AI Chip Production 2026: Map of New U.S. Chip Plants

Top photo: TSMC’s Fab 21 campus under construction in north Phoenix, Arizona, Nov. 5, 2023. Photo by Hunter Trick (TrickHunter), licensed under CC BY-SA 4.0, via Wikimedia Commons. Cropped.

Key facts, as of Oct. 18, 2026

Record output, still short: TSMC reported record third-quarter net income of NT$763.06 billion on Oct. 15. It expects fourth-quarter revenue of $52.3 billion to $54.3 billion.
Arizona speeds up: TSMC’s second Arizona fab is finished and tools are moving in. Volume production of 3-nanometer chips is now set for the third quarter of 2027, slightly ahead of plan.
Memory is the other choke point: Micron says much of its 2027 high-bandwidth memory (HBM) output is already sold, at prices “much higher” than in 2026.
11 U.S. sites mapped: Four are producing. The rest target first output between 2027 and 2031, and five have slipped behind their first public schedules.
Policy: A 25% tariff on a narrow group of advanced chips has applied since Jan. 15. Since Jan. 13, Nvidia H200-class exports to China have been reviewed case by case.

AI chip production is at record levels and still falling short of demand. Taiwan Semiconductor Manufacturing Co. (TSMC) makes most of the world’s advanced AI processors. On Oct. 15 it reported third-quarter revenue of $46.72 billion, up 41.2% from a year earlier, and record net income of NT$763.06 billion, up 68.7%. “Even after those checks, TSMC says demand remains well above its supply capacity,” SemiAnalysis analyst Sravan Kundojjala told CNBC. He was referring to TSMC’s checks on whether customers have the financing and other components needed to deploy AI systems.

The quarter also brought news from TSMC’s U.S. plants. “In Arizona, we have completed the construction of our second fab, and are in the process of tool move-in and installation,” Chairman and CEO C.C. Wei said on the earnings call. “Volume production will begin in 3Q’27, slightly ahead of our previous schedule.” A day earlier, on Oct. 14, GlobalFoundries said it will operate two new fabs in Malta, N.Y., funded by up to $21 billion from Japan. The first, costing about $10 billion, is expected online in 2029. We covered that deal in Japan’s $21 billion plan for two GlobalFoundries fabs in New York.

The interactive map below shows where new U.S. chipmaking and packaging capacity is being built and when each site is due to produce. After it comes a look at the bottlenecks in AI chip production and at U.S. policy. More coverage is on our Tech & AI hub and our Business & Tech hub.

Who makes AI chips, and why capacity is so tight

Nvidia, AMD and Broadcom design most AI accelerators, but none of them owns a factory. According to CNBC, TSMC makes the vast majority of the world’s most advanced semiconductors, including those companies’ graphics processors and central processors. Apple, Google, Amazon, Microsoft, Meta and OpenAI also rely on TSMC for their custom in-house AI chips. Sajiv Dalal, who heads TSMC North America, told CNBC he gets at least 10 calls a day about wafer capacity. U.S. cloud giants are projecting well over $700 billion in capital spending this year.

Leading-edge chips now drive TSMC’s business. In the third quarter, 3-nanometer chips made up 33% of its wafer revenue and the new 2-nanometer node made up 8%. Chips at 7 nanometers and below made up 79% in total. High-performance computing, the category that includes AI accelerators, brought in 68% of revenue.

Each chip designer is lining up supply years ahead. Nvidia said on May 31 that its Vera Rubin platform was ramping into full production, with production shipments starting this fall. Taiwan’s Economic Daily News, as summarized by TrendForce, reported that AMD’s MI450 series entered production at TSMC in the third quarter on the 2-nanometer process with CoWoS packaging. Higher-volume shipments are expected in 2027. AMD CEO Lisa Su said capacity planning that once looked one to two years ahead now runs three to five years out.

TSMC’s two rivals at the leading edge are both building in the U.S. Intel’s Fab 52 in Chandler, Ariz., has been making chips on Intel’s 18A process since October 2025. Tom’s Hardware reports that Intel is capping output there until yields reach industry-standard levels, which it expects in early 2027. In Taylor, Texas, industry sources told Seoul Economic Daily that Samsung began prototype production on a 2-nanometer process in September for Tesla’s AI5 chip, with full supply planned for next year.

