A Detailed Explanation of Terafab—Why Musk Insists on Building a "Chip Factory"

Tesla’s growth narrative is shifting its underlying foundation: batteries and vehicles remain cash flow sources, but the next phase’s imagination is focused on physical AI—Robotaxi, Optimus, AI models shared between humans and vehicles, and the computing power stack. The bottleneck is also shifting: not battery capacity, but chip supply, manufacturing capability, and supply chain security under geopolitical tensions.

According to ChaseWind Trading Desk, Dan Levy, an analyst at Barclays covering US automotive and mobility, straightforwardly wrote in the latest research report: “Chips will become the pillar of Tesla’s next growth phase,” and to push this growth to “mega-scale,” Tesla needs a Terafab.

Barclays considers Terafab as Tesla’s next round of vertical integration “extreme move”: building a closed-loop in the US for logic chips, memory, and advanced packaging, aiming to cover AI6/AI7+ needed for Robotaxi/Optimus, as well as Dojo-related chips for data centers.

The problem is, once the scale is opened up according to Tesla’s standards, the bill becomes daunting: Barclays estimates that even at a lower target, capital expenditure would be around $20-25 billion, with more aggressive capacity plans pushing it to $40-50 billion. Bank of America bluntly states that the Terafab project faces a capital expenditure threshold of over $60 billion, and even in the most ideal scenario, its 2nm wafer pricing will still be higher than TSMC’s, making ROI difficult to justify.

More critically, execution path. Tesla lacks experience in large-scale chip manufacturing; leading process, tooling, yield, and packaging are not problems that can be solved simply by “building a factory.” Compared to doing it alone, Barclays favors a “collaborative Terafab”—Tesla funds capacity locking, while fabs handle manufacturing, with potential partners including Samsung, TSMC, or even Intel.

What exactly would Terafab do? Logic, memory, and packaging in a US domestic closed loop

The report traces the timeline back to Battery Day 2020: Tesla then used battery capacity as the “strategic pillar for the next decade.” Now, in the new phase driven by Robotaxi and humanoid robots, Barclays believes the new pillar becomes chips—it directly determines inference at the vehicle end, inference at robots, and the expansion of data center training/validation chains.

Elon Musk repeatedly emphasized at the November 2025 shareholder meeting and Q4 2025 call that: over the next 3-4 years, chips (including AI logic and memory) will be the growth-limiting factor. Recently, he also mentioned on X that Terafab will start within days, and the report takes this as a “next big move” time window for discussion.

The ambition of Terafab is not just capacity expansion but to create a complete closed loop of “logic + memory + advanced packaging” in the US.

Logic chips: Tesla has accumulated considerable design experience from HW3 to AI6, and has established manufacturing anchors at TSMC Arizona and Samsung Texas, making this the relatively mature part among the three.

Packaging: Tesla has design experience but limited manufacturing experience—packaging is crucial for tightly coupling logic and memory, achieving higher efficiency, and is a key variable for overall compute unit performance.

Memory: the biggest gap—Tesla has neither design nor manufacturing experience, and the supply of advanced memory in the US is severely lacking—Micron’s Idaho plant won’t produce until mid-2027, and multiple factories in New York are not expected to start until 2030.

This makes “full supply chain localization” not just a linear increase but a multiplicative challenge.

Behind Terafab are two strategic motivations that cannot be ignored. First is de-risking geopolitics: potential disruptions in the Taiwan Strait and US-China trade tensions keep the logic chip supply chain dependent on TSMC under a sword of Damocles, while the lack of advanced memory supply in the US further increases the appeal of “full supply chain localization.” Second is design control: Tesla aims to tightly couple logic and memory through packaging, reducing discrete components, and creating truly customized compute units for its software stack—management expects AI5 performance comparable to Nvidia’s Blackwell (Thor), but with only one-third the power consumption and less than 10% the cost.

This unified “one chip, multiple scenarios” strategy converges inference at the vehicle, Optimus, and some data center applications onto self-developed chips, making AI5/AI6 the hub connecting vehicle, robot, and data center business lines.

