Tesla's growth narrative is undergoing a fundamental shift. While batteries and vehicles remain its cash flow sources, the next phase of its ambition is pinned on physical AI—encompassing Robotaxi, the Optimus humanoid robot, and shared AI models and computing stacks for both vehicles and robots. Consequently, the primary bottleneck is shifting: it is no longer battery production capacity, but rather the supply of chips, manufacturing capability, and supply chain security amid geopolitical tensions.
According to analysis, Barclays US Auto & Mobility analyst Dan Levy stated plainly in a recent report: "Chips will become the pillar of Tesla's next growth phase." To push this growth to a "mega-scale," Tesla requires a Terafab facility.
Barclays views Terafab as Tesla's ultimate move in vertical integration: creating a largely closed-loop system within the United States for logic chips, memory, and advanced packaging. The goal is to supply the future AI6/AI7+ chips required for Robotaxi and Optimus, as well as chips related to the Dojo supercomputer for data centers.
However, scaling to Tesla's envisioned levels brings staggering costs. Barclays estimates capital expenditures of approximately $20 billion to $25 billion for a lower target, while more aggressive capacity plans could reach $40 billion to $50 billion. Bank of America has stated directly that the Terafab project faces a capital expenditure hurdle exceeding $60 billion. Even under the most ideal scenario, its pricing for 2nm wafers would likely remain higher than those from TSMC, making it difficult to rationalize the investment returns.
The execution path is even more critical. Tesla lacks experience in large-scale semiconductor manufacturing. Challenges related to leading-edge process technology, toolchains, yield rates, and packaging are not solved simply by "building a factory." Compared to going it alone, Barclays favors a "collaborative Terafab" model—where Tesla provides funding to secure capacity, and an established foundry handles manufacturing. Potential partners include Samsung, TSMC, and even Intel.
So what exactly would Terafab do? Create a US-based closed loop for logic, memory, and packaging.
The report draws a parallel to Tesla's 2020 Battery Day, where battery capacity was positioned as the strategic pivot for the following decade. In the new phase driven by Robotaxi and humanoid robots, Barclays believes the new pivot is chips—they directly determine the scalability of in-vehicle inference, robot inference, and the data center training/validation pipeline.
Elon Musk emphasized during the November 2025 Annual Shareholder Meeting and the Q4 2025 earnings call in January 2026 that chips (including AI logic and memory) would become a limiting factor for growth over the next 3-4 years. His recent mention on X that Terafab would launch "within days" leads the report to discuss Terafab as the imminent "next big move."
Terafab's ambition extends beyond mere expansion; it aims to create a complete domestic US loop for "logic + memory + advanced packaging."
In logic chips, Tesla has accumulated significant design experience from HW3 to AI6 and has already established manufacturing anchors with TSMC in Arizona and Samsung in Texas, making this the most mature segment of the three.
Regarding packaging, Tesla has design-level expertise but limited manufacturing experience. Yet, packaging is crucial for tightly coupling logic and memory to achieve higher energy efficiency, making it a key variable for the performance of the entire compute unit.
Memory represents the largest gap: Tesla has neither design nor manufacturing experience here. Furthermore, the US supply of advanced memory is severely lacking—Micron's Idaho facility isn't expected to have initial output until mid-2027, and its multiple New York plants are projected for 2030.
This makes the difficulty of achieving "full-chain localization" not just linear, but multiplicative.
Behind Terafab lie two strategic motives that cannot be ignored. First is geopolitical de-risking: potential disruptions in the Taiwan Strait and ongoing US-China trade tensions keep the supply chain for logic chips dependent on TSMC under a cloud of uncertainty. The absence of advanced memory supply in the US further increases the appeal of full-chain localization. Second is design control: Tesla aims to tightly couple logic and memory through packaging, reducing discrete components to create compute units truly customized for its software stack. The report cites management expectations that the AI5 chip's performance could rival Nvidia's Blackwell (Thor), but with one-third the power consumption and a cost over 10% lower.
