The automaker with zero fabrication experience targets the world's most advanced semiconductor process
March 21st, 2026: Tesla formally launches TeraFab projectNew here? Follow stories to track developments over time. Create a free account to get updates when stories you care about change.
Every company designing custom artificial intelligence chips today — Apple, Google, Amazon, Microsoft — pays someone else to manufacture them. Tesla just announced it will build and operate its own semiconductor fabrication plant, a $20 billion facility called TeraFab targeting the 2-nanometer process node, the most advanced manufacturing technology in existence. No company without decades of chipmaking experience has ever attempted this.
The stakes are enormous and asymmetric. If TeraFab works, Tesla secures an independent supply of custom silicon for its self-driving cars, Optimus humanoid robots, and AI training infrastructure — breaking its dependence on Taiwan-based TSMC and Samsung at a moment when chip supply constraints loom within three to four years. If it fails, Tesla will have burned through a significant fraction of its cash reserves on a capability that took Intel, Samsung, and TSMC generations of engineers to develop. Intel's foundry division, backed by fifty years of fabrication expertise, lost $13.4 billion in 2024 alone.
Why it matters
If a carmaker can build cutting-edge chips, it reshapes who controls the AI hardware supply chain.
Elon Musk, Tesla Factory, Fremont (CA, USA) (8765031426)
Tesla factory with parked cars during sunset, showcasing modern automotive industry vibes.
From above of circuit boards of modern smartphones placed in plastic box in electronics factory
Elon Musk Reveals Tesla TeraFab Chip Factory
The Tesla Space
Elon Musk announces Tesla's AI chip factory and bets on ...
TIMES BRASIL - LICENCIADO EXCLUSIVO CNBC
Elon Musk Building "Tera Fab" to Produce AI Chips in US ...
Eli the Computer Guy
Images from Openverse under Creative Commons licenses. Videos from YouTube.
Exploring all sides of a story is often best achieved with Play.
Fictional AI pastiche — not real quote.
"A man who dares to build the very furnace that forges his steel, rather than beg it from another's hearth, has already won half the battle in his own mind — and that audacity, gentlemen, is precisely the sort of vertical integration I staked my own fortune upon in Pittsburgh."
Fictional AI pastiche — not real quote.
"A twenty-billion-dollar wager on vertical integration is either the boldest act of industrial statesmanship since Carnegie built his own steel mills, or the most expensive lesson in the difference between vision and expertise. One notes with some irony that the gentlemen at Intel, possessed of half a century's hard-won knowledge, managed to lose thirteen billions in a single year — a figure that ought to concentrate Mr. Musk's mind most wonderfully."
Your progress in this debate will be lost.
Choose one persona for each side of the debate
Select debater for this side:
No debate personas available right now.
Select debater for this side:
No debate personas available right now.
Make your prediction before the referee scores
Pick the question both personas must answer in the final round
Debate Oracle! You called every round!
Sharp Instincts! You know your debaters!
The Coin Flip Strategist! Perfectly balanced!
The Contrarian! Bold predictions!
Inverse Genius! Try betting the opposite next time!
Can you match the quotes to the right people?
— Who said this?
Tip: press 1– to answer.
Tesla designs electric vehicles, energy storage systems, and AI-powered autonomous driving software, and is now attempting to add semiconductor fabrication to its vertically integrated manufacturing model.
TSMC manufactures roughly 56% of the world's contract chips, including processors for Apple, Nvidia, AMD, and Qualcomm, and is the only company currently producing 2-nanometer chips at scale.
Samsung fabricates Tesla's current autopilot and Full Self-Driving chips and signed a $16.5 billion contract in April 2025 to manufacture Tesla's next-generation AI6 chip at its Taylor, Texas facility.
Intel's foundry division lost $13.4 billion in 2024 despite fifty years of fabrication experience, making it the most relevant cautionary example for Tesla's TeraFab ambitions.
Tesla officially launched its TeraFab semiconductor fabrication initiative, a $20 billion project to build a 2nm chip fab at the North Campus of Gigafactory Texas in Austin. The facility aims to produce 100 to 200 billion custom AI and memory chips annually for autonomous vehicles, Optimus robots, and AI infrastructure.
Musk posted on X: "Terafab Project launches in 7 days," generating intense media coverage and speculation about the scope of the announcement.
Musk formally confirmed the TeraFab semiconductor fabrication project during Tesla's earnings call, stating "We will break the chip wall" and projecting a supply constraint within three to four years without in-house manufacturing. Tesla's 2026 capital expenditure plan exceeded $20 billion.
At a shareholder event, Musk said Tesla must build its own semiconductor fabrication facility to avoid supply constraints and publicly floated a partnership with Intel, though no deal was signed.
Tesla killed its Dojo custom supercomputer chip program and Pete Bannon, the vice president who had led all custom silicon development since 2016, left the company. The AI6 chip was repositioned to replace Dojo's training role.
Tesla committed $16.5 billion to have Samsung manufacture its next-generation AI6 chip at Samsung's Taylor, Texas fabrication facility using 2nm gate-all-around transistor technology.
The next-generation HW4 computer began shipping in Tesla vehicles with an improved chip on Samsung's 7nm process, featuring more compute cores and enhanced neural network performance.
