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

```

Tesla’s growth narrative is shifting at its core: batteries and vehicles are still sources of cash flow, but the next stage of imagination is pinned on physical AI—Robotaxi, Optimus, and AI models and compute stacks shared between humans and cars. The bottleneck is also shifting: it is no longer battery production capacity, but chip supply, manufacturing capability, and supply chain security amid geopolitical tensions.

According to Wind Chaser Trading Desk, Barclays US automotive and mobility analyst Dan Levy wrote plainly in his latest report: “Chips will become the pillar of Tesla’s next stage of growth,” and to push this growth to “mega-scale,” Tesla needs a Terafab.

Barclays regards Terafab as Tesla’s next “extreme move” in vertical integration: to build a closed loop in the US as much as possible for logic chips, memory, and advanced packaging, with the goal of covering the AI6/AI7+ chips required for Robotaxi/Optimus, as well as Dojo-related chips for data centers.

The problem is, once the scale is expanded to Tesla’s standards, the bill becomes staggering: Barclays estimates that even at a lower target, capital expenditures will need to be about $20-25 billion, while more aggressive capacity planning would raise that to $40-50 billion. Bank of America bluntly stated that the Terafab project faces a capital outlay threshold of over $60 billion, and even in the best scenario, the unit price for 2nm wafers will still be higher than TSMC, making the returns for this investment hard to justify.

Even more important is the execution path. Tesla lacks experience in large-scale chip manufacturing, and challenges in leading process nodes, tool chains, yields, and packaging are not issues that can be solved by simply “building a factory.” Compared with going it alone, Barclays prefers a “collaborative Terafab”—Tesla provides the funds to lock in capacity, while foundries handle manufacturing. Potential partners could include Samsung, TSMC, or even Intel.

What exactly does Terafab want to do? Creating a closed domestic loop for logic, memory, and packaging in the US

The report brings the timeline back to Battery Day 2020: at that time, Tesla used battery production capacity as the “strategic pivot for the next decade.” Now, in the new stage driven by Robotaxi and humanoid robots, Barclays believes the new pivot is chips—which directly determine whether inference on the vehicle side, robot side, and training/validation chains in data centers can expand.

Musk has emphasized multiple times at the November 2025 shareholder meeting and the January 2026 4Q25 conference call: in the next 3-4 years, chips (including AI logic and memory) will become the limiting factor for growth. Recently, he also mentioned on X that Terafab will launch in a few days; based on this, the report discusses Terafab’s timeframe as Tesla’s “next big move.”

Terafab’s ambition is not just to expand production, but to create a complete closed loop for “logic + memory + advanced packaging” in the US.

Logic chips: Tesla has accumulated significant design experience from HW3 to AI6, and has already set manufacturing anchors at TSMC Arizona and Samsung Texas—this is the relatively mature part among the three segments.

Packaging: Tesla has some design expertise, but limited manufacturing experience—packaging is decisive for whether logic and memory can be tightly integrated and achieve higher energy efficiency, being a key variable for the computing unit’s performance.

Memory is the biggest gap: Tesla has neither design nor manufacturing experience, and the supply of advanced memory in the US is also seriously lacking—Micron’s Idaho plant is expected to have initial output only by mid-2027, and multiple plants in New York are projected to come online only by 2030.

This means the difficulty of “full-chain localization” does not rise linearly, but grows multiplicatively.

There are two layers of strategic motivation behind Terafab that cannot be ignored. The first is geopolitical risk mitigation: potential disruptions in the Taiwan Strait and US-China trade tensions mean that relying on TSMC for the logic chip supply chain is always a sword hanging overhead, and the lack of advanced memory supply in the US further increases the appeal of “domestic end-to-end” solutions. The second reason is design control: Tesla wants to more tightly integrate logic and memory through packaging, to reduce discrete components and build a computing unit truly customized for its own software stack—the report cites management expectations that AI5’s performance could rival Nvidia Blackwell (Thor) but with only one-third of the power consumption and less than 10% of the cost.

