Tesla is racing back into the AI hardware arms race, with Elon Musk reversing course on a flagship supercomputing project that he had previously sidelined. The company is reviving its in-house Dojo effort at the same time as it finishes a new AI chip design, signaling a deeper bet that custom silicon and massive compute will define its future beyond electric cars.
I see this as more than a technical pivot. It is a strategic U turn that ties Tesla’s survival to whether it can turn AI, robotics, and supercomputing into profitable platforms just as its core vehicle business comes under pressure.
Dojo3 returns as Tesla doubles down on custom silicon
The clearest sign of Musk’s reversal is his decision to restart work on Dojo3, the third generation of Tesla’s in-house supercomputer project. Elon Musk posted on X that Tesla will restart work on the system after previously halting the earlier Dojo effort when the economics and performance “simply did not pan out,” a shift that underscores how quickly the company’s AI priorities are evolving and how willing Musk is to change course when he thinks the technology has caught up to his ambitions, as reflected in his recent Dojo3 comments. In a separate post on X, Musk said that Now that the AI5 chip design is in good shape, Tesla will restart work on Dojo3, making it clear that the supercomputer’s revival is directly tied to progress on the company’s next generation of AI silicon, a linkage he emphasized when he described how many training chips he wants on a single board in remarks captured in Musk.
That AI5 chip is central to the U turn. On January 18, 2026, Musk said Tesla is almost done with the AI5 chip design, which he claimed will deliver 10 times the computing power of the company’s current HW4 hardware and will be manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), a leap he framed as the foundation for what he called the world’s highest performance AI training systems in his AI5 chip update. Reporting on Tesla’s silicon roadmap describes this as a New Chapter in Tesla’s Silicon Strategy Tesla, with the company going all in on chipmaking and positioning AI5 to rival NVIDIA’s Blackwell architecture with what Musk argues will be much better performance per dollar, a claim that highlights how aggressively Tesla is trying to control its own AI stack rather than rely solely on external GPU suppliers, as detailed in the New Chapter analysis.
Cortex 2.0 turns Giga Texas into an AI factory
While Dojo3 and AI5 grab headlines, the physical backbone of Tesla’s AI push is rising in Austin. At Giga Texas, Musk has already named the company’s supercomputing cluster Cortex, a label he revealed after a walkthrough of the facility when he told followers that the Giga Texas supercomputing cluster is named Cortex and shared that he had just completed a tour of the site, a moment that cemented the factory’s identity as more than a car plant in his Cortex comments. Despite the presence of an existing supercomputer cluster at Tesla’s Giga Texas facility, satellite images and company plans show construction of Cortex 2.0, an expanded complex that will host an even larger number of GPUs and is expected to come online in Austin within the coming weeks, a build out that illustrates how quickly Tesla is scaling its training capacity according to Despite the reporting.
The company has described Cortex 2.0 as a new supercomputer facility in Texas that will power Optimus humanoid robots and autonomous electric vehicles, with the site expected to become operational either later this year or in early 2026 and then continue expanding through the end of the decade as Tesla adds an even larger number of GPUs to the cluster, a long term roadmap that underlines how central AI training has become to its manufacturing footprint, as outlined in Cortex 2.0. In parallel, the company is struggling with vehicle sales and is shifting focus to AI and robotics, investing in autonomous mobility through its upcoming Cybercab robotaxi concept and planning to launch Robotaxis and related services in Austin by April 2026, a timeline that shows how the supercomputing build out at Giga Texas is meant to support real world products like Cybercab and other Robotaxis and services rather than remain a purely internal research tool, as detailed in the Robotaxis and coverage.
More from Morning Overview