Intel's headquarters recently hosted an unusual visitor: Elon Musk, the world's richest man. Intel's official X account posted photos of CEO Chen Liwu shaking hands with Musk. The company announced its participation in the Terafab project, collaborating with SpaceX, xAI, and Tesla to redefine chip manufacturing technology. Chen Liwu emphasized Musk's track record of disrupting industries, calling Terafab a paradigm shift for logic chips, memory, and packaging. The announcement caused significant industry震动, with Intel's stock rising over 4% that day.
This partnership continues Musk's chip manufacturing ambitions. Two weeks earlier, he revealed plans for Terafab—a $20-25 billion project to produce AI chips consuming 1 terawatt annually, targeting 2nm technology. Reports indicate Musk has begun recruiting senior engineers from TSMC and Samsung.
Intel now joins this high-stakes gamble. Despite U.S. government support and hundreds of billions in investment, Intel's foundry business hasn't achieved true competitiveness. Former CEO Pat Gelsinger departed amid stock and performance declines. Chen Liwu's main challenge is balancing chip manufacturing and design operations.
Why is Musk determined to challenge TSMC's near-monopoly in advanced chip manufacturing? Instead of going alone, he's partnering with Intel. However, Intel trails TSMC significantly in yield rates. TSMC's N2 process starts at 65-70% yield, while Intel's 18A begins at 55-60%. Intel's CFO stated commercial viability won't come until late 2026, with industry standards reached by 2027.
Chips represent Tesla's lifeline. The company's three core businesses—self-driving vehicles, robotaxis, and Optimus robots—all depend on inference chips. Musk warned that chip supply could become Tesla's biggest bottleneck within three to four years. He compared chip-less Optimus robots to the Tin Man without a heart from The Wizard of Oz.
Tesla's next-generation AI5 chip requires 40-50 times more power than current AI4 chips, serving Cybercab fleets, Optimus production lines, and FSD training centers. AI5's mass production has been delayed to mid-2027 due to capacity constraints. Tesla has secured agreements with both TSMC and Samsung for AI5 production, with AI6 exclusively contracted to Samsung for $16.5 billion.
Optimus robots represent the most critical variable. Musk's vision requires millions of humanoid robots annually, each needing a high-performance chip. Neither TSMC nor Samsung can dedicate such massive capacity to a single customer. To fulfill his ambitious plans, Musk must build his own chip factory.
SpaceX adds another dimension—80% of Terafab's output would target space applications, producing radiation-resistant chips for orbital AI satellites powered by space-based solar energy. Though considered unrealistic by many, this narrative supports SpaceX's projected $1.25 trillion valuation and $2 trillion IPO target. Terafab serves as the manufacturing anchor for this vision.
Musk summarizes his logic simply: "Either build Terafab or have no chips. We must have chips, so we must build Terafab." This represents both Tesla's greatest threat and most pressing reality.
Musk's strategy centers on extreme vertical integration, following the Gigafactory model that consolidated battery, motor, and vehicle production under one roof. Terafab would replicate this for chips—integrating design, lithography, memory production, advanced packaging, and mask production at Austin's Giga Texas campus.
Musk claims internal mask production can reduce design iteration cycles from 12-18 months to under 9 months. Each Terafab module would focus on single chip designs, simplifying manufacturing and improving yields. Masks function as "negatives" for circuit printing in lithography. Traditional outsourcing to foundries like TSMC creates scheduling bottlenecks when designs need modification.
By keeping mask production in-house, Musk gains absolute control over design changes. Discovering architectural flaws or needing next-generation AI chips no longer requires waiting for foundry availability. This internal testing loop dramatically compresses iteration cycles.
Traditional foundries like TSMC serve multiple clients on single production lines, frequently switching between orders from Nvidia, Apple, and AMD. Each changeover requires recalibration, causing yield fluctuations. Terafab's approach resembles single-model assembly lines—each module produces one chip type exclusively, minimizing variables and maximizing yield stability.
For Intel, joining Terafab represents redemption rather than mere collaboration. Musk's visit holds particular significance for Chen Liwu, marking his most important potential client since taking over last year. Intel's foundry business generated only $307 million in external revenue for 2025, negligible compared to TSMC's tens of billions.
Over the past year, Chen Liwu has restructured Intel through layoffs, attracted strategic investments from the U.S. government, Nvidia, and SoftBank, and repurchased an Irish fab from Apollo Capital for $14.2 billion. Improved financial performance prompted additional U.S. government funding—$3.5 billion beyond the original $8.5 billion subsidy—specifically supporting foundry operations.
Intel brings its 18A process node to Terafab—a 1.8nm-class technology featuring RibbonFET transistors and PowerVia backside power delivery. Fab 52 in Arizona is fully operational, with Panther Lake CPUs shipping since late 2025 and Clearwater Forest server chips scheduled for 2026. However, yield rates remain below industry standards, not expected to reach optimal levels until 2027. Meanwhile, some capacity sits idle.
This represents the most advanced chip technology currently manufactured entirely in the U.S., potentially supplying Tesla's AI chips and SpaceX satellite chips. Intel's packaging technologies—EMIB and Foveros—also fit Terafab's vertical integration strategy. Rather than developing technology from scratch, Musk pragmatically leverages Intel's existing capabilities.
Some analysts view Intel's involvement as revealing Terafab's true nature—a capacity procurement agreement wrapped in technological ambition. Notably, the partnership announcement lacked detailed production commitments, timelines, or legal binding documents.
