Foxconn × NVIDIA: Inside the World’s Biggest GB200 AI Server Build in Mexico and Taiwan
Foxconn (Hon Hai) is building a major NVIDIA GB200 superchip server production facility in Guadalajara, Mexico, alongside a 64-rack GB200 NVL72 supercomputer at its Kaohsiung Super Computing Center in Taiwan. The Taiwan system is designed to deliver more than 90 exaflops of AI performance, with full deployment targeted for 2026.
Key Facts at a Glance
| Item | Details |
|---|---|
| Announcement | 8 October 2024, Hon Hai Tech Day, Taipei |
| Mexico production site | Guadalajara, Jalisco; reported 450-metre-long plant |
| Taiwan facility | Hon Hai Kaohsiung Super Computing Center |
| AI platform | NVIDIA GB200 NVL72, based on Blackwell |
| Taiwan system | 64 racks and 4,608 GPUs |
| Claimed AI performance | More than 90 exaflops |
| Foxconn’s Mexico investment | More than US$500 million to date, with an additional US$241.2 million announced in August 2024 |
| Guadalajara land purchase | 453 million Mexican pesos in 2023 |
Foxconn: From Plastic Components to Global Electronics Manufacturing
Terry Gou founded Hon Hai in Taiwan in 1974. The company reportedly began with a small loan from his mother and a handful of workers manufacturing plastic parts for black-and-white television sets. It later expanded into connectors, computer cases and complete electronics assembly for major technology companies, including Apple and Dell.
Foxconn’s involvement in iPhone assembly from 2007 became a major turning point, helping establish the company as the world’s largest electronics contract manufacturer. Gou stepped back from the chairman’s role in 2019 and later entered Taiwan’s presidential politics.
NVIDIA: From a Denny’s Booth to the AI Chip Industry
Jensen Huang, Chris Malachowsky and Curtis Priem founded NVIDIA in April 1993. The company’s early planning famously took place in a Denny’s diner in San Jose, California.
Its first chip, the NV1, struggled commercially, and NVIDIA came close to running out of money in 1996. The RIVA 128 helped the company recover. NVIDIA went public in 1999 and subsequently introduced the GeForce 256, marketed as the world’s first GPU. Decades later, GPUs became essential to artificial intelligence training and high-performance computing.
Timeline of the Foxconn–NVIDIA Partnership
| Year | Milestone |
|---|---|
| 1974 | Terry Gou founded Hon Hai |
| 1993 | NVIDIA was founded |
| 2007 | Foxconn began iPhone assembly |
| 2022 | Foxconn opened a high-tech electronics board plant in Jalisco |
| 2023 | Foxconn purchased land in Guadalajara for 453 million pesos |
| October 2024 | Guadalajara GB200 production and the Kaohsiung supercomputer were announced |
| 2025 | Mexico plant operations were scheduled to begin, alongside the first phase of the Kaohsiung system |
| 2026 | Full Kaohsiung deployment was targeted |
What Is the NVIDIA GB200 NVL72?
The GB200 NVL72 is a liquid-cooled AI server rack containing 36 Grace CPUs and 72 Blackwell GPUs, arranged as 36 GB200 superchips. The components communicate through NVLink, with a stated bandwidth of 130 TB/s.
Manufacturing these systems requires precise component placement, dense circuit-board assembly, liquid-cooling connections, heat-dissipation equipment and extensive burn-in stress testing before shipment. Their weight and technical complexity make specialised production facilities important.
The Numbers Behind the Taiwan Supercomputer
| Calculation | Result |
|---|---|
| 64 racks × 72 GPUs | 4,608 GPUs |
| 64 racks × 36 Grace CPUs | 2,304 CPUs |
| 64 racks × 36 GB200 superchips | 2,304 superchips |
The stated performance of more than 90 exaflops uses low-precision AI arithmetic. It should not be compared directly with traditional supercomputer benchmark scores that use different numerical formats.
Why Guadalajara Matters
Guadalajara’s location places production closer to US customers and supports the diversification of technology supply chains away from China. Foxconn has also said the site will support cooling solutions for GB200 infrastructure. Jalisco’s government described the announcement as highly significant for the state.
People Behind the Project
| Person | Role | Birth Date and Age in October 2026 | Background |
|---|---|---|---|
| Jensen Huang | NVIDIA co-founder and CEO | 17 February 1963; 63 | Born in Tainan, Taiwan, he moved to the US as a child, earned engineering degrees from Oregon State University and Stanford, and worked at LSI Logic and AMD before founding NVIDIA |
| Terry Gou | Founder of Hon Hai (Foxconn) | 18 October 1950; 76 | Built Foxconn from 1974, served as chairman and later entered Taiwan’s presidential politics |
| Young Liu | Foxconn chairman | Not stated | Confirmed the Mexico plant to customers at Tech Day |
| Benjamin Ting | Foxconn senior vice president | Not stated | Described the project as the largest GB200 production facility on the planet |
Net Worth and Family
Jensen Huang’s net worth moves with NVIDIA’s share price and has been in the low-to-mid hundreds of billions of US dollars on major wealth trackers. Terry Gou’s estimated wealth is generally in the single-digit billions of US dollars.
Jensen Huang is married to Lori Huang, whom he met at Oregon State University. Their children, Madison and Spencer, both work at NVIDIA. AMD CEO Lisa Su is reported to be a distant relative of Huang, adding an unusual family connection to the rivalry between NVIDIA and AMD.
Frequently Asked Questions
Where is Foxconn’s NVIDIA production facility?
The main production site is in Guadalajara, Jalisco, Mexico, while the companion supercomputer project is located in Kaohsiung, Taiwan.
What is the GB200 NVL72?
It is a liquid-cooled AI server rack containing 72 Blackwell GPUs and 36 Grace CPUs connected through NVLink.
How many GPUs will the Taiwan system have?
The planned 64-rack system contains 4,608 GPUs, alongside 2,304 Grace CPUs and 2,304 GB200 superchips.
Foxconn’s manufacturing expertise and NVIDIA’s AI computing technology place both companies at the centre of the expanding market for advanced AI infrastructure.
