Elon Musk, the world's wealthiest individual and CEO of Tesla, unveiled an ambitious plan to enter the chip manufacturing sector last Saturday, describing it in grand terms. His "Terafab" vision highlights the severe shortage and production capacity crisis in artificial intelligence chips. Musk announced his intention to venture into high-end semiconductor manufacturing, calling it the "most epic chip manufacturing project in history" and naming it "Terafab." He appears to be attempting to transform the "AI computing power shortage" into an extraordinary industrial gamble that could reshape the semiconductor supply landscape. Musk's goal is not merely to produce higher-performance chips but to fundamentally alter the supply and demand dynamics of computing power for the era of AI, robotics, and space data centers through unprecedented vertical integration. However, such a monumental, era-defining vision also exposes the harsh realities of global advanced semiconductor capacity shortages and the extremely high barriers to manufacturing.

"I have an important announcement to make; this will be the most epic human chip-making endeavor in history so far," he stated publicly before a small group in Austin, Texas. "This will really take things to the next level, a level people probably haven't even conceived of yet. We are going to raise the entire benchmark parameters by several orders of magnitude here." No one has previously proposed anything truly comparable to Musk's so-called "Terafab" concept. Based on the disclosed plan, it would be an immensely large-scale project aimed at producing the most advanced semiconductors for AI training/inference systems, humanoid robots, and frontier space exploration. He aims not only to challenge the world's premier chip manufacturers, like the foundry giant Taiwan Semiconductor Manufacturing (TSM), but also to achieve this goal on a scale far exceeding current industry capacity. The projected scale has left investors astonished.

According to a forecast from Bernstein analysts, the "Terafab" project could require capital expenditure of approximately $5 trillion to $13 trillion. This funding would be used to build 140 to 360 new factories, each producing 50,000 wafers per month, to achieve Musk's ambitious target of 1 terawatt of comprehensive computing power annually. As the world's richest person, Musk has previously accomplished feats deemed impossible—creating a commercially viable high-frequency rocket launch business with SpaceX, mainstreaming electric vehicles with Tesla Inc., and providing internet infrastructure from space with Starlink. However, some doubt whether Musk can, or even intends to, build everything he outlined in Austin as the "most epic" chip-making endeavor.

NVIDIA's CEO Jensen Huang, at the GTC conference, showcased an "unprecedented AI computing revenue super blueprint." He informed global investors that, driven by strong demand for Blackwell architecture GPUs and the upcoming Vera Rubin architecture, AI chip supply is lagging far behind sustained robust demand. Future revenue in the AI chip sector could reach at least $1 trillion by 2027, significantly higher than the previous $500 billion AI infrastructure blueprint for 2026. As model scale, inference chains, and multimodal/agentic AI workloads push computing power consumption to expand exponentially, tech giants' capital expenditure is increasingly focused on AI computing infrastructure. Global investors continue to anchor the "AI bull market narrative"—centered on NVIDIA, Google's TPU clusters, and AMD's new product iterations and AI cluster deliveries—as one of the most certain growth investment stories in global stock markets. This also means investment themes closely related to AI training/inference, such as power, liquid cooling systems, and optical interconnect supply chains, will remain among the hottest investment sectors, following leaders like NVIDIA, AMD, Broadcom, Taiwan Semiconductor Manufacturing, and Micron, even amid geopolitical uncertainties in the Middle East.

According to Wall Street firms like Morgan Stanley, Citi, Loop Capital, and Wedbush, the global AI infrastructure investment wave, centered on AI computing hardware, is far from over; it is only at the beginning. Driven by an unprecedented "storm of AI inference-side computing demand," this wave of global AI infrastructure investment, lasting until 2030, could reach a scale of $3 trillion to $4 trillion. Is there an ulterior motive? "A true Terafab, in our view, feels like a stretch," wrote Bernstein analysts, including Stacy Rasgon. The required computing power would "be equivalent to the entire scale of currently installed global semiconductor capacity and would actually require several times the current installed 'relevant' semiconductor capacity." Patrick Moorhead, a senior analyst at Moor Insights & Strategy, suggested there is a significant possibility Musk will not ultimately build chip fabrication facilities. Instead, the goal of announcing Terafab might be different: to emphasize the worsening global chip capacity shortage or to incentivize chipmakers, especially large U.S. fabs, to accelerate expansion. Alternatively, it could boost external perceptions of SpaceX's future influence and growth prospects as the company approaches what could be the largest IPO later this year.

