Overlook Sensors, Tesla’s AI Coaching Prices Are Hovering

Assist CleanTechnica’s work by a Substack subscription or on Stripe.


Or help our Kickstarter marketing campaign!

One argument Elon Musk and Tesla followers have made for ages — for a couple of decade — is that the additional prices of sensors like lidar and radar for self-driving automobiles are usually not price it, are illogical, and would be the loss of life of an organization like Waymo. Simply use cameras and reduce prices. The AI software program will do the remainder.

Nicely, one factor we’ve seen lately, particularly in 2025, as Tesla has tried to succeed in true self-driving and robotaxi deployment, is rising AI infrastructure prices. Like, severely rising prices.

It’s not 100% clear what Tesla is spending on this, as a result of the prices are rolled right into a broader class of prices. Nonetheless, as I famous in my article earlier immediately, Tesla’s working bills rose a whopping 39% within the 4th quarter whereas revenue from operations dropped 11%, and Tesla really acknowledged that these rising prices have been “pushed by AI and different R&D tasks and SG&A.” Imprecise, however you get the purpose.

Digging into Tesla’s shareholder letter slightly additional, there’s extra on this matter. To start with, there’s this graph:

That graph made my jaw drop and my eyes pop. You may see a big rise in AI coaching capability all through 2025, but it surely’s nothing in comparison with what the corporate is projecting for 2026! How a lot is all of that going to value? (And, by the best way, is Tesla powering that with extremely polluting fossil gasoline energy vegetation?)

Within the textual content portion of the shareholder letter, Tesla wrote the next underneath the “AI Coaching Compute” subheading:

“We’re at the moment constructing Cortex 2 at Gigafactory Texas to additional improve our AI coaching compute capability. Within the first half of 2026, we plan to greater than double the scale of onsite compute in Texas (by way of H100 equivalents). We goal to maximise capital effectivity by scaling coaching compute judiciously, together with when the coaching backlog will get too lengthy or in anticipation of better demand from our engineers to help our AI-related choices.”

Catch that on the finish? Clearly, that is very pricey, and so the thought is that they may simply develop capability when it’s actually, actually wanted. Apparently, they will’t actually afford to develop capability forward of time — or, at the least, which may look actually dangerous.

Regardless of the hype, Tesla doesn’t but have absolutely driverless robotaxis deployed with out human supervisors. They’ve simply eliminated human supervisors from a few vehicles in Austin, however there are nonetheless trailing vehicles with people in them following these robotaxis. All of this rising AI compute capability has been in an effort to get to robotaxi deployment.

Additionally, regardless of saying for a couple of decade that they’d principally have the ability to flip a swap and activate robotaxi functionality throughout the fleet unexpectedly — counter to how Waymo does it, one metropolis at a time — it seems that Tesla has to go the city-by-city route, too. It’s at one stage in Austin, one other stage in San Francisco, and goals to roll out in fewer than 10 cities by the top of 2026. And keep in mind that’s the plan based on somebody who has missed such rollout targets for a couple of decade and who stated roughly 6 months in the past that Tesla robotaxis would cowl round half of the US inhabitants by the top of 2025.

I stated final yr that it appeared Tesla was in a form of race towards time or race towards itself on this regard. It wants to succeed in deployment quickly to stimulate extra automobile gross sales and make more cash on buyer companies, or else these rising AI prices are going to place the corporate’s funds within the detrimental. In fact, even that could possibly be wonderful for some time, since Tesla is sitting on tons of money, but it surely actually wouldn’t be a superb look or a optimistic factor to go from quarterly income again to quarterly losses.

We’ll have to attend and see what Q1 2025 brings, however I feel we are able to rely on these working bills persevering with to rise.

Oh yeah, then there’s additionally Tesla’s $2 billion coming funding xAI, regardless of TSLA shareholders voting towards it in 2015. That’s one other story, although.