This unified strategy of "one chip family, multiple applications" converges in-vehicle inference, Optimus inference, and some data center uses onto Tesla's own chips, making AI5/AI6 the hub connecting its automotive, robotics, and data center business lines.
The report estimates Tesla procured approximately 3 to 4 million chips in 2025. Therefore, if demand continues rising at this rate, suppliers would need to commit not to a "minor expansion" but to a high-risk capital investment. Musk has mentioned a vision of 160,000 wafer starts per month (WSPM), which Barclays estimates could translate to about 24 million chips annually under good yield conditions.
$20-$50 billion is just the starting point; it may be economically unsustainable.
The cost of this vision is a daunting bill. The report outlines two cost tiers: covering roughly 12 million chips per year requires about $20-$25 billion in capital expenditure; scaling the target to 24 million chips annually raises the requirement to $40-$50 billion. For context, Samsung's Taylor project announced $17 billion for approximately 20,000 WSPM capacity; TSMC's total investment in Arizona is disclosed at up to $165 billion; Micron plans $100 billion for multiple memory fabs in New York. Musk's mentioned 160,000 WSPM vision approaches the scale of "multiple large fabs combined"—a typical fab ranges from 20,000 to 40,000 WSPM.
A more critical issue is that these investments are not reflected in Tesla's existing 2026 capital expenditure guidance of $20 billion. Barclays' own forecast for Tesla's 2026 free cash flow is already negative $3 billion; adding fab-level investments would multiply the cash flow pressure significantly.
Consequently, the report is highly skeptical of Tesla building a fab independently. Tesla's lack of large-scale chip manufacturing experience, the process expertise required for leading-edge nodes, the cycle times for EUV equipment, and the complexity of advanced packaging are not challenges solved merely by constructing a building. The report even cites precedents like the Dojo project being canceled and written down, and the 4680 battery's production volume and performance falling short of expectations, cautioning the market not to equate "design capability" directly with "manufacturing ramp-up capability."
Bank of America's report from March 23rd stated plainly that the Terafab project faces a capital expenditure threshold exceeding $60 billion. Even under the most ideal assumptions, its 2nm wafer pricing would remain higher than TSMC's, making investment returns difficult to justify.
Specifically, BofA indicated that under optimal assumptions of 100% utilization and yield, Terafab's front-end wafer fixed cost would be approximately $6,000, still 1.3 to 1.5 times higher than TSMC's advanced nodes. To maintain the roughly 45% gross margin needed to fund its technology and capacity roadmap, Terafab would have to price its 2nm wafers at around $32,000, higher than TSMC's approximate $30,000.
Even setting technical challenges aside, the facility construction alone would take 3 to 5 years: about 1.5-2 years for the building, roughly 1 year for tool installation, and 1-2 years for risk production and product qualification. The report estimates that if construction started today, Terafab would not achieve volume production until 2029 at the earliest.
Going solo is virtually without precedent; a more likely version is "Tesla provides funding, a giant handles manufacturing."
Barclays leans towards a "collaborative Terafab" model: Tesla provides capital to secure capacity, while a foundry handles manufacturing. This would involve capital contributions, guaranteed offtake, or loss guarantees to incentivize suppliers to commit to more aggressive production ramps.
Among potential partners, Samsung is seen as the most natural fit—covering logic, packaging, and DRAM. The report even suggests Tesla might partially influence the planning of Samsung's second fab in Taylor. TSMC is also a natural partner, though it does not produce automotive-grade DRAM. Intel aligns with Musk's preference for US supply chains and might have available capacity.
Regarding the stock price, Barclays advises caution: grand chip capacity goals can fuel the long-term growth narrative, but once it comes down to capital expenditure and free cash flow, the market will require more concrete execution details.
Barclays maintains an Equal Weight rating on Tesla with a $360 price target, implying approximately 5% downside from the report's referenced closing price of $380. The chip story can be expansive, but until a viable execution path and clear cost boundaries are established, Terafab remains an open proposition characterized as "high-risk, high-reward, and more likely collaborative."
Comments