Recently, more and more people are reckoning that AI may be more hype, less reality. For example, 75% of the executives said so in a 2021 KPMG survey and IBM Watson’s overpromised & underdelivered on AI health care. In my opinion, it’s just a natural cycle of new technology adoption in business, despite news typically dramatize them. But it does affect the investment into new ventures. Which drives the market selection process, similar to natural selection. At a time like this, a new species may arise to adapt to the changes. Just like Amazon & etc. survived very well from the Dot-Com bubble. Tesla is likely a survivor, and its AI Day provides a peek of a better future. Let’s see how it’s a different kind of beast, that may eat dinosaurs’ lunch.
A killer AI application
Self-driving is a mass consumer AI use case, or the mother of all AI projects as Apple CEO Tim Cook called it. It’s not only very technically challenged. But more importantly, it has a high-value use case many consumers need & want. It’s a huge gold mine for AI & seems to be reachable. So even if the AI market music stops, the drum beats on self-driving ventures to race commercialization will continue. tl;dr technology advance is indeed a critical mean. But “necessity is the mother of invention”.
Sure, Tesla’s Full Self-Driving is only Level 2 Autonomous today. The question is what’s a better & faster way to get the higher levels massively adopted. Tesla is gaining a unique advantage by combining 2 “secret weapons”: harvesting mass production data & fast software iterations. They will be the critical driving functions of the self-driving virtuous cycle: more FSD usages, more production data, better FSD capability…
Whole new full-stack innovations
In the AI arms race, Tesla AI day is also most a recruitment show of a rising superpower. Yes, it’s not just showing off their powerful “toys”, but also attracting new talents because this is a talent-intensive game. To be honest, it’s very appealing. Not only Tesla has a few leading techs, but more importantly putting all critical bits under the same roof & dare to innovate the end-to-end stack from scratch. Just name a few:
- Top self-driving cars from sensors, controls, compute Hardware 3(Tesla chip inside) to software.
- Data platform, huge annotated dataset & productive toolset.
- The state of the art simulation to solve edge cases to reach 99.9999…%
- Optimized & most powerful Neural Network cloud training compute stack from the chip(D1), Dojo server farm to software.
Bonuses: 2 more extrapolations down to the road.
- For most moving to the cloud is a no-brainer now, but not Tesla. Tesla is likely building Dojo data centers soon or later. Psst Micro Data Center may be the “Agile” way to get started.
- Many have been trying to make the last mile connectivity better, e.g. Google Fiber & 5G. Starlink is still in Musk’s pocket. Go figure 🤔.
Prosperable business model
If to open source of not was asked in the Q&A. Musk answered these are all expensive & need to pay for. But he thinks: “If other car companies want to license it and use it in their cars. That would be cool. This is not intended to be just limited to Tesla cars.” It is nice for Musk to be open on that, but the barriers are most likely how much carmakers are open to trust & work with their competitors. For history, they typically don’t.
Furthermore, founding is a real problem even in tech. Most people don’t want to work on a business for free, right? “Luckily” Tesla has 3 unique advantages:
- Making Self-Driving as a subscription service. In other words, Tesla is the “Netflix of Self-Driving”.
- Betting on Camera, a high potential disruptive innovation to LIDAR, because it’s much cheaper. Sure LIDAR is much more accurate for now, but computer vision improves much quicker. There is no doubt that LIDAR bets will always do better in some environments. But for most of the cases, Camera eyes will be “good enough”. Furthermore, thanks to camera & computer vision are highly valued in mobile, the spillover effect help to accelerate. Just try to imagine what their new architect of the Mult-Cam Vector Space Prediction with pre-fusion can achieve & what else to come next year.
- Adjacent opportunity — Tesla Bot. The doubt of Tesla’s “overpromise” on Tesla Bot is indeed grounded with track records. Nevertheless, it’s a promising adjacent opportunity for Tesla. Because that problem domain nicely overlaps with its Self-Driving solution stack. Anyway, kudos to team Tesla & whoever walks the talk: “If you’re not failing at something, you’re not trying hard enough.”
Respectful opponents
Yep, I sound like a Tesla fanboy. To be honest I’m not. I’m just helpless optimistic & also admire their efforts in pushing the boundaries of Self-Driving to make the way we move better earlier. It’s good Tesla is leading the pack. Nevertheless, I guess Tesla might also appreciate it if there are real respectful opponents. So what could be a good one does? I would try:
- An open platform levels the playing field to help the rest of the carmakers to compete at the same level. So that consumers may have different & even better choices.
- An open data platform facilitates interoperability & unleashes 3rd party innovations e.g. Driver Journey Simulated Reality Universe.
Wish the best team wins 😉.
Full Disclosure
The opinions stated here are my own, not those of my company. They are mostly extrapolations from public information. I don’t have insider knowledge of those companies, nor a whatever expert.