There is growing concern in Silicon Valley that the semiconductor industry's capacity expansion pace is insufficient to produce the core AI chips needed by AI companies to realize their ambitious commercial plans. Amazon, Alphabet Inc., and other hyperscale cloud providers are projected to spend roughly $650 billion to $700 billion this year alone to expand or build new large-scale AI data center infrastructure. This has already caused severe shortages in NVIDIA AI chips, various self-developed AI ASICs, memory chips related to HBM and eSSD, and is now spilling over to data center CPUs. NVIDIA's founder and CEO Jensen Huang has expressed his hope that Taiwan Semiconductor Manufacturing—the chipmaking giant that produces nearly all of NVIDIA's advanced semiconductors—will provide more capacity. However, this key pillar of global chip manufacturing has maintained a cautious approach to expansion.

Taiwan Semiconductor Manufacturing's CEO C.C. Wei told Wall Street analysts on an October earnings call, "We are receiving very strong signals directly from our customers' customers, demanding more capacity to support their AI computing infrastructure-related businesses." He added, "We will also continue to maintain strict discipline and thoroughness in our capacity planning methodology to ensure a stronger profitability growth curve for our shareholders." Musk explicitly stated that Tesla, xAI, and SpaceX will require a staggering number of silicon chips in the coming years. His ultimate goal is to possess 1 terawatt of computing power for AI supercomputing systems and to produce hundreds of millions of new Optimus AI humanoid robots annually. He identified the "key missing element" as computing power, estimating that current AI output is only about 2% of what his companies need. He emphasized that he has urged key suppliers, including Taiwan Semiconductor Manufacturing, Samsung Electronics Co., and U.S. memory chip giant Micron Technology Inc., to expand capacity as quickly as possible—promising to buy almost everything they produce. "We will buy every chip they make; that's exactly what I told them," he said.

Recent statements from NVIDIA's Jensen Huang, urging memory chip makers that NVIDIA will buy all the HBM capacity they can produce, and AMD CEO Lisa Su's plans to partner with Samsung to secure large HBM supplies, highlight the exponential surge in demand for memory chips, including HBM, DRAM, and NAND. This is driving accelerated expansion by the three major memory chip manufacturers: Micron, SK Hynix, and Samsung Electronics. However, semiconductor manufacturing companies have generally chosen to expand capacity at a more restrained pace, far below the level Musk deems necessary for Tesla, SpaceX, and xAI. "So, either we build Terafab, or we don't have the chips we desperately need," he stated. "And we need those chips now."

Analysts are skeptical not only because Musk has never manufactured chips before but also because his proposal defies the economics of the chip industry. Building factories to meet one's own demand was the model during the industry's infancy. However, due to the extreme costs—a state-of-the-art fab costs about $30 billion and can become obsolete within just five years—such fixed semiconductor infrastructure costs are only justified by producing and selling massive volumes of chips over the long term. A team from Barclays Capital believes there is logic in Musk's recognition of supply shortage risks and their potential impact on his ambitions for electric vehicles, AI robots, and data center AI infrastructure. However, as the Barclays analysts wrote in a research note, "Yet, as is often the case with many Tesla endeavors, this attempt is unprecedented in multiple respects. Tesla severely lacks direct chip manufacturing experience and faces significant cost and execution risks. Therefore, deeper collaboration with Samsung or Taiwan Semiconductor Manufacturing, or even Intel, might be a more likely path."

It is noteworthy that even a company as powerful as NVIDIA, the "AI chip superpower" with a market capitalization of approximately $4.3 trillion, is a major tech company that has never directly owned chip fabrication plants. Most of its peers are in the same situation. They primarily outsource almost all chip production to Taiwan Semiconductor Manufacturing and, to a lesser extent, Samsung Electronics. These large chip companies, led by NVIDIA, make this economic model work by essentially consolidating a large portion of the industry's business into chip orders vast enough to cover capital expenditure and R&D costs. Musk recently proposed that Terafab would adopt the Integrated Device Manufacturer model, handling both chip design and manufacturing internally. This is the classic model that enabled U.S. chip giant Intel to dominate for decades. However, that company has long since lost its former vigor—largely due to strategic missteps highlighting the enormous costs and unpredictable complexities inherent in chip manufacturing. IDM model companies possess the three core capabilities of chip design, manufacturing, and packaging/testing, representing the highest technical barriers, largest capital consumption, and highest risk factors in the global chip supply chain. Samsung Electronics and the former Intel are the world's top IDM chip companies. Taiwan Semiconductor Manufacturing focuses on the core manufacturing环节, belonging to the Foundry model in the chip supply chain. Since Morris Chang founded TSMC in 1987 and pioneered the pure-play foundry business model, allowing tech companies to focus on fabless chip design, the in-house chip-making model has declined. Today, the Taiwan-based chipmaking giant is the preferred foundry for numerous U.S. tech behemoths like NVIDIA, Apple, AMD, Broadcom, Microsoft, and Google.