Advanced packaging: the CoWoS squeeze

A modern AI accelerator is a package of several chips, not one. TSMC’s CoWoS (chip-on-wafer-on-substrate) process places the processor beside stacks of high-bandwidth memory on a silicon interposer, a thin base that carries signals between the chips. Packaging capacity has been one of the industry’s hardest bottlenecks.

Institutional investors cited by Economic Daily News estimated in June that TSMC’s CoWoS capacity could reach 120,000 to 140,000 wafers a month in 2026. With outside packaging firms included, industry capacity would be near 200,000. The same investors expected the gap between CoWoS supply and demand to narrow from about 20% to about 10% by the end of 2026. Later reports said TSMC’s CoWoS capacity could double to about 260,000 wafers a month by the end of 2028, with Arizona among the sites.

This is why several of the U.S. projects on the map are packaging plants rather than wafer fabs. They include Amkor’s campus in Peoria, Ariz., SK hynix’s HBM plant in West Lafayette, Ind., and TSMC’s first Arizona packaging fab, which began early construction in 2026. GlobalFoundries’ expansion in New York is set to become the first U.S. source of silicon interposers.

The memory bottleneck: HBM

High-bandwidth memory stacks DRAM chips vertically next to the processor, and every Nvidia and AMD accelerator needs it. Only three companies make it at scale. Counterpoint Research, as cited by Seoul Economic Daily, put SK hynix’s second-quarter share of HBM revenue at 50%, Samsung’s at 33% and Micron’s at 18%.

Micron CEO Sanjay Mehrotra told investors on Sept. 30 that “a large part of the volume is already sold out for 2027 for HBM, and the prices are much higher than 2026 prices,” Ars Technica reported. Samsung executive Kim Taewoo said HBM will take almost 30% of memory makers’ DRAM wafer capacity in 2027, up from 20% this year, according to Reuters. That leaves less memory for consumer PCs and phones. TrendForce projects that HBM’s average selling price will rise 121% in 2027. Micron’s own fiscal 2026 revenue reached $133.19 billion, more than triple the year before.

Interactive map: new U.S. chip plants and their dates

The map shows 11 new or expanding U.S. chip sites. Ten make logic chips, memory or advanced packaging. The 11th is Texas Instruments’ new fab in Sherman, Texas, which makes analog and embedded chips that also go into data centers. Each marker gives the company, location, type of plant, investment, CHIPS Act award, announcement date, groundbreaking date, production start and current status, with links to sources. Actual dates have happened. Target dates are company-stated goals. Where a cited source gives no date, the field says so instead of estimating one. All data are current as of Oct. 18, 2026.

New and expanding U.S. chip plants, as of Oct. 18, 2026

Click or tap a marker for the company, investment, CHIPS Act award, key dates and status. Use the buttons to jump to a site. Marker positions are approximate.

Loading the map. Every site is also listed in the table below.

LogicMemoryPackagingAnalog/embedded

Map: The Boston Report, using company announcements, U.S. Commerce Department CHIPS award notices and news reports linked in each marker and in Sources below. Map tiles © OpenStreetMap contributors. "Target" dates are company-stated goals, not guarantees.

Timeline: from announcement to first production

Intel Rio Rancho, NM

TSMC Phoenix, AZ

Intel Chandler, AZ

Texas Instruments Sherman, TX

Samsung Taylor, TX

Micron Boise, ID

Amkor Peoria, AZ

GlobalFoundries Malta, NY

SK hynix West Lafayette, IN

Micron Clay, NY

Intel New Albany, OH

Announcement to production start (actual)Announcement to company-stated production target

Bars run from each project’s announcement to its first production milestone, sorted by production date. Intel Fab 9 shows its opening only, because the cited source gives no announcement date. Sources: see table and Sources section.