This unified “one chip, multiple scenarios” strategy brings vehicle inference, Optimus inference, and some data center uses all onto in-house chips, making AI5/AI6 the central hub connecting Tesla’s vehicle, robot, and data center businesses.

The report estimates Tesla will purchase about 3-4 million chips in 2025. In other words, if this demand curve rises at the same slope, what suppliers need to secure for Tesla is not a “small expansion,” but a round of high-risk capital investment. Musk has mentioned the concept of 160,000 WSPM (wafer starts per month); Barclays believes, with good yields, this would equate to about 24 million chips per year.

$20-50 billion is just the starting point—the economics may not be sustainable!

The cost of the vision is a staggering bill.

The report gives two cost ranges: covering about 12 million units per year requires $20-25 billion in capital expenditure; if the target rises to 24 million units per year, then $40-50 billion is needed. For comparison: Samsung’s Taylor project announced $17 billion for about 20,000 WSPM capacity, TSMC Arizona’s total investment is reported at $165 billion, and Micron plans $100 billion in New York to build multiple memory plants. Musk’s mentioned 160,000 WSPM concept is close to “multiple megafactory scale” in the industry—a typical fab is about 20,000 to 40,000 WSPM.

The sharper issue: none of these investments are reflected in Tesla’s current 2026 capital expenditure forecast ($20 billion). Barclays’ own forecast for Tesla’s 2026 free cash flow is already -$3 billion; if fab-level investments are added, cash flow pressure would multiply several times.

For this reason, the report is highly cautious about “building independently.” Tesla lacks experience in large-scale chip manufacturing; the process expertise, EUV lead times, and advanced packaging complexities required for advanced process nodes are not things that can be resolved by just building a plant. The report even cites the previously cancelled and impaired Dojo project and the underperformance of the 4680 battery project as precedents, warning investors not to equate “design capability” directly with “manufacturing ramp-up capability.”

Bank of America’s research report from March 23 directly pointed out that the Terafab project faces a capital requirement threshold of over $60 billion, and even under the best scenario, the unit price for 2nm wafers will still be higher than TSMC, making the investment return difficult to justify.

Specifically, BofA says under assumptions of 100% utilization and top yields, the fixed cost of Terafab’s front-end wafer is around $6,000, still 1.3-1.5 times higher than TSMC’s advanced nodes. To maintain the roughly 45% gross margin required by the technical and capacity roadmap, Terafab would need to price 2nm wafers at around $32,000—higher than TSMC’s $30,000.

Even without considering technical challenges, just building the facilities would take 3-5 years: constructing the building would take about 1.5-2 years, equipment installation about 1 year, while risk production and product certification would require another 1-2 years. The report estimates that, if construction started today, Terafab could only begin mass production in 2029 at the earliest.

Almost no precedent for going it alone: the more likely scenario is “Tesla funds it, industry giants build it”

Barclays prefers the “collaborative Terafab” version: Tesla provides the funds to lock in capacity, foundries handle manufacturing, and through capital investment and underwriting of losses, suppliers are incentivized to commit to a 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 become involved in planning Samsung’s second Taylor site; TSMC is also a natural partner but does not do automotive-grade DRAM; Intel fits Musk’s preference for a US supply chain, and might have available capacity.

For Tesla’s stock, Barclays remains cautious: the grand ambitions for chip capacity can continue to fuel the long-term growth narrative, but once it comes down to capital spending and free cash flow, the market will need much firmer execution details.

Barclays maintains its Equal Weight rating for Tesla, with a target price of $360. Compared with the report’s quoted closing price of $380, this implies about 5% downside—the chip story could grow larger, but without clear executable paths and cost bounds, Terafab remains an “open question: high risk, high reward, more likely via collaboration.”

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content is from Wind Chaser Trading Desk.

For more detailed interpretation, including real-time insights, in-depth research, etc., please subscribe to [Wind Chaser Trading Desk · Annual Membership]

Risk Disclaimer and TermsMarkets have risks, investments need caution. This article does not constitute personal investment advice and has not taken into account any individual user's specific investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their particular circumstances. Invest accordingly, at your own risk. ```