Beyond a different business model, Musk also seems to envision a manufacturing configuration vastly different from those used by Taiwan Semiconductor Manufacturing and Samsung Electronics. "Modern fabs are doing cleanrooms wrong. I'll bet right here that Tesla will have a state-of-the-art large fab at the 2nm level, and I'll be able to eat a cheeseburger and smoke a cigar inside it," he said during a podcast in January. Ultra-clean cleanrooms are critical for modern chip production. A single transistor is many times smaller than a virus. Just a speck of dust can cause significant damage, leading to millions of dollars in wasted wafers. In last Saturday's presentation, Musk stated that the Austin mega-fab would co-locate logic semiconductors, memory chips, packaging, testing, and lithography mask production equipment within the same chip manufacturing building. No existing chip manufacturing company operates such an integrated model. Due to decades of specialized分工 and differences in advanced manufacturing processes across the global chip supply chain, this is typically economically unfeasible in the "silicon economy."

Simultaneously, Musk appears to be framing his new chip venture in terms of U.S. national interests. He reposted a message on X platform stating that Terafab is crucial for national security because a large portion of global chips are still manufactured in Taiwan and South Korea. Of course, Musk's semiconductor ambitions should be viewed in the context of his series of bold构想 over the years: helping humanity colonize Mars; transporting people from New York to Washington via sealed tunnels; reusing space rockets. At least one of these once seemingly distant dreams has now become reality.

Musk's grand vision resonates within the industry because it highlights one of its most pressing concerns: the urgent need for enterprises and governments to deploy disruptive AI super-systems, coupled with a severe shortage of high-end semiconductor capacity. The Bernstein analyst team wrote that Musk "has indeed accomplished more than one grand thing initially deemed impossible by skeptics." "Perhaps Elon has some even more bizarre, far-fetched way to improve the current situation where the supply of core AI computing infrastructure—like AI chips, high-performance memory chips, and data center CPUs—is lagging far behind demand. We leave that for readers to ponder, as we don't know what that might be either." The essence of Terafab is not an ordinary "Tesla expansion" story but Musk defining the "AI chip shortage" as the core constraint for the next phase of his empire's expansion. Musk's assessment is that the current global growth rate of chip supply cannot keep up with his demand curve for robotics, autonomous driving, AI training/inference, and space computing data centers. From an industrial logic perspective, Tesla's long-term growth narrative is shifting from electric vehicle manufacturing to becoming a "physical AI-level super platform company." Once the growth engines become Robotaxi, Optimus, on-vehicle inference, robot inference, and the xAI/SpaceX space training and deployment compute chain, the bottleneck shifts away from primarily batteries and vehicles to high-end logic chips, advanced packaging, memory, and supply chain security under geopolitics.

Musk has explicitly stated he has urged Taiwan Semiconductor Manufacturing, Samsung, and Micron to expand capacity quickly, even promising to "buy all the chips they produce," yet global advanced capacity remains tight. Concurrently, Taiwan Semiconductor Manufacturing itself has repeatedly emphasized disciplined capacity planning, and chip companies like NVIDIA and Broadcom have recently publicly acknowledged that incredibly strong AI chip demand is pushing the key data center infrastructure capacity of TSMC and other chipmakers to its limits. In other words, the real driver behind Terafab is Musk's unwillingness to continue entrusting the next phase's growth-critical path to the capacity节奏 of external foundries. In the view of analysts from Wall Street giants like Barclays and Bernstein, the most likely path for "Terafab" construction is not Tesla single-handedly building a mature advanced process system from scratch. Instead, it would involve driving Samsung, Taiwan Semiconductor Manufacturing, or potentially even Intel's capacity—through prepayments, capacity locking, shared capital expenditure, or deeper joint development—to accelerate the establishment of dedicated supply for Tesla/xAI/SpaceX. This speculation is not mere fantasy—Tesla's next-generation AI6 chip is reportedly being developed for manufacturing using Samsung's 2nm process, targeting mass production in the second half of 2027. Meanwhile, Musk, while announcing the Terafab ambition, continues to reinforce the narrative of "U.S. domestic chip security" and the SpaceX future capital market story. This indicates that Terafab is both a manufacturing strategy and a form of "pressure-signaling engineering" aimed at suppliers, Wall Street, Silicon Valley financiers, and policymakers.

The Terafab ambition further validates that the AI computing infrastructure shortage is not short-term noise but a new super cycle of capital expenditure that will continue to spill over into logic, memory, advanced packaging, semiconductor equipment, core semiconductor materials, U.S. domestic manufacturing, and even space AI computing infrastructure. Consequently, for themes like chip manufacturing foundries, HBM/DRAM/eSSD, advanced packaging, the semiconductor equipment chain, and certain U.S. high-end manufacturing sectors, it represents a powerful medium-term catalyst.