All sites in one table

New and expanding U.S. chip manufacturing and packaging sites, as of Oct. 18, 2026
Site	Company	Location	Type	Investment	CHIPS Act funding	Announced	Ground broken	Production	Status	Sources
TSMC Arizona (Fab 21 campus)	TSMC	Phoenix, AZ	Logic (Advanced packaging planned)	$265 billion total planned Arizona commitment (six logic fabs, two packaging plants, R&D center, plus more fabs announced July 2026)	Up to $6.6 billion (finalized Nov. 15, 2024)	May 15, 2020	Construction under way by June 1, 2021	Actual: Fab 1 (N4): high-volume production began Q4 2024	Fab 1 producing. Fab 2 (3nm) construction complete, tools moving in; volume production targeted for Q3 2027, slightly ahead of the earlier plan. Fab 3 (2nm/A16) broke ground April 2025, targeting volume production by the end of the decade. Fab 4 and the first advanced packaging fab started early construction in 2026.	TSMC Arizona; TSMC 2020 announcement; Reuters, June 2021; TSMC Q3 2026 call; Commerce CHIPS award
Intel Ocotillo campus (Fab 52 and Fab 62)	Intel	Chandler, AZ	Logic	$20 billion for two fabs (2021 plan)	Intel award of $7.86 billion; unpaid portion converted to a 9.9% U.S. equity stake in Aug. 2025 (company-wide, not site-specific)	March 23, 2021	Sept. 24, 2021	Actual: Fab 52: fully operational, ramping Intel 18A since Oct. 2025	Delayed. Fab 52 is producing 18A chips (Panther Lake, Clearwater Forest). Intel originally said both fabs would be fully operational in 2024. Fab 62 is under construction and expected to be ready around 2028, per Tom's Hardware; Intel has not assigned it a node.	Intel, March 2021; Intel groundbreaking; Intel, Oct. 2025; Tom's Hardware, June 2026
TI Sherman SM1 (up to four fabs)	Texas Instruments	Sherman, TX	Analog/embedded (300mm mature-node)	About $30 billion for up to four fabs	Up to $1.61 billion (Texas and Utah, Dec. 2024)	Nov. 17, 2021	May 18, 2022	Actual: SM1 production began Dec. 17, 2025	SM1 producing and ramping with customer demand. TI first expected production in 2025, and it started in December 2025. Later fabs will be built in line with market demand.	TI, Nov. 2021; TI groundbreaking; TI, Dec. 2025; Commerce CHIPS award
Samsung Taylor fab	Samsung	Taylor, TX	Logic (2nm foundry)	$17 billion at announcement; more than $37 billion across Taylor and Austin	Up to $4.745 billion (finalized Dec. 20, 2024)	Nov. 23, 2021 (U.S. time)	Construction began in 2022 (utilities and foundations from June 2022)	Target: Prototype 2nm output for Tesla AI5 began by Sept. 2026; full supply targeted for 2027	Delayed. Originally due to operate in the second half of 2024. Industry sources told Seoul Economic Daily that prototype production and yield checks were under way in Sept. 2026, with a second fab groundbreaking targeted for late 2026.	Samsung, Nov. 2021; Korea JoongAng Daily; Commerce CHIPS award; Seoul Economic Daily, Sept. 2026
Intel Ohio One (Mod 1 and Mod 2)	Intel	New Albany, OH	Logic	$28 billion (Intel)	Part of Intel's company-wide award (see Intel Arizona)	January 2022	Sept. 9, 2022	Target: Mod 1 operations 2030 to 2031; Mod 2 in 2032	Delayed. Production was first planned for 2025. Intel slowed construction in Feb. 2025 and says timing depends on customer demand. A Licking County commissioner said in Aug. 2026 that chipmaking tools had not yet been ordered.	Intel groundbreaking; Intel timeline update; WOSU, Aug. 2026; New Albany-Gahanna Beacon
Micron Boise fabs (ID1 and ID2)	Micron	Boise, ID	Memory (DRAM, including for AI)	About $50 billion across two fabs ($15 billion at 2022 announcement)	Part of Micron's award of up to $6.165 billion for Idaho and New York (Dec. 10, 2024)	Sept. 1, 2022	Sept. 12, 2022	Target: ID1 first wafer output mid-2027; ID2 late 2028	Delayed. Under construction. In 2022 Micron said DRAM output would start in 2025. Micron planned to move tools into ID1 later in 2026 and reaffirmed the mid-2027 first-wafer target in July 2026.	Reuters, Sept. 2022; Micron groundbreaking; Idaho Statesman, May 2026; Micron, July 2026; Commerce CHIPS award
Micron Central New York megafab	Micron	Clay, NY	Memory (DRAM, up to four fabs)	Up to $100 billion	Part of Micron's award of up to $6.165 billion for Idaho and New York (Dec. 10, 2024)	Oct. 4, 2022	Jan. 16, 2026	Target: Production expected to start in 2030	Vertical construction began with a first concrete pour on July 9, 2026, more than a quarter ahead of plan. Bechtel is the main contractor for the first fab.	Micron, Oct. 2022; Micron groundbreaking; Micron, July 2026; Commerce CHIPS award
Amkor Arizona advanced packaging and test campus	Amkor	Peoria, AZ	Packaging (Packaging and test (TSMC, Apple, Nvidia customers))	About $12 billion with phase 2 ($2 billion at 2023 announcement)	Up to $407 million (Dec. 2024)	Nov. 30, 2023	October 2025	Target: Phase 1 production in early 2028	Under construction. Phase 1 building due mid-2027. Phase 2, announced Sept. 8, 2026, would start construction in late 2027 and finish by the end of 2029.	Amkor, Nov. 2023; DCD, Oct. 2025; GPEC/Amkor, Sept. 2026; Commerce CHIPS award
SK hynix HBM packaging plant	SK hynix	West Lafayette, IN	Packaging (HBM memory)	More than $4 billion ($3.87 billion at 2024 announcement)	Up to $458 million (finalized Dec. 2024)	April 3, 2024	Aug. 27, 2026	Target: HBM mass production in the second half of 2029	Delayed. Under construction. The cleanroom is due by Oct. 2028. In 2024 SK hynix targeted mass production in the second half of 2028.	SK hynix, April 2024; SK hynix groundbreaking; Commerce CHIPS award
GlobalFoundries Malta (new Japan-funded fabs and expansion)	GlobalFoundries	Malta, NY	Logic (Specialty logic, silicon photonics, interposers)	About $10 billion for a first new fab and about $10 billion for a possible second (up to $21 billion from Japan); plus $5 billion in Malta and Burlington, Vt.	$1.45 billion award (New York); $955 million reallocated Oct. 14, 2026; U.S. gets about 4% equity	Oct. 14, 2026	Expected later in 2026	Target: First new fab expected online in 2029	Announced. GF says the site is permitted and shovel-ready. The $5 billion program adds silicon photonics, silicon-germanium and the first U.S. source of silicon interposers for advanced packaging.	GlobalFoundries, Oct. 2026; NIST, Oct. 2026; NIST CHIPS awards
Intel Fab 9 (Rio Rancho)	Intel	Rio Rancho, NM	Packaging (Foveros 3D packaging)	$3.5 billion (New Mexico investment)	Not disclosed separately	Not stated in cited sources	Not stated in cited sources	Actual: Opened Jan. 24, 2024	Operating. Intel's first high-volume Foveros 3D packaging site; the only one in the U.S., per Tom's Hardware.	Intel, Jan. 2024; Tom's Hardware