The report estimates Tesla will purchase about 3-4 million chips in 2025. That means, if demand follows this trajectory, suppliers need to lock in not just “small capacity increases” but a high-risk round of capital investment. Musk has mentioned a 160,000 WSPM (wafer starts per month) plan, and Barclays believes this could correspond to about 24 million chips per year under good yield conditions.

$20-50 billion is just the starting point, and may be economically unsustainable!

The cost of the vision is a terrifying bill.

The report provides two cost ranges: approximately $20-25 billion for a capacity of about 12 million chips/year; and $40-50 billion if scaled to 24 million chips/year. For reference: Samsung’s Taylor project announced $17 billion for about 20,000 WSPM capacity, TSMC’s Arizona total investment disclosed at $165 billion, and Micron’s planned $100 billion in New York for multiple storage fabs. Musk’s 160,000 WSPM plan is close to “multiple large fabs stacked”—typical wafer factories produce about 20,000 to 40,000 WSPM.

A sharper issue is that these investments are not reflected in Tesla’s current 2026 capital expenditure guidance of $20 billion. Barclays’ own forecast for Tesla’s free cash flow in 2026 is already -$3 billion, and adding wafer fab-level investments would multiply cash flow pressures.

For this reason, the report is highly cautious about “building factories independently.” Tesla lacks large-scale chip manufacturing experience; the process complexity, EUV equipment cycles, and advanced packaging are not problems that can be solved simply by “building a factory.” The report even cites the Dojo project, which was canceled and impaired, and the 4680 battery’s underwhelming volume and performance, as lessons to remind the market not to equate “design capability” directly with “manufacturing ramp-up ability.”

Bank of America Merrill Lynch’s March 23 report bluntly states that the Terafab project faces a capital expenditure threshold of over $60 billion, and even in the most optimistic scenario, its 2nm wafer pricing will still be higher than TSMC’s, making ROI difficult to justify.

Specifically, BofA notes that under optimal assumptions of 100% utilization and yield, the front-end wafer fixed costs are about $6,000, still 1.3 to 1.5 times higher than TSMC’s advanced nodes. To maintain the required gross margin of about 45% for the technology and capacity roadmap, Terafab would need to price 2nm wafers at around $32,000, above TSMC’s roughly $30,000.

Even without technical difficulties, facility construction alone takes 3-5 years: about 1.5-2 years for factory readiness, 1 year for equipment installation, and 1-2 years for risk production and certification. The report estimates that if construction starts today, Terafab would earliest reach mass production by 2029.

Almost no precedent for doing it alone: a more likely scenario is “Tesla funds, giants build”

Barclays favors a “collaborative Terafab”: Tesla funds capacity locking, while fabs handle manufacturing, through capital investments, guarantees of losses, and other means to encourage suppliers to commit to more aggressive capacity ramp-up.

Among potential partners, Samsung is seen as the most natural choice—covering logic, packaging, and DRAM, and the report even suggests Tesla could potentially participate in planning its second Taylor fab; TSMC is also a natural partner but does not produce automotive-grade DRAM; Intel aligns with Musk’s preference for an American supply chain and may have available capacity.

For the stock market, Barclays advises restraint: the grand capacity goals for chips can continue fueling the long-term growth narrative, but once it comes to capital expenditure and free cash flow, the market needs more concrete execution details.

Barclays maintains an Equal Weight rating on Tesla, with a target price of $360. Compared to the report’s implied downside of about 5% from the closing price of $380, the chip story can be bigger, but until the execution path and cost boundaries are clearer, Terafab remains a “high risk, high reward, more likely collaboration” open proposition.

All the above insights are from ChaseWind Trading Desk.

For more detailed analysis, real-time commentary, and frontline research, please join 【**ChaseWind Trading Desk - Annual Membership**】

![](https://img-cdn.gateio.im/social/moments-1666a924d9-b5039a48c4-8b7abd-ceda62)

Risk Disclosure and Disclaimer

Market risks exist; investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions herein are suitable for their circumstances. Invest at your own risk.