The argument of this essay is narrow. The West is rebuilding chip fabrication on its own soil. The build-out is real, partly funded, partly producing, partly still a press release. It is happening along three distinct threads, and a serious account of any one of them has to keep the other two in view. The first is the Terafab project that Elon Musk announced in Austin on 21 March 2026 and that Intel joined in April. The second is the wave of Asian-headquartered chipmakers — TSMC, Samsung, SK Hynix, and the supporting OSAT and equipment companies — who are building factories inside the United States for the first time at scale. The third is the new American quantum foundries: a standalone subsidiary inside IBM, a multimodality foundry inside GlobalFoundries, IonQ’s just-closed acquisition of SkyWater, and the wider $2 billion CHIPS quantum package announced on 21 May 2026. None of these threads replaces the Asian fabs the West has depended on for two decades. They sit alongside them.
By “the West” I mean the United States, the United Kingdom, Germany, and Japan. By “repatriation” I mean new chipmaking capacity being built on Western soil by Western firms and by Asian firms with Western government subsidies, rather than the existing capacity in Taiwan, South Korea, and Japan that already serves the global market. By “fabrication” I mean both front-end wafer production (the work of printing transistors onto silicon in cleanrooms) and the back-end work of advanced packaging, where the silicon is sliced, stacked, and assembled into finished chip modules. The China frame matters because most of the Western policy and most of the Western money that is currently flowing into these three threads is justified as a response to the same US-China semiconductor competition that drove the CHIPS Act of August 2022 and the export controls that followed it in October 2022, October 2023, and December 2024. But the Western fabs do not by themselves solve the China chip story, and an honest account should not pretend they do.
Terafab: what it is and is not
Terafab is a planned semiconductor fabrication complex in Texas, jointly developed by Tesla, SpaceX, xAI (which SpaceX has since folded into a unit called SpaceXAI), and Intel. Elon Musk announced the project on 21 March 2026 at the defunct Seaholm Power Plant in Austin. Intel joined in April 2026. The public framing is unusually bold. The full-scale facility is designed to produce one terawatt of artificial intelligence compute per year, against roughly half a terawatt that the United States currently generates as a country. Initial wafer output is targeted at 100,000 wafer starts per month, scaling to 1 million wafer starts per month at full capacity. For comparison, that full-scale target would represent something on the order of 70 percent of TSMC’s current global output, and it would come from a single facility operated by companies that have never fabricated a chip at scale.
The site plan has two pieces. The first is a research and pilot fab at the North Campus of Tesla’s existing Giga Texas complex in Austin, costing around $3 billion and “intended to try out ideas” in Musk’s phrasing, with capacity for only a few thousand wafers per month. The second is a permanent fab in Grimes County, Texas, on the site of a former electric generating plant, where initial filings in May 2026 disclosed about $55 billion of investment for the first phase, scaling to as much as $119 billion across all phases. In August 2026 SpaceX and Tesla said they would commit about $16.8 billion in the first phase at that site. The project is designed as two physically separate fabs, each making only one chip design. One fab will produce inference chips for Tesla vehicles and the Optimus humanoid robot line, including the AI5 chip that Tesla expects to ship in small batches in 2026 and in volume in 2027. The other fab will produce D3 chips custom-designed for SpaceX’s planned orbital artificial intelligence data centres. Musk has said that 80 percent of Terafab’s compute output is intended for space.
The process technology target is 2 nanometres initially, with Intel’s forthcoming 14A node (1.4 nanometres) intended for the full-scale build. The pilot fab will let Tesla iterate rapidly in one place: design a chip, fabricate it, test it, revise the mask, and repeat, without shipping wafers between sites. Tesla frames that integrated capability as something that no other fabrication site in the world currently offers.
What Terafab is not is a confirmed project at scale. Reuters noted in May 2026: who pays for the lithography and metrology equipment, who operates the factory, when it will come online, and whether the scale assumptions hold. Bernstein analysts estimate that building enough chip capacity to power one terawatt of annual compute would cost somewhere between $5 trillion and $13 trillion in capital expenditure. Musk’s stated rationale is blunt: the global chip industry is not supplying chips fast enough to meet what his companies will need, so Tesla, SpaceX, and Intel are building their own. That is a statement of intent, not a statement of capacity.