Select a column heading with a dotted underline to sort. On phones, swipe the table sideways. “Delayed” means the current target is later than the company’s first public target.

What the map shows: delays, and a few accelerations

Four of the 11 sites are producing: TSMC’s first Arizona fab (since the fourth quarter of 2024), Intel’s Fab 9 packaging plant in New Mexico (opened January 2024), Intel’s Fab 52 (since October 2025) and TI’s SM1 in Sherman (since Dec. 17, 2025). Most of the others are running behind their first public schedules:

Intel Ohio One was announced in 2022 with production originally planned for 2025. Intel now targets 2030 to 2031 for the first fab and 2032 for the second. A Licking County commissioner said in August that the chipmaking tools had not yet been ordered.
Intel Arizona said in 2021 that Fab 52 and Fab 62 would be fully operational in 2024. Fab 52 reached 18A production in October 2025, and Fab 62 is expected to be ready around 2028.
Samsung Taylor was meant to be operating in the second half of 2024. It is now in prototype production, with full supply planned for 2027.
Micron Boise planned DRAM output in 2025 when it broke ground. Its first new fab now targets first wafers in mid-2027, and the second in late 2028.
SK hynix Indiana moved its mass-production target from the second half of 2028 to the second half of 2029 by the time it broke ground on Aug. 27, 2026.

A few projects are moving faster. TSMC’s second Arizona fab is slightly ahead of schedule, and Micron poured first concrete in Clay, N.Y., on July 9, more than a quarter ahead of plan. U.S. plants cost more to run. TSMC said the dilution of its gross margin from overseas fabs will be close to 2% in 2026, widening to 3% to 4% in the next few years as more U.S. fabs ramp up.

CHIPS Act, tariffs and export controls: U.S. policy in October 2026

CHIPS Act money. The 2022 CHIPS and Science Act funds most of the awards listed in the table: up to $6.6 billion for TSMC, $6.165 billion for Micron, $4.745 billion for Samsung, $1.61 billion for Texas Instruments, $458 million for SK hynix and $407 million for Amkor. The Commerce Department pays them out as companies reach construction and production milestones. Under the current administration, Commerce has also taken equity. Intel’s unpaid CHIPS grants and Secure Enclave funds were converted into a 9.9% government stake in August 2025, and the government will receive about 4% of GlobalFoundries under an agreement announced Oct. 14.

