Tesla Plans 400 MW of AI Training Compute, From Over 225 MW

Tesla plans to reach 400 MW of AI training compute by the end of 2026, the company said in its Q3 2026 Update on 21 October 2026. Its two Texas training clusters, Cortex 1 and Cortex 2, are listed at more than 105 MW and more than 120 MW, together more than 225 MW (our arithmetic).

Tesla says the compute supports “the development of both vehicle and humanoid robot autonomy software.” In the same quarter its research and development spending rose to $2.76 billion, up about 69% from $1.63 billion a year earlier (our arithmetic), and Optimus production began on high-volume lines in Fremont.

Business Pill · PHYSICAL AI

A short explainer on physical AI: what changes when AI systems control cars, robots and machines. It teaches the general idea only and says nothing about Tesla’s results.

Tesla Q3 2026 explained · video

The quarter in about a minute and a half, from our published pieces (21 Oct 2026).

What Tesla Reported

Tesla’s infrastructure table lists “SI Training Compute” in Texas: Cortex 1 at more than 105 MW and Cortex 2 at more than 120 MW, both in production. A footnote says the figure “Excludes 40 MW of other compute not part of Cortex 1 or Cortex 2.” SI is Tesla’s term for Super Intelligence. Measuring compute in megawatts means counting the power the computing hardware draws, not the number of chips.

The update says: “We plan to reach 400 MW of SI training compute by the end of the year, supporting the development of both vehicle and humanoid robot autonomy software.” It also says Tesla “continued construction for the Austin Fab, where we plan to start commissioning equipment in 2027.” Its outlook says capacity build-out for “SI compute, solar, battery material and semiconductor manufacturing” is underway.

Going from more than 225 MW to 400 MW by year-end means adding up to about 175 MW in the fourth quarter (our arithmetic, using the stated minimums).

What Was Posted on X

Tesla’s own post on 21 October summarising its Q3 Shareholder Update: under SI Compute it says Tesla is on track for 400 MW of training compute by year end, feeding both vehicle and humanoid autonomy, and that Austin Fab equipment commissioning is planned for 2027; under Robotics it says Optimus production started on its high-volume lines in Fremont. This piece takes its figures from the update itself.

Where the Spending Shows

Research and development expense was $2,761 million in the quarter, against $1,630 million a year earlier and $2,371 million in the second quarter. Tesla says the increase in operating expenses was “driven by SI and other R&D projects, SBC (including 2025 CEO Performance Award) and SG&A.”

Capital expenditures were $4.73 billion, against $2.25 billion a year earlier. Over the trailing twelve months they were $15.41 billion, against $8.91 billion a year earlier, an increase of about 73% (our arithmetic). Tesla says it continues to invest “through R&D and Capex spend across our businesses, including SI, robotics and energy.”

Robots on the Line

“Optimus production began in Fremont on our high-volume lines,” the update says. “The bots coming off this line are the first production-quality bots that we have manufactured.” Tesla adds that it has “proven that the current manufacturing process for Optimus is suitable for a commercial product.” Its capacity table lists Optimus in early ramp in California and under construction in Texas, with no unit capacity given.

The cash flow statement also shows a $2,002 million purchase of a SpaceX equity investment in the first quarter of 2026. Tesla recorded an unrealised gain of $1,005 million on it in the second quarter and an unrealised loss of $142 million in the third.

Connecting the Coverage

Our results piece on the same release covers Tesla’s headline numbers: capital expenditures of $4.73 billion, against $2.25 billion a year earlier, and operating income down 56% to $718 million (our results piece).

On 20 October CNBC reported from SpaceX’s earnings call that its AI business is on track to be able to bring 200,000 of Nvidia’s Grace Blackwell systems online every month for the rest of the year (our Blackwell piece). SpaceX’s own release said third-quarter revenue rose 161% to $14.6 billion, “fueled by our Super Intelligence (SI) segment” (our SpaceX results piece).

On 20 October the Memphis City Council passed a 135-day pause on new data centers in the city where SpaceXAI built Colossus 1 (our Memphis piece). On 19 October Tesla’s proxy listed “1 million bots delivered” among Musk’s pay-award milestones (our Optimus piece). And on 20 October Musk said SpaceX and Tesla have formed special committees to discuss a potential merger, with no agreement on terms (our merger piece).

The Structural Read

As we read it, Tesla now reports AI compute the way utilities report power, in megawatts, and ties it to two products: vehicle autonomy and humanoid robot autonomy software.

As we read it, the spending shows in operating costs before it shows in products: research and development rose to $2.76 billion, and Tesla says the rise in operating expenses was driven by SI and other R&D projects, SBC and SG&A.

As we read it, Tesla is building more of the stack itself: the same update lists Austin Fab construction, cathode and lithium refining in Texas and LFP cells in Nevada alongside the Cortex clusters.

Three Implications

FOR AI SUPPLIERS Tesla’s plan implies up to about 175 MW of training compute added in the fourth quarter (our arithmetic); the update names no chip supplier.

FOR TESLA SHAREHOLDERS Research and development expense rose to $2,761 million from $1,630 million a year earlier, and trailing-twelve-month capital expenditures reached $15.41 billion, per the update.

FOR MUSK’S OTHER COMPANIES The update shows a $2,002 million SpaceX equity investment bought in the first quarter, and our earlier pieces cover SpaceX’s own compute build-out and the merger committees.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Nine Layers of AI.

The framework’s framing: “We are entering a phase where four scaling laws are simultaneously shaping the AI ecosystem, while several converging forces are radically transforming the physical layers underpinning AI itself.”

As we read it, Tesla’s update spans several of those physical layers at once, from training compute in megawatts to a chip fab, battery materials and robots on a production line. The framework was published in May 2026 and does not discuss this update.

What Is Not Established

The update, which we read in full, does not say how many chips the 400 MW represents, which chips Tesla uses, what the build costs, or whether it includes the 40 MW of other compute. It gives no Optimus production volume. Comments from the earnings call are not in this version. We did not contact Tesla.

The Bottom Line

Tesla plans 400 MW of AI training compute by the end of 2026, against more than 225 MW across Cortex 1 and Cortex 2 at the end of September (our arithmetic), as its quarterly research and development spending rose to $2.76 billion and Optimus production began in Fremont.