Our top 4 Tesla AI day predictions

Tesla’s second AI Day is fast approaching, and with it a flurry of speculations as to what will be discussed at the event — a phenomenon that we at TechCrunch are certainly not above.

AI Day is the luxury electric vehicle maker’s annual event to hype up its artificial intelligence capabilities across different business segments, with the ultimate goal of recruiting the brightest minds and engineers.

Even without Tesla’s star head of AI Andrej Karpathy, who left the company in July, expect this event to be full of technical jargon and big promises from Tesla’s executives, including the big dog himself, CEO Elon Musk.

We’ll talk about what we’re expecting to hear about based on recent news, as well as last year’s AI day event. But first, how to watch.

Tesla’s AI Day is scheduled for September 30 and should begin around 5 p.m. PT, although Tesla’s events are often notorious for running late. Remote viewers can watch it livestreamed on Tesla’s YouTube channel or on its website.

Now, onto the forecasts.

An Optimus prototype


The Tesla Bot is on display at the Tesla Giga Texas manufacturing facility during the “Cyber Rodeo” grand opening party on April 7, 2022 in Austin, Texas. Image Credits: Getty Images

At last year’s AI Day, Tesla announced plans to produce a humanoid robot named Optimus, one that is “intended to be friendly and navigate through a world built for humans,” according to Musk. 

Tesla AI Day was originally scheduled for August 19, but Musk pushed the date back in the hopes of having a working Optimus prototype by this week. On September 20, Musk reiterated his goal to have the prototype ready by the end of the month.

During Tesla’s first-quarter earnings call this year, the celebrity executive told investors the company continues to work on the 5’8″ robot, which will rely on Tesla’s work in neural networks and its Dojo advanced supercomputer (more on that in a bit), to handle the drudgery of human existence, AKA everyday tasks like grocery shopping. At the time, Musk said Optimus would be “worth more than the car business, worth more than FSD.”

FSD, or Full Self-Driving, is Tesla’s advanced driver assistance system (ADAS) that is definitely not fully autonomous, but can perform some autonomous driving tasks for drivers willing to shell out $15,000. Musk has also said that Tesla, without FSD, is worth “basically zero.

During Tesla’s Austin gigafactory opening in April, Musk said Optimus, along with a new wave of products, will be introduced in 2023. Which brings us to the alleged prototype…

If it manages to make it onstage, we’re expecting Optimus to weigh in at around 125 pounds, walk at about 5 miles per hour, have a screen for a face and be weak enough that a human could overpower it if the bot decided to have a mind of its own one day.

Unless Tesla’s AI team has made extremely rapid progress on getting this robot to perform human-like tasks, it seems unlikely that a consumer product will roll out in the next year. And even if one does, will it be any good?

Musk has been promising to solve full self-driving for years now — in 2016, he said Tesla was less than two years away from solving “complete autonomy.” FSD still does not have full self-driving capabilities, and those shortcomings are causing problems for Tesla. The automaker is facing state and federal scrutiny for Autopilot, its lesser ADAS version, and FSD, even as it expands access to the latter, as well as multiple lawsuits claiming Tesla has falsely advertised the capabilities of its ADAS.

Updates to Autopilot and FSD

What Tesla’s Autopilot sees. Image Credits: Tesla

Musk also recently tweeted that Tesla’s AI team has an end-of-month deadline for “actually smart” summon and auto park, both of which are features of Tesla’s Enhanced Autopilot, its $6,000 ADAS. Smart summon, which Musk’s tweet implies has been operating at a less than perfect performance, promises to navigate the car in complex environments and parking spaces, “maneuvering around objects as necessary to come find you in a parking lot.” Autopark helps “automatically parallel or perpendicular park your car, with a single touch,” according to Tesla’s website.

Tesla is likely to reiterate its goal to get to a widespread rollout of FSD beta in North America by the end of this year. We’re expecting the automaker to also provide specific updates to the next version of FSD beta, like having the vehicle take note of surrounding cars and increase speed to keep up with traffic.

A Tesla robotaxi

During Tesla’s Cyber Rodeo to celebrate the opening of the Austin gigafactory in April,  Musk said 2022 would be the year of scaling up, while 2023 would garner a “massive wave of new products.”

At the time, he mentioned a “dedicated robotaxi” that would look “quite futuristic.” A couple weeks later, during Tesla’s Q1 earnings call, Musk spoke more about the robotaxi, saying that it would be purpose-built for autonomy — meaning it will be built without a steering wheel or pedals. It will also offer riders the lowest cost per mile of any other transport, and be cheaper than a subsidized bus ticket, the CEO promised.

Tesla has been talking about a robotaxi concept since 2016, when Musk came out with his Master Plan Part Deux. The CEO imagined a Tesla Network taxi service that would allow Tesla owners to throw their cars in the autonomous ring to generate income on their behalf while they’re at work or on vacation.

Will Musk finally provide specifics about what these robotaxis will actually look like, how they will be commercialized, or whether Tesla has set a date for production? Find out on Friday.

Updates to the Dojo supercomputer

Screenshot from Tesla AI Day 2021 presentation on D1 Chip for Dojo supercomputer. Image Credits: Tesla

This is where stuff will get really jargony.

Last year, Tesla unveiled its in-house-built chip, the D1, that the company is using to run its supercomputer, Dojo. Tesla’s AI architecture is dependent on Dojo, the neural network training supercomputer that Musk says can process vast amounts of camera imaging data four times faster than other computing systems. Having a super fast computer is important for Tesla because it relies solely on cameras, rather than lidar and radar, for autonomous operations. This speed depends on the processing power of D1.

In late August, Tesla participated in Hot Chips 34, a conference on semiconductors, during which the company revealed new information about Dojo’s AI system architecture and supercomputer. Serve the Home, an industry blog, published Tesla’s slides, which detail specifics about how Dojo’s distributed system facilitates machine learning and how this will lead Tesla to generalized autonomy. The presentations were highly technical, so we’ll skip the specifics on hardware and capabilities, but we’re expecting much of the same information to be shared on AI Day, as well as how it all contributes to performance.

Surprise announcements

Again, Musk has repeatedly hinted at a wave of new products coming forth next year, so we wouldn’t be surprised if he came out with something truly wacky and eyebrow-raising.

We’ll be watching the event so you don’t have to, so check back in with TechCrunch for updates at what news is actually shared during Tesla’s AI Day.