Asian fabs in America: front-end, packaging, and equipment
The second thread is the wave of Asian-headquartered chipmakers who are now building inside the United States. Their work is not repatriation in the sense that the West is building its own fabs from scratch. It is the relocation of capacity from existing Asian headquarters to new American sites, with US government subsidies underwriting the move. Three sites dominate the front-end story.
TSMC’s Arizona cluster in Phoenix is the largest. TSMC’s first Arizona fab is operational and, by the company’s Q2 2026 earnings call, is achieving yields that Chief Financial Officer Wendell Huang described as “as good as” TSMC’s flagship facility in Taiwan. The second Arizona fab is moving in equipment, the third is under construction, and a fourth fab together with the site’s first advanced packaging facility is in the preparatory stage. Total committed investment is now $265 billion across twelve fabrication and advanced packaging facilities and an R&D centre, on land that TSMC has had to expand from its original 1,100 acres by an additional 900 acres. The November 2024 finalisation of TSMC’s original $6.6 billion CHIPS Act award, with upside-sharing terms tied to Intel 14A-class milestones and a five-year buyback waiver, marked the first major award under the programme; the July 2026 announcement of an additional $100 billion in investment raised the total commitment further.
Samsung’s Taylor, Texas campus is the second. Samsung has committed roughly $37 billion to a two-fab site, supported by about $4.7 billion in direct CHIPS Act funding. The first fab runs Samsung’s 2 nanometre SF2P+ process and, in July 2026, completed the tape-out of Tesla’s AI5 chip — the first working silicon from the Texas campus, with customer production targeted for 2027 under a multi-year supply agreement with Tesla worth approximately $16.5 billion through 2033. Samsung confirmed in August 2026 that a second fab at Taylor will break ground by the end of 2026 and enter mass production by 2030, with industry sources indicating that the second fab is being pushed toward a 1.4 nanometre node in response to TSMC’s 1.6 nanometre roadmap.
SK Hynix’s West Lafayette, Indiana facility at Purdue Research Park is the third front-end story, with an important distinction: it is currently a back-end, advanced packaging, and research facility, not a wafer fab. SK Hynix broke ground on 27 August 2026 on a $3.87 billion project designed to package high-bandwidth memory (HBM) using wafers produced in South Korea, with HBM4E volume production targeted for the third quarter of 2029 and cleanroom operations beginning in the second half of 2028. SK Hynix is also considering its first front-end fab in the United States, but at the site-selection stage, with no commitment yet. The site’s strategic logic is that advanced packaging has become the bottleneck for AI accelerator supply, and putting packaging capacity close to the American hyperscaler customers matters more than putting wafer production close to them.
Two smaller pieces complete the picture. Hanmi Semiconductor, the Korean maker of thermo-compression bonders used in HBM stacking, announced in May 2026 that it would establish Hanmi USA in San Jose, California, by year-end 2026, with its first US presence aimed at SK Hynix’s Indiana facility and at equipment demand from Micron, Intel, and Tesla. Amkor Technology, the world’s second-largest outsourced semiconductor assembly and test provider, broke ground in October 2025 on a $7 billion advanced packaging campus in Peoria, Arizona — the first US high-volume advanced packaging facility — with Apple as the first and largest customer and NVIDIA also an anchor customer. Amkor previously received up to $407 million in proposed CHIPS Act funding. The Amkor facility complements TSMC’s Arizona front-end fabs and is scheduled to begin production in early 2028.
The honest picture is uneven. TSMC Arizona is producing. Samsung Taylor is producing tape-outs and will produce chips next year. SK Hynix Indiana is still a building site. Hanmi USA is a sales and support office. Amkor Arizona is still under construction. And the back-end supply chain that supports Samsung Taylor specifically is largely absent from Texas: the OSAT services, the HBM substrates, and the TC bonder equipment remain in Korea or in California, not on Taylor’s doorstep. The Texas fab is real; the Texas semiconductor cluster around it is not yet.
Quantum foundries in America
The third thread is the new American quantum foundries. Until 2026, the phrase “quantum foundry” usually meant an academic lab capable of fabricating a handful of qubit wafers for research. The 2026 expectation is a true wafer fab producing quantum-grade devices at volume for multiple vendors, with the process design kits, in-line characterisation, and yield discipline that logic-chip customers expect from TSMC or Samsung.
