Tesla Goes for Chips: Terafab Launches This Week

On Saturday, March 14, Elon Musk posted a brief but loaded announcement on X: "Terafab Project launches in 7 days." That countdown ends on March 21, 2026 — and with it, Tesla officially enters the most complex manufacturing domain in human industry: semiconductor fabrication.

The project, which Musk has been telegraphing for years, is Tesla's bid to design and manufacture its own AI chips — starting with the AI5, the fifth-generation chip intended to power its Full Self-Driving and autonomous robotaxi systems. The ambition is staggering: a vertically integrated "gigantic chip fab" that combines logic processing, memory, and packaging under one roof.

Why Tesla Wants Its Own Fab

The strategic logic is straightforward. Tesla's AI roadmap requires enormous volumes of custom silicon, and even with TSMC and Samsung as manufacturing partners, Musk has argued for years that supply simply cannot keep pace with demand.

"Even when we extrapolate the best-case scenario for chip production from our suppliers, it's still not enough," Musk said at Tesla's AGM last year. "So I think we may have to do a Tesla Terafab. It's like giga but way bigger. I can't see any other way to get to the volume of chips that we're looking for."

The Tesla AI5 chip is among the first products Terafab is designed to produce, with small-batch production expected in 2026 and volume production projected for 2027. Musk has also left the door open to a partnership with Intel, though no deal has been signed.

The Case for Skepticism

Not everyone is convinced. Electrek published a pointed analysis on Monday noting that Tesla "has absolutely zero experience manufacturing semiconductors" — and that its closest comparable venture, the 4680 battery cell program, should give investors serious pause.

The parallels are uncomfortable. At Battery Day in September 2020, Musk promised 100 GWh of in-house cell production by 2022, a 56% battery cost reduction, and a $25,000 vehicle. None of it materialized on schedule. By early 2025 — five years after the announcement — actual 4680 production was estimated at roughly 20 GWh per year, a fraction of the original target. The 4680's dry electrode process proved far harder than anticipated. A key supply chain partner, L&F Co., wrote down its Tesla deal by 99%.

Semiconductor fabrication is, by most engineering measures, orders of magnitude more complex than battery cell production. Building a competitive fab requires process engineers specializing in EUV lithography, chemical-mechanical planarization, yield management, and dozens of other highly specialized disciplines that Tesla has never employed.

The Talent Gap

Tesla did once have legitimate chip design talent. The company hired legendary architect Jim Keller in 2016 and built a team that produced the HW3 and HW4 Autopilot inference chips — a genuine achievement. But much of that talent is gone. Keller left in 2018. Ganesh Venkataramanan, who led the Dojo supercomputer project, departed in 2023. Then in August 2025, Musk canceled the entire Dojo program, and Peter Bannon — the architect in charge of all custom silicon at Tesla — left the company, taking roughly 20 engineers to his new startup, DensityAI.

What Tesla had was chip design talent. Chip manufacturing talent is an entirely different workforce.

Even Jensen Huang Is Skeptical

Perhaps the most stinging rebuke came not from a critic, but from the person with the most incentive to want more chip production capacity in the world. According to Electrek's analysis, Nvidia CEO Jensen Huang has privately indicated that building a competitive semiconductor fab from scratch is "virtually impossible" — a remarkable statement from the man who runs the world's most valuable chip company and would benefit from more competition in chip supply.

What's Actually Launching

Details on what exactly "launches" on March 21 remain thin. Reuters noted that Tesla did not respond to requests for more information about the project. The announcement appears to refer to the formal commencement of construction or operations at a designated Terafab facility, not to any chip rolling off a production line. Volume manufacturing of the AI5 chip is still projected for 2027 at the earliest.

For now, Tesla's Terafab occupies a familiar Musk position: a bold bet on a future that may transform an industry — or may follow the same pattern of overpromising, years of delays, and results that fall short of the original vision. The semiconductor industry will be watching closely.