Tax-credit deadline. A July 2025 law raised the advanced manufacturing investment credit from 25% to 35%. The credit still does not cover projects whose construction begins after Dec. 31, 2026, as Tom’s Hardware has noted, so companies have a reason to start building before year’s end.

Tariffs. A presidential proclamation imposed a 25% duty on certain advanced computing chips from Jan. 15, 2026. Imports for U.S. data centers, U.S. research, startups and several other uses are exempt. The proclamation also set up a second phase tied to trade negotiations. As of Oct. 1, no broader semiconductor tariff had taken effect, according to a review of Federal Register notices by the trade-compliance site Sourceics. Our explainer on U.S. economic policy and tariffs covers the wider tariff picture.

Export controls. On Jan. 13 the Commerce Department’s Bureau of Industry and Security (BIS) began reviewing license applications to export Nvidia’s H200, AMD’s MI325X and similar chips to China case by case. Applicants must show that the sales will not reduce chip capacity available to U.S. customers, that the Chinese buyer screens its customers, and that the product has passed independent testing in the United States. Reuters reported in July that a U.S. official said H200 shipments to China had begun. Taiwan is also expanding in Texas: see our report on Taiwan’s planned $4 billion park for AI chip suppliers.

The Massachusetts and Boston angle

None of the plants on the map is in Massachusetts. The state’s part in the chip build-out is research, design and prototyping. The Northeast Microelectronics Coalition (NEMC), a division of the Massachusetts Technology Collaborative, is one of eight regional hubs in the Defense Department’s Microelectronics Commons program. On April 28, the Healey-Driscoll administration announced that NEMC had won $22.5 million for four projects, after $37.8 million in September 2024.

The largest of the four, at $7.04 million, is an AI hardware project led by UMass Amherst. It aims to move chips that combine standard CMOS logic with memristors, used for AI processing in devices, from the lab into production. GlobalFoundries, Raytheon and BAE Systems are among its partners. MIT leads a $2.5 million project on 5G and 6G antenna arrays, and Analog Devices leads a $6.3 million project on high-power gallium nitride electronics. NEMC says its hub has more than 300 member organizations.

The AI build-out also needs power, and some of that work happens near Boston. Woburn-based VEIR recently raised money to bring superconducting power lines to AI data centers; see our VEIR report. This past week’s tech stock sell-off showed that investors are also watching how the AI build-out is financed.

Reactions

“Moving into fourth quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2-nanometer technology,” TSMC Chief Financial Officer Wendell Huang said on Oct. 15.

“Overall, supply-demand environment is only getting tighter,” Micron’s Mehrotra told investors on Sept. 30, according to Ars Technica.

“The competition is extremely fierce,” Mizuho Securities analyst Jordan Klein told CNBC of the race for TSMC capacity. “It’s a game of who can get the most leverage and supply.” Dan Hutcheson of TechInsights told CNBC that “people want a second source to be able to twist TSMC’s arm and keep them honest on pricing.”

“Today marks the day when the United States and SK hynix embark on a new future of AI together,” SK hynix CEO Kwak Noh-Jung said at the Aug. 27 groundbreaking in Indiana.

“With Japan’s investment funding and GlobalFoundries’ partnership, two large semiconductor chip manufacturing facilities will be built in the U.S., creating thousands of high-paying jobs for American workers,” Commerce Secretary Howard Lutnick said in GlobalFoundries’ Oct. 14 announcement.

In Ohio, New Albany Mayor Sloan Spalding told WOSU that “Intel’s been pretty transparent about their need to get customers lined up in order to keep the project moving along at a good pace.” Licking County Commissioner Tim Bubb warned about the timeline: “You better make sure you know what you’re ordering, because it’s like a two-year lead time and (it’s) very expensive to put the equipment there to make their chips.”

Sources

Like this story? Get the newsletter

New Boston Report stories delivered to your inbox, with links to the full articles.

Free. Confirm by email, unsubscribe anytime. Signups are handled by Jetpack/WordPress.com. Privacy Policy

Latest

All latest news →

AI Chip Production in 2026: Record Demand, Tight Supply and a Map of New U.S. Chip Plants