The single largest piece is IBM’s new subsidiary Anderon, announced on 21 May 2026 as a standalone company headquartered in Albany, New York, designed as a 300-millimetre quantum wafer foundry — the same wafer format as leading-edge classical fabs. The deal pairs $1 billion in proposed CHIPS incentives from the US Department of Commerce with $1 billion in cash from IBM, plus intellectual property, assets, and staff. Anderon will focus initially on superconducting qubit wafers and supporting electronics wafers, with the explicit goal of expanding to other quantum modalities. On the same day the Department of Commerce signed a separate letter of intent with GlobalFoundries for $375 million in CHIPS funding, to establish a multi-modality quantum foundry spanning superconducting, trapped-ion, photonic, topological, and silicon-spin qubits, extending GlobalFoundries’ existing work with PsiQuantum.
The smaller recipients of the same CHIPS quantum package include Atom Computing, D-Wave, Infleqtion, PsiQuantum, Quantinuum, and Rigetti at $100 million each, plus Diraq at up to $38 million. Rigetti runs the only US-owned superconducting quantum fab in commercial operation, Fab-1; that fab is the recipient of the company’s $100 million CHIPS award. The portfolio deliberately spans modalities rather than betting on one architecture. The Department of Commerce, in a structural departure from conventional grant programmes, will take a minority, non-controlling equity stake in each of the nine recipients.
The most consequential 2026 development in the quantum foundry space, however, was IonQ’s acquisition of SkyWater Technology, which closed on 31 July 2026 after SkyWater stockholders approved the $1.8 billion deal on 8 May. SkyWater is the largest exclusively US-based pure-play semiconductor foundry, with Department of Defence Microelectronics Activity accreditation at Category 1A Trusted, and sites in Bloomington Minnesota, Sanford Florida, and Austin Texas. IonQ is using the SkyWater foundry to fabricate the first fully integrated 256-qubit quantum processing units for its Superion 256 platform, completing six tapeouts in the first half of 2026 and compressing IonQ’s chip design cycle from nine months to two. The acquisition converts SkyWater’s three sites into Regional Quantum Production Hubs and gives IonQ — and, through SkyWater’s existing merchant foundry relationships, D-Wave and others — a domestic, trusted supply chain.
The total capital flowing into US quantum fabrication is now $2.013 billion in CHIPS quantum incentives alone, with total public and private US quantum capital estimated at $13.8 billion. The clearest summary of where this leaves things came in the Futurum Group’s analysis: 50-to-1 funding ratio between Anderon’s $1 billion and Diraq’s $38 million reflects an assessment by the US government that superconducting silicon is the only quantum modality currently capable of leveraging production-grade semiconductor fabrication infrastructure at 300 millimetre scale.
Why now: the mechanism behind the repatriation
Three conditions lined up in 2026 to make the repatriation fundable, designable, and politically durable. The first was the CHIPS and Science Act of August 2022, which committed roughly $280 billion in new federal funding, of which $52.7 billion was appropriated for semiconductor manufacturing incentives and research, and the additional 25 percent investment tax credit. The second was the series of US export controls tightening access to advanced computing chips and semiconductor manufacturing items for end uses in mainland China — the October 2022 rule, the October 2023 update, and the December 2024 expansion. The third was the demand curve for AI compute, which the lifting of large-language-model capability since late 2022 had converted from a research curiosity into a corporate procurement priority. Together these three conditions made it possible for the first time in the modern semiconductor era to justify on cost-and-strategy grounds what had previously been justified only on national-security grounds: building advanced fabs on Western soil, with Western subsidies, for Western customers.
The shortfall these moves were designed to close was severe. Before 2022 the United States produced roughly 0 percent of the world’s leading-edge logic chips. Under current trajectories the US share is expected to rise toward 10 to 15 percent by 2030, with South Korea and Taiwan still producing the bulk of the world’s most advanced silicon. The Asian fabs now arriving on American soil are not repatriating the leading edge in the strict sense — most of their American capacity is one to two process generations behind their Asian headquarters — but they are repatriating capacity at a meaningful slice of the production stack, and they are doing it under conditions of US government partnership that did not exist a decade ago.