Intel's participation makes Musk's challenge more concrete: How does U.S. chip manufacturing compare to industry leader TSMC? Technologically, Intel's 18A entered production in mid-2025, about six months ahead of TSMC's N2. However, transistor density tells a different story—18A achieves 238 million transistors per square millimeter, while N2 reaches 313 million, a 30% gap.
Yield rates show even greater disparity. TSMC's N2 starts at 65-70%, while Intel's 18A begins at 55-60%. Industry profitability typically requires 80% yields, with TSMC's mature processes often exceeding 90%. Intel's current rate means nearly one in two wafers becomes scrap, creating enormous cost pressure.
Capacity differences are equally stark. TSMC shipped approximately 15 million 300mm equivalent wafers in 2025, backed by 50+ fab modules accumulated over decades. Intel's Fab 52 currently produces about 40,000 wafers monthly—an order of magnitude smaller. Samsung's 2nm production remains in risk production phase with approximately 40% yield.
According to Taiwanese reports, TSMC's 3nm and 2nm capacity is fully booked through 2028, prioritizing AI-related orders from Nvidia, AMD, Broadcom, and Marvell over longstanding clients like Apple and MediaTek. This creates extreme shortages for non-AI chip designers, forcing them to secure capacity years in advance, accept delays, or turn to Samsung—currently holding only 7% market share compared to TSMC's 72%.
If Intel can improve 18A yields promptly, it could capture market share during this severe supply shortage. However, the most critical gap remains TSMC's customer ecosystem. Apple, Nvidia, AMD, and Qualcomm have deep, decades-long relationships with TSMC, with process IP, design kits, and yield improvement experience embedded within this network.
While Intel has external clients like Microsoft's Maia 2, its foundry business remains fractional compared to TSMC's. This gap can't be closed quickly with money or equipment—it requires time and trust accumulation. Compounding the issue, Intel's own chip design business creates concerns among potential competitors.
Harvard Business School professor David Yoffie highlighted the core conflict: "If you're Nvidia, AMD, or Qualcomm, would you really trust your core designs to Intel—your competitor?"
Why is Musk determined to tackle challenges where Intel previously failed? Perhaps Intel's struggles stem from the cultural crisis of an old-school IDM (Integrated Device Manufacturer). During its peak, Intel dominated the PC era through internal design and manufacturing. But it missed key transitions to mobile chips and foundry models, particularly with its delayed 10nm node, allowing TSMC to build structural advantages in yield and scale.
Despite multiple CEO changes, Intel struggled against bureaucratic inertia. Pat Gelsinger's IDM 2.0 vision, after repeated delays and hundreds of billions in investment, left behind an 18A process still climbing toward acceptable yields and a foundry business with minimal external clients.
Musk's confidence comes from his history of redesigning manufacturing systems as a disruptor. Whether building rockets at SpaceX or battery factories at Tesla, he rejects existing industry cost structures and engineering boundaries, applying first principles reasoning.
Critically, Terafab isn't a pure-play foundry. It serves only Musk's companies—Tesla, SpaceX, and xAI—meaning it doesn't need to beat TSMC in market competition, only produce chips more cheaply and quickly than external sourcing. This goal is far more realistic than building a commercial foundry, similar to how Gigafactory supports Tesla's expansion without becoming the world's largest battery supplier.
Still, Musk's targets remain extremely aggressive. Bernstein analysis suggests achieving 1 terawatt annual production would require 142-358 modern fabs at approximately $5 trillion—exceeding 70% of the U.S. federal budget. TSMC's current output required two decades and about 50 fab modules to achieve. Terafab aims to more than double this capacity from scratch.
Musk's announced $20-25 billion initial investment, according to Bernstein's model, would barely cover one 7nm fab at 2018 technology levels. This may explain his urgency in partnering with Intel.
Capital represents only part of the challenge. EUV lithography machines face physical supply constraints—ASML produces only 50-60 annually, with all chip manufacturers competing for them. Terafab would struggle to secure equipment against established players like TSMC, Samsung, and Intel.
The handshake between Musk and Chen Liwu reflects broader anxiety in U.S. semiconductor manufacturing: TSMC virtually monopolizes global advanced chip production capacity. Intel spent years and hundreds of billions to restore partial advanced manufacturing capability in the U.S. Musk believes faster iteration, aggressive vertical integration, and massive demand can create a different path.
This isn't merely an engineering plan—it's a long-term bet on U.S. chip sovereignty. Intel's 18A process demonstrates that near-world-class chips can be manufactured domestically, but scale, yield, and customer ecosystems remain far behind TSMC, lacking commercial competitiveness.
Musk needs Intel's manufacturing experience; Intel needs Musk's demand scale and narrative power. Both companies understand each other's weaknesses and their own bargaining chips.
The chip industry follows one iron law: nothing proves more costly than underestimating difficulty. Intel learned this when transitioning from planar transistors to FinFETs, surrendering process leadership to TSMC and spending nearly a decade realizing the gap had become unbridgeable.
But equally true: Musk has repeatedly achieved the "impossible." SpaceX was initially considered unrealistic, yet Musk overcame challenges through iteration speed and vertical integration.
In advanced nodes, manufacturing yields, and customer ecosystems, TSMC stands as an imposing wall that makes competitors despair. Yet the rocket-building Musk believes his vertical integration model can颠覆行业定律.
Comments