Tesla Is Snatching Apple’s Stars to Make Itself the New Apple

If you don’t follow the ins and outs of Silicon Valley personnel moves, you might have missed the news. Even if you saw it, it may not have made much sense. Chris Lattner is leaving Apple for Tesla? Chris who?

Lattner doesn’t enjoy the name recognition of a Tim Cook or a Jony Ive. But he’s a rock star among software engineers. As the guy who built Swift, Apple’s iPhone-centric programming language, he’s one of those coders that other coders put on a pedestal. He personifies Silicon Valley’s relentless push toward technology ca
Now, he’s moving on, becoming the head of software engineering for Autopilot, the technology that’s transforming Tesla’s electric vehicles into autonomous vehicles. Apple’s innovation machine is losing another key cog to a company that has lured so many others away from the House That Jobs Built. And that provides an obvious storyline for the tech press and so many other Silicon Valley watchers: Tesla is the new Apple.
In many ways, the story is true. Apple vice president of Mac engineering Doug Field, director of alloy engineering Rich Heley, and MacBook Air engineer Matt Casebolt are among those who’ve left Cupertino for Tesla. But the truth goes deeper than that. Sure, Tesla is snagging some of Apple’s mojo, becoming a defining symbol of Silicon Valley innovation. “Apple is stuck in the world of phones and watches,” the argument goes, “while Tesla is well down the path to self-driving cars, for Jobs’ sake!” Look beyond the A-list talent and sexy public image, though, and you’ll see that Tesla is mimicking Apple (and Google and Amazon and Facebook) in a more meaningful way. Lattner’s arrival is just the latest evidence of this. Like those other tech giants, Tesla is not just building new products. It’s building them from entirely new parts, remaking them from top to bottom. Apple did this with phones. Now, Tesla is doing it with cars—and with computers, too.

Extreme Engineering

Google didn’t just build a new search engine. It built a new kind of global computer capable of running that search engine at unprecedented scale, fashioning everything from the software to the servers to the network switches to the data centers. That’s what made Google so successful: It could serve far more people, far more quickly than anyone else. Facebook and Amazon soon followed suit.
In similar fashion, Apple didn’t just build the iPhone. It built so many of the individual parts inside the iPhone, including the microprocessor at the heart of this iconic device. That allowed the company to not only build a phone no one else could, but do it with unprecedented speed and efficiency.
Tesla knows that autonomous vehicles require the same kind of extreme engineering. Tesla is not just building a car, it’s building an entirely new kind of computer. Today, computers are designed to send data into the world. Autonomous vehicles require computers that can draw data from the world and use it to understand what’s happening around it. That is a very different kind of computer, and it hasn’t yet been built—not to the degree that anyone can be sure it will work with unerring accuracy and safety.

Fundamentals

Tesla hired Lattner because he has a history of building enormously complex, enormously successful software projects. As his PhD advisor, Vikram Adve, points out, Swift is just one example of this. Before that, Lattner built two other software projects on a similar scale: LLVM and Clang. Never heard of those? Then you’re not a software engineer. These are foundational tools at both Apple and Google, underpinning everything that happens on the iPhone and on every Google online service. “Chris is really good at managing and running a large software project, with very high quality, attracting really good programmers, and getting results,” says Adve, a computer science professor at the University of Illinois at Urbana-Champaign.
Lattner did not respond to my request for an interview. But his previous work is telling not just because it was so successful, but because it was foundational. LLVM and Clang are software tools that engineers use to build other software—and build it at the most fundamental level, the level where it talks to the hardware. Tesla needs people like Lattner in part because it has split with Mobileye, the Israeli company that makes the image recognition system currently underpinning Autopilot. It must build a whole new system, apparently around GPU chips from nVidia, the graphics processing chips that now do double-duty at the heart of many artificially intelligent systems, And Lattner is the kind of engineer who can lead the development of this system. Not to mention: the foundational software that drives nVidia GPUs is based on LLVM.
But it’s not just the Lattner hire that shows how deep Tesla could go in building its own technology. While Lattner will run the software side, the head of hardware engineering for Autopilot is Jim Keller, the former Apple chip designer who helped build many of the microprocessors that went into the world’s iPhones. (Several of Keller’s old colleagues are now also working for Tesla.) Tesla has the key talent needed to build its own AI technology from top to bottom. And it may have to. Mobileye is off the table, and it’s still unclear whether GPUs are the future of autonomous vehicles. Mobileye apparently dumped Tesla for moving too fast down the road to self-driving cars, but Tesla has no intention of slowing down.
No one outside of Tesla knows whether Tesla is building an AI chip. But hiring Keller suggests that it will. And if it does, it will need people like Lattner to build the low-level software that can drive those chips. Fundamentally, LLVM is a tool for building software that could run on any piece of hardware. It lets you readily target the chips you want to target, says Todd Mostak, the CEO of MapD, a company that uses LLVM to build database software for multiple processors.
The world is already headed toward a new kind of AI chip. Last year, Google revealed that it built one called the Tensor Processing Unit, or TPU, and commercial chip makers like nVidia and Intel are building dedicate AI chips as well. Such chips are needed to serve the demands of the neural networks and other techniques behind so many of the AI-driven services moving into the market, from image and speech recognition to the nerve centers that drive autonomous cars.
Tesla could get this technology elsewhere, but its ambition is often far more extreme. After all, it’s already building its own batteries in its own factory. Building your own stuff lets you do things no one else can do, at a scale and cost no one else can achieve. That’s where the real advantage comes. So, yes, Tesla hired Chris Lattner because it wants to be more like Apple. But it also hired him because it wants to be something very different.