Hey, how's it going? — Episode 444

Hey, how's it going?

I've been digging through today's Tesla updates, and it's one of those days where the picture is genuinely mixed. Let's catch up on what's actually moving the needle.

The European market threw Tesla a bit of good news today. Sales across Europe were up 11 percent in March, with both Tesla and the Chinese brands driving a lot of that surge. It shows that in some regions the EV story is still growing, even if the pace feels uneven depending on where you look.

At the same time, California is telling a different story. Tesla sales have dropped there as hybrids gain ground and overall EV demand cools off. It's a reminder that incentives, fuel prices, and local preferences can shift the ground pretty quickly. Tesla's strength has always been adapting to these regional differences, but the California dip is worth watching because that state's been a bellwether for years.

On the competitive front, Ford's CEO was pretty blunt. He said Tesla lacks recent product updates and that the real rival to watch is BYD. Whether you agree or not, it highlights how the conversation has moved from "Tesla versus legacy automakers" to "Tesla versus really efficient Chinese manufacturers." Different battlegrounds, different strengths.

On the supply side, Samsung is reportedly tripling its DRAM supply to Tesla amid rising demand. This lines up with the heavy compute needs for everything from in-car systems to the training clusters. It's a quiet but important signal that the hardware foundation for the AI side of the business keeps expanding.

Tesla is also rolling out a new voice model in its vehicles. The articles don't give every technical detail, but anything that makes the in-car assistant feel more natural tends to improve the everyday ownership experience. These software improvements are the kind of thing owners notice immediately.

There's a new piece out on the Tesla Model Y L Premium that walks through five key highlights. From what I can tell it's tailored for certain markets, likely focusing on range, build quality, and the features that make the Y such a strong seller globally. Nice to see continued refinement on the vehicle that's carrying a big chunk of their volume.

One more note on the product side. The Tesla Model S is currently leading NHTSA complaints per 10,000 vehicles, but the context in the report is important: most of the top ten vehicles on that list are still gas-powered. It doesn't erase customer frustrations, but it does put the numbers in perspective against the broader industry.

Finally, on the AI front, SpaceX just secured an option to acquire Cursor AI for $60 billion ahead of its IPO. While it's technically a SpaceX move, the overlap in talent, ambition, and Elon's focus on coding tools that accelerate development is hard to ignore. These big bets show where the real arms race is heading.

That's the mix today — some regional wins, some clear headwinds, and steady progress on the tech that will matter longer term.

Let's talk about how battery chemistry has evolved and why different parts of the world are making very different bets.

Early lithium-ion cells were mostly variations on lithium cobalt oxide. They gave decent energy density but relied on expensive cobalt, which created both cost and ethical issues around mining. As EVs moved from proof-of-concept to mass market, the industry needed to balance three things: how far the car could go on a charge, how much the pack would cost, and how many cycles it would last before noticeable degradation.

That search led to NMC chemistries — nickel manganese cobalt. By tweaking the ratios (NMC 811 has a lot more nickel), manufacturers could push energy density higher, giving longer range in the same package weight. Tesla used these in early Model 3 and Y packs, especially in vehicles sold in markets that prioritized highway range. The trade-off is that nickel-heavy cells can run hotter and sometimes show slightly faster calendar aging if not carefully managed.

Meanwhile, China went all-in on LFP — lithium iron phosphate. The chemistry is cheaper, inherently more stable, and far less dependent on scarce metals. Range per kilogram is lower, so vehicles need bigger packs, but the cost advantage and longevity are compelling. Chinese manufacturers could hit aggressive price points while promising 3,000+ cycles with minimal degradation. That approach helped them dominate domestic sales and export markets that care more about total ownership cost than matching a 400-mile EPA number.

Europe sits somewhere in the middle. Strict CO2 fleet rules and a cultural emphasis on efficiency pushed many brands toward high-nickel NMC for smaller packs that still deliver decent range. At the same time, European regulators are increasingly concerned about supply-chain resilience and raw-material ethics, so there's growing interest in LFP and even sodium-ion for entry-level cars. The result is a patchwork: German premium brands stick with high-energy cells for performance image, while volume brands and Chinese transplants bring LFP options to hit price targets.

