mile driven autonomously.

So, revenue per mile is the key metric. How do you envision that working for consumers? Will it be through subscriptions or robotaxis?

Both models will coexist. Robotaxis are already successful in various markets, and we anticipate more growth in that area. However, many consumers will still prefer personal vehicles, similar to how people choose to own homes over renting.

Legacy automakers have realized they’ve become more like insurance and financing companies, losing control over car design. It seems Nvidia has an opportunity to become a primary supplier as they seek to regain that control. Is that dynamic shifting in your favor?

Nvidia’s business model is open, allowing OEMs to choose the level of collaboration they want. Some manufacturers prefer to build their own inference systems, while others seek a more turnkey solution. We adapt our approach based on each OEM’s capabilities, working closely with them to ensure our technology integrates seamlessly.

You’ve mentioned training models and synthetic data. With companies like Waymo and Tesla leading in autonomous miles, how does Nvidia position itself as a third-party provider to help automakers catch up?

That’s a compelling point for OEMs to engage with Nvidia. Our Hyperion ecosystem allows data sharing among partners, enabling us to accumulate millions of hours of driving data. This collective effort helps us build robust models and close the data gap.

You’re using synthetic data to enhance training. How does that work, and why would automakers participate in data sharing?

Data collection is costly, and sharing data through our Drive platform can save OEMs significant resources. By collaborating, they can leverage our extensive data collection efforts without incurring the same expenses.

Nvidia’s approach seems to focus on creating a smart vehicle capable of handling various scenarios without relying solely on mapped roads. Is that the future direction?

Yes, we’re moving towards a model that can operate effectively without extensive mapping. While more data is essential for training, we’re also leveraging foundation models to enhance reasoning capabilities, which will be crucial for achieving higher levels of autonomy.

Safety is paramount in autonomous driving. How does Nvidia ensure the safety of its models, especially with the complexities of AI reasoning?

Safety is critical, and we adhere to rigorous development and validation protocols. We have a redundant stack for our L2++ and ADAS functions, ensuring safety at every level. Our classical stack acts as a safety guardrail, verifying trajectories generated by the end-to-end model.

Is the reasoning process of the model similar to how humans think? Does it operate in a language-like manner?

Yes, our next-generation model will incorporate language reasoning, allowing it to communicate its thought process while driving. This multi-modal approach combines visual signals with language-based reasoning.

What’s the latency like for this reasoning process? Is it manageable for real-time driving?

Latency is a key consideration. Our model aims to keep latency under control, with current systems operating within 100 milliseconds. The reasoning component is just one part of the overall system, which relies on visual input for immediate reactions.

Is all this processing done locally in the car?

No, validation occurs offline, but the safety guardrails operate in real-time within the vehicle. We compare outputs from both the classical stack and the end-to-end model to ensure safe trajectories.

Do autonomous vehicles require constant connectivity, or can they operate independently?

While some connectivity is necessary for navigation and mapping, the vehicle must be capable of driving autonomously without relying on constant connectivity. Level 4 systems must have sensor redundancy to ensure safe operation even in the absence of GPS.

What happens if connectivity is lost during Level 4 autonomy?

Level 4 systems must be able to handle connectivity loss safely. The vehicle should be able to navigate to a safe location if it loses GPS or other critical inputs.

As vehicles become more complex and expensive, how do you see the cost of autonomy playing out? Will consumers be willing to pay for these advanced features?

Building autonomous vehicles requires significant hardware, but costs are decreasing as technology matures. Prices for sensors and computing power are dropping, making it more feasible to integrate advanced features into vehicles.

Do you face challenges in securing fabrication capacity for chips amid rising demand?

Yes, securing fabrication capacity is a challenge, especially with the intense demand for chips. We’re committed to investing in the future of autonomous vehicles, balancing our needs with the broader market.

As we wrap up, what should we expect from Nvidia in the coming years?

We’re rolling out our technology across all Mercedes vehicles and collaborating with partners like Uber to launch L4 services. Our goal is to build a robust ecosystem with a wide range of OEMs, and we’re excited about the future developments on the horizon.

2026-07-13T20:42:49Z