Tesla's HW3 computer, powered by its custom Full Self-Driving chip fabricated by Samsung on a 14-nanometer process, began shipping in Model S, 3, X, and Y vehicles. The chip delivered 72 trillion operations per second.
Tesla recruited legendary chip architect Jim Keller and Apple veteran Pete Bannon to build an in-house semiconductor design team, signaling dissatisfaction with off-the-shelf Nvidia hardware for autonomous driving.
Predict which scenario wins. Contrarian picks score more — points lock in when the scenario resolves.
Tesla successfully constructs and ramps TeraFab over four to six years, achieving competitive yields on 2nm chips. The facility supplies custom silicon for Tesla's vehicle fleet, Optimus robots, and AI training clusters, eliminating dependence on TSMC and Samsung. This would validate Musk's vertical integration philosophy and potentially open Tesla to selling excess capacity as a contract foundry. The scenario requires Tesla to recruit thousands of experienced fab engineers, master extreme ultraviolet lithography, and sustain $20 billion or more in capital investment through a period of negative free cash flow.
Facing the reality that building a leading-edge fab from scratch takes a decade or more and requires expertise Tesla does not have, TeraFab is quietly rescoped. Instead of manufacturing chips from raw silicon wafers, the facility focuses on advanced packaging — combining chips fabricated by Samsung and TSMC into custom multi-chip modules. This would still provide meaningful vertical integration benefits at a fraction of the cost and technical risk, following a path similar to what Apple has pursued with TSMC's advanced packaging services.
Rather than building fabrication capability from zero, Tesla strikes a deep partnership with or acquires a struggling fab operator — Intel's foundry division being the most frequently discussed candidate. Musk's November 2025 comments about Intel suggest this path has been explored. An acquisition would give Tesla instant access to process engineers, existing clean rooms, and institutional manufacturing knowledge, dramatically accelerating the timeline while adding enormous financial and integration risk.
TeraFab follows the trajectory of Tesla's 4680 battery cell program and original Dojo supercomputer — bold announcements followed by years of delays, technical setbacks, and results that fall well short of initial claims. Construction stretches years beyond schedule, yields prove uneconomical, and Tesla continues relying on Samsung and TSMC for chip supply. The capital expenditure generates negative returns and contributes to Tesla's projected negative free cash flow, potentially triggering a capital raise that dilutes existing shareholders.
Intel CEO Pat Gelsinger announced IDM 2.0 in March 2021, committing over $100 billion to rebuild Intel's manufacturing leadership and open its fabs to outside customers. Despite five decades of fabrication expertise and billions in CHIPS Act subsidies, Intel's foundry division posted a $13.4 billion operating loss in 2024. Yield rates on advanced nodes lagged TSMC by 20 percentage points or more.
Gelsinger was ousted in late 2024. Intel's stock lost more than half its value. The company began exploring a separation of its foundry and design businesses.
Intel's experience demonstrated that even established chipmakers with massive resources struggle to compete with TSMC on advanced process nodes, raising the bar for any new entrant.
If Intel — with 50 years of fab experience, thousands of process engineers, and billions in government subsidies — lost $13 billion in a single year trying to catch TSMC, Tesla's prospects as a company with zero fabrication experience are the central question for TeraFab's viability.
At Tesla's September 2020 Battery Day, Musk announced the 4680 battery cell format, promising a 56% cost reduction and five times the energy output. Tesla built dedicated production lines at Giga Texas and Giga Nevada to manufacture cells in-house rather than relying solely on suppliers like Panasonic and CATL. Production consistently fell short of targets, with yield and quality issues persisting for years.
The Cybertruck launched in late 2023 using 4680 cells, but production volumes were limited. Tesla continued relying heavily on external cell suppliers.
By 2025, Tesla had improved 4680 yields but never achieved the cost and performance breakthroughs promised at Battery Day. The program illustrated both Tesla's willingness to attempt difficult vertical integration and its tendency to underestimate manufacturing timelines.
TeraFab skeptics, led by Electrek, explicitly compare the project to the 4680 program: ambitious vertical integration announcements followed by years of delays and results falling short of promises. Semiconductor fabrication is arguably orders of magnitude more complex than battery cell production.
Apple began designing custom chips in 2008, shipping the A4 processor in the original iPad in 2010. Over the next fifteen years, Apple expanded custom silicon across every product category — A-series for iPhone, M-series for Mac, and specialized chips for AirPods, Apple Watch, and security. Apple invested in chip design talent rather than fabrication, relying entirely on TSMC to manufacture its processors. The partnership grew so deep that Apple effectively funded TSMC's advance to leading-edge nodes.
Apple's M1 chip, launched in 2020, demonstrated that custom silicon could deliver dramatic performance and efficiency gains over commodity processors.
Apple proved that world-class chip capability does not require owning a fab. The fabless model let Apple capture silicon's strategic benefits — performance differentiation, power efficiency, hardware-software integration — without the capital intensity and technical risk of manufacturing.
Apple's success raises the question of whether Tesla's decision to build its own fab is strategically necessary or a case of vertical integration past the point of diminishing returns. Tesla could potentially achieve most of TeraFab's goals through deeper partnerships with TSMC and Samsung, as Apple does, without the $20 billion fabrication gamble.