Tesla's 4680 format is its own bet. By going tabless and much larger in diameter, the cell reduces manufacturing cost per kilowatt-hour and improves thermal characteristics. Pair that with a chemistry that can be either high-nickel or moving toward more manganese-rich formulations, and you start to see the ambition: a cell that is cheaper to make, safer, and still delivers competitive energy density. The structural pack that 4680 enables also cuts weight and parts count, which improves vehicle efficiency — something that matters whether you're in California, China, or Germany.

The surprising international contrast is how perspectives on "good enough" differ. In North America, many consumers still chase the biggest range number on the window sticker. In China, the conversation is often about cost per kilometre over the vehicle's entire life. European buyers seem to land between the two — they want meaningful range but also expect the car to hold its value and degrade slowly in varied climates.

What this means is there won't be one winning chemistry. Different markets reward different trade-offs. Tesla's vertical integration lets it adjust the recipe factory by factory — using LFP in Shanghai-built cars for certain markets while pushing 4680 in the US and Berlin. That flexibility is becoming a genuine competitive advantage as the rest of the industry wrestles with which bet to double down on.

The evolution isn't linear; it's regional and economic. The chemistry that wins in one country may lose in another, which is why watching how Tesla deploys its different cell formats around the world tells us more about the real strategy than any single headline.

(Word count: 612)

If you're new to Tesla or just starting to pay attention as an investor, the Full Self-Driving story can sound like a lot of jargon. Let's walk through it simply, the way I'd explain it to a friend who's never read a single technical paper.

Start with the hardware journey. Early cars had Autopilot Hardware 2, which was basically a solid foundation but limited in computing power. HW3 stepped things up with custom chips designed specifically for neural networks. Then came HW4, with more cameras, better resolution, and significantly more processing muscle. HW5 is on the horizon, but the important pattern is that Tesla keeps iterating the "eyes and brain" of the car while promising that earlier hardware can still improve through software.

The biggest philosophical shift happened when Tesla moved to vision-only. Most companies were fusing radar, lidar, ultrasonic sensors — a whole buffet of different technologies. Tesla decided that because humans drive using primarily vision, a well-trained neural net should be able to do the same. They removed radar from new cars and doubled down on eight cameras that see in every direction. To a new listener this can sound risky, but the bet is that software intelligence beats adding more imperfect sensors that can disagree with each other in rain, snow, or bright sunlight.

Inside that vision system lives something called the occupancy network. Think of it as the car building a real-time 3D model of the world around it. Instead of just detecting a car or a pedestrian as a flat bounding box, the network figures out the actual space that object occupies and how it might move. It's like the difference between seeing a photo and understanding the physical scene.

Then comes the planning layer, which now uses transformer models — the same family of AI tech behind tools like ChatGPT. These transformers are excellent at paying attention to the most relevant parts of a complex scene. Instead of following a long list of hand-coded rules ("if a pedestrian is within X metres, do Y"), the system learns from millions of miles of real driving data what usually happens next and chooses the safest, most natural action.

So where does Tesla stand compared with Waymo and Cruise? Those companies use a lot more sensors, including lidar, and operate in limited, mapped cities with safety drivers or remote operators ready to step in. Their approach is more conservative and geofenced. Tesla's bet is different: a pure-vision system that improves everywhere at once because every customer car is collecting data. No geofence. The car has to handle new cities, construction zones, and weird weather without ever having been explicitly programmed for them.

For a new investor, the practical question is simple: which bet scales? The sensor-heavy approach can work beautifully in a few cities, but rolling it out globally is expensive and slow. The vision-plus-data approach is harder at first but, if the neural nets keep improving, becomes cheaper and more universal over time.

Tesla still has work to do — regulatory approval, handling edge cases, and proving reliability in the real world. But the architecture is built on the idea that the best way to solve driving is to copy the most successful driver on the planet (the collective human fleet) and train a neural net on an ever-growing pool of real-world examples.

It's an engineering philosophy as much as a product roadmap: use simple sensors, massive real-world data, and powerful learning algorithms rather than complexity at the hardware level. Whether that bet pays off is still unfolding, but understanding the architecture helps you see why Tesla keeps talking about data advantage and why every mile driven by their customers matters.

(Word count: 598)

Let me know what you think — always appreciate your take. Talk soon.