Limits and caveats
Several limits deserve honest acknowledgement. Terafab’s headline numbers are targets, not capacities, and the gap between announcement and production in fab projects is historically wide. The back-end gap around Samsung Taylor is real and not yet addressed by the suppliers on the ground. SK Hynix’s Indiana timeline is years behind the AI demand curve. The quantum commercialisation story is still early; IBM’s roadmap targets a verified demonstration of quantum advantage by the end of 2026 on its Nighthawk platform, and the first large-scale fault-tolerant system, Starling, by 2029. TSMC’s Arizona build is hitting a construction-worker shortage that the company has flagged in public, with the Chief Financial Officer warning in July 2026 that the shortage of skilled construction labour in Phoenix is now a binding constraint on the speed of the build. The wider bottleneck on every front-end project is the supply chain for the most advanced lithography, metrology, and process control tools, which is dominated by ASML in the Netherlands, Applied Materials and Lam Research in California, and KLA in California, and which cannot be expanded quickly no matter how much capital is committed to building fabs. And the SMIC parallel inside mainland China, covered in earlier huaxia.click essays on TSMC and on China’s excess capacity, is the part of this story these Western fabs do not address. The Western build-out is a response to a strategic problem, not a solution to it.
Closing
What changed in 2026 was not that Western fabrication suddenly became competitive with the Asian fabs that the global chip industry has depended on for two decades. It was that the cost calculus, the strategic calculus, and the artificial intelligence demand curve finally lined up at the same time. The CHIPS Act and the export controls gave the Western governments both the money and the policy rationale. The Asian chipmakers accepted the subsidies and committed to American sites under multi-decade horizons. The AI compute demand gave those sites customers. The quantum foundries are an attempt to do the same thing a generation earlier, before the technology locks in around foreign supply chains. None of it removes the dependence on Taiwan. None of it removes the dependence on Korea. None of it removes the dependence on SMIC’s parallel push inside mainland China. What it does is begin, slowly and expensively, the work of building the alternative.
Sources
References
[1] Reuters — Elon Musk lays out Terafab AI chip project plan (6 May 2026). Primary reporting on SpaceX’s $55B first-phase Texas filing and Bernstein’s $5–13 trillion full-buildout estimate.
[2] CNBC — Elon Musk’s SpaceX chip fab in Texas to cost up to $119 billion (6 May 2026). Primary reporting on Grimes County public-hearing filing.
[3] Wikipedia — Terafab (retrieved 19 September 2026). Encyclopaedic summary of the joint venture, sites, and timelines.
[4] Forbes — Intel Joins Terafab To Build Elon Musk’s $25 Billion AI Chip Project (10 April 2026). Reporting on Intel’s foundry-partner announcement.
[5] Reuters — TSMC expects strong, multi-year demand for AI chips as it ramps up Arizona investment (19 July 2026). Reporting on the Q2 2026 earnings call and the additional $100 billion US investment.
[6] Korean Tech Texas — The Back-End Gap, Revisited: Samsung Taylor’s Front-End Is Now Real (August 2026). Reporting on the Tesla AI5 tape-out at Taylor Fab 1 and the backend supply-chain gap.
[7] Seoul Economic Daily — Samsung, SK hynix Race to Build Chip Plants in Korea, U.S. (21 August 2026). Reporting on Samsung’s 1.4nm Taylor expansion and SK Hynix’s front-end fab consideration.
[8] Reuters — SK Hynix holds groundbreaking ceremony for $4 billion Indiana AI chip packaging facility (27 August 2026). Primary reporting on the West Lafayette groundbreaking and HBM4E production timeline.
[9] Amkor Technology — Breaks Ground on New Semiconductor Advanced Packaging and Test Campus in Arizona (6 October 2025). Primary press release on the $7 billion Peoria campus.
[10] IBM Newsroom — IBM and U.S. Department of Commerce Announce America’s First Purpose-Built Quantum Foundry (21 May 2026). Primary press release on the Anderon letter of intent.
[11] NIST — Department of Commerce Announces Letters of Intent With 9 Companies for $2 Billion to Accelerate U.S. Leadership in Quantum Computing (21 May 2026). Primary press release on the CHIPS quantum package.
[12] IonQ — IonQ Debuts Superion 256 Quantum Computing Platform (September 2026). Primary press release on the SkyWater-fabricated 256-qubit QPUs and the deal close.
[13] Futurum Group — Will a U.S. Quantum Foundry Leverage $4.6 Billion in New Capital to Become the Next TSMC? (August 2026). Industry analysis on the quantum foundry landscape.
[14] CHIPS for America — U.S. Department of Commerce. Programme portal for awards and CHIPS Act documentation.