Tesla is aggressively scaling up its artificial intelligence infrastructure to train next-generation autonomous models. Last week’s Q2 2026 earnings report revealed that Tesla’s installed compute capacity at Gigafactory Texas expanded dramatically over the first half of the year.
According to Tesla’s financial filings, on-site AI training compute in Texas more than doubled in terms of megawatts (MW) from January to June, with the bulk of the additional capacity coming online in Q2. Looking at the historical chart shared in the report, Tesla finished 2025 at roughly 115 MW of training capacity. That number edged up slightly to around 135 MW at the end of Q1 2026 before nearly doubling to approximately 260 MW by the end of Q2, driven primarily by the activation of new supercomputing clusters.
Not a Tesla App
Tesla’s existing AI training capacity is now orders of magnitude higher than what it was when the graph saw its first uptick all the way back in Q2 2022. Tesla also outlined its planned future capacity, projecting nearly 400 MW of AI compute at Giga Texas by the end of this year. If Tesla achieves those numbers, it will have tripled its Texas AI infrastructure over the course of 2026.
Detailing the purpose of this hardware expansion, Tesla said in the report:
“Cortex 2 supports the development of both vehicle and humanoid robot autonomy software and will ramp further over the rest of the year to ensure we have sufficient compute resources.”
What Tesla Uses Its Massive Compute Power For
Training complex neural networks requires immense raw processing power. Tesla relies on these massive clusters to ingest billions of video frames collected from its global vehicle fleet, training the end-to-end neural nets that power consumer Full Self-Driving software and the driverless Robotaxi service.
The expanded compute capacity also feeds directly into Optimus, helping train AI models for physical dexterity, spatial awareness, and real-world task execution for humanoid robots.
Having more processing power allows engineers to train larger models faster and iterate on software builds much more frequently.
Fueling the Next Generation of Autonomous Vehicles
This expansion in compute capacity directly supports several major software pushes across Tesla’s autonomous ecosystem.
Not a Tesla App
During last week’s earnings call, executives confirmed that Tesla Robotaxis are already running FSD v15, bringing the next architectural leap in self-driving software to active fleets early. That software stack is already seeing heavy real-world validation, given that Tesla revealed its Robotaxi vehicles have driven over 380,000 unsupervised miles across multiple cities.
Additionally, Elon Musk announced during the call that FSD is coming to the Tesla Semi. Tesla’s expanded processing resources will likely also contribute to training neural nets to handle a massive Class 8 commercial truck.
Interestingly, this extra power also fuels Tesla’s shifting model training strategy. We recently learned that current-gen consumer HW4 vehicles actually run distilled FSD models derived from flagship models trained for the Cybercab. With Cortex 2 continuing to ramp up, we expect faster model iterations for both commercial fleets and consumer vehicles alike.