Tesla (NASDAQ:TSLA) has announced that going forward all Tesla cars will be equipped with sufficient hardware for fully autonomous driving once the software is fully matured. This "second generation" Autopilot system will probably be based on Nvidia's (NASDAQ:NVDA) Drive PX 2 hardware and its DriveWorks software for self-driving vehicles. Tesla critics decry the announcement as a mere publicity stunt, and whether the system will lead to a truly autonomous vehicle remains to be seen. But the system will certainly lead to a much more capable and safe Autopilot than the previous one based on Mobileye (NYSE:MBLY) hardware.
Game Over, Mobileye
For those who have been following my articles on the blooming partnership between Tesla and Nvidia, the announcement on October 19 must have seemed anti-climactic. I first speculated on September 22 that Nvidia had come between Tesla and Mobileye, leading to the termination by Mobileye of its supplier agreement with Tesla back in July.
The report in Electrek provided even more confirmation, but the formal announcement by Tesla has value in laying to rest any remaining doubt. It also provides some important clues as to the nature of the breakup with Mobileye and the Nvidia partnership.
One of the questions outstanding from the Mobileye breakup was how quickly Tesla would have to start replacing Mobileye hardware with something else. We now know the answer: about three months. Recognizing that Tesla might be under a time crunch, I wrote:
Both companies [Tesla and Nvidia] realize what's at stake, both realize that they have to get this right on the first try. A gap in availability of Autopilot in future production vehicles would be far preferable to a release of an inferior system.
Tesla could probably mitigate any Autopilot availability gap by including the Nvidia hardware during manufacturing, and providing an over-the-air software installation once the software development effort was completed.
And that's exactly what has happened. The short turnaround in replacing the Mobileye hardware lends credence to the reported accusation of a Tesla spokesperson that Mobileye "attempted to force Tesla to discontinue this development, pay them more, and use their products in future hardware."
If Tesla indeed only had about three months of Mobileye parts inventory when the supplier agreement was terminated, then Mobileye would certainly have assumed that it could disrupt Tesla automobile production and force Tesla to agree to its demands. But we now know that this would be a miscalculation to some degree. Mobileye could only disrupt near-term availability of Autopilot.
Neither Tesla nor Nvidia have said much about the hardware, except that Musk allowed during a media Q&A that it is based on the "Titan GPU", and the press release says it's 40 times faster than the previous (Mobileye) system. The system is equipped with a wealth of sensors including eight cameras, 12 ultrasonic sensors, and an upgraded forward-looking radar.
Musk was apparently being a little imprecise (gee, when does that ever happen?). There is no Titan GPU per se, and I'm sure he did not mean that the system incorporates the Titan X PC add-in board. The Titan X uses the Pascal GP102 graphics processor, second only to the GP100 Nvidia uses in its Pascal Tesla P100 accelerators for data center use. The GP102 probably would be a good choice for the application.
Given the short time frame, I'm convinced that this GP102 is embedded in some variant of the Drive PX 2 that Nvidia unveiled at CES this year. A stock PX 2 features two Parker generation Tegra SOCs as well as two unspecified Pascal GPUs. Depending on Tesla's requirements, Nvidia may have been called upon to customize the system, possibly reducing the number of processors. The PX 2 might also have been customized to accommodate Tesla's specific suite of sensors.
The use of Drive PX 2 (in some form) is significant for Nvidia investors, since it means that Nvidia isn't merely supplying chips to an OEM, a business Nvidia has been trying to get out of. Instead, Nvidia is providing a complete system, with PC board, processors, and other chips for camera and sensor interfaces.
Along with an integrated hardware solution, Nvidia also provides Driveworks, a software development kit (NYSEARCA:SDK) that provides algorithms for object detection, path planning, and map localization. This is in addition to the "deep learning" AI tools available for the GPUs.
Nvidia's SDKs would put Tesla well along the path of replacing Mobileye, especially since Tesla had already taken over "higher level" Autopilot functions. But even the SDKs would probably not be enough for Tesla to build an equivalent of the current Autopilot system without a head start.
Fortunately for Tesla, Nvidia has been working on self-driving car technology for some time, as demonstrated in this video released in February:
And then there was the quick interview between Nvidia CEO Jen-Hsun Huang and Elon Musk during the keynote of the 2015 GPU Technology Conference hosted by Nvidia (shown in the photo at the top of the article). At the conference Musk stated:
We'll take autonomous cars for granted in quite a short time. I almost view it as a solved problem. We know what to do, and we'll be there in a few years.
At the time, this was passed off as idle speculation on Musk's part, but in retrospect, it appears he was basing his observations on the work that Nvidia was doing. And Jen-Hsun did not take issue with his prediction.
I've noticed a tendency of Tesla's critics to disparage Tesla on technical grounds. While I have my own misgivings about Tesla's financial management, I've never doubted Tesla's technical capabilities. Elon Musk has proved very astute technically, which is why SpaceX is landing rockets safely on barges in the middle of the ocean. Generally, I think it's a big mistake for non-technical analysts to try to second guess Tesla on technical grounds.
I don't believe that the software development effort is so immature that the Nvidia hardware will be obsolete before it is ready to be deployed. I expect that software capability equivalent to first generation Autopilot will be ready in 3-4 months, with future upgrades adding more capability in the coming months and years.
Whether the systems are ever capable of full autonomy remains to be seen. It's entirely possible that Tesla is overselling this. On the other hand, Musk and Huang may know something we don't.
All of this discussion is by way of motivating the value to the companies of the collaboration. The economic benefit to Nvidia now appears to be immediate and significant. Lacking more specific information, I'm ball parking $1,000 per unit revenue for Nvidia. At 50,000 systems delivered every quarter, that adds about $50 million to Nvidia's top line.
The deeper value to both companies is that if they pull this off, or even get close, Autopilot II (or whatever it's called) will exert tremendous gravitational pull throughout the industry. Automakers are not going to be willing to wait for an Intel (NASDAQ:INTC) Mobileye collaboration to yield results some time around 2020 if the Nvidia/Tesla collaboration already has a product ready for delivery in 2018. There's clearly revenue potential for both companies.
The near term revenue boost for Nvidia make its a strong buy. I maintain a sell rating on Tesla due to the misguided SolarCity (NASDAQ:SCTY) merger, with the utmost regret.
Disclosure: I am/we are long NVDA.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.