Implications of Increasing Mexican Exports of AI Data Servers to the United States

Introduction

During the past several years, the composition of Mexican exports to the United States has significantly changed. Auto and auto parts exports have decreased because of high Trump administration tariffs designed to shift production for the U.S. market from Mexico (and Canada) to the United States. (Trump tariff policy) Simultaneously, Mexican exports of the servers required to satisfy the apparently insatiable demand for AI data centers has skyrocketed. An estimated $650 billion investment in U.S. data centers is expected in 2026. (data center investment) Mexico now provides an estimated 40% of U.S. server imports, with sales reaching $46.9 billion for the first seven months, second to Taiwan, with sales of $53.5 billion. (Vietnam is third, with those three country suppliers largely replacing U.S. imports from China.) (US AI Server imports)

Servers and related hardware made up almost one-fifth of the $317 billion of goods Mexico exported to the United States between January and May 2026. This reflected a more than 100% increase compared to the same period a year earlier. (Mexican AI server exports)

Although specific import/export data is not available, it appears that at least some companies producing AI data servers in Mexico are exporting them to Canada as well as to the United States. Nvidia is reported as importing servers for its Canadian operations from Mexico under the USMCA. (Canadian server imports) Although volumes are not available, Nvidia recently announced that it had partnered with Ciara Technologies to assemble some AI servers in Montreal. (Nvidia Canadian server production) AI data center operators in Canada including Nvidia have greater sourcing flexibility since Imports of servers from third countries such as Taiwan, unlike into the United States, are generally duty free under the WTO’s Information Technology Agreement. (Zero Server Tariffs)

In the short and medium-term at least, server exports have thus helped to keep the Mexican economy from tanking. (Mexico's economic outlook) However, the server production phenomenon in Mexico raises major questions. Can the mostly Taiwanese new investment be sustained despite such concerns as Mexico's poor investment climate and the uncertainties that are a product of Mr. Trump's tariff policies, including the uncertain future of the United States-Mexico-Canada Agreement (USMCA)? Can server production help to replace lost automotive jobs? Will a significant percentage of the materials and components required for server production ever be sourced in Mexico, or will most of the value added continued to be imported from Taiwan, China, and the United States? Finally, what happens to the industry in the unlikely event that U.S. demand for new data centers diminishes because of electric power shortages or public opposition or, worse, if the AI craze turns out to be a "bubble" that bursts sooner rather than later? I seek to provide at least partial answers below.

Mexican Server Production and Exports

Development of Mexico as a major source of server production for U.S. AI data centers is a recent phenomenon. (Mexico: cornerstone of American AI boom) The servers, specialized computers used to store, manage, and process data inside AI data centers, are the major building blocks for data centers. “An AI data center is purpose-built … to handle AI training and inference.” AI workloads are estimated to require ten times more computational power than traditional applications. Three thousand AI centers are estimated to be under construction or planned globally by 2030, representing an estimated $7 trillion in total investment, with $3 trillion in the United States. For example, Texas currently has more than four hundred projects in the pipeline. (AI data centers)

Several considerations appear to have encouraged nearshoring of AI data server production to Mexico. Taiwanese enterprises, the principal source of Mexican server investment, were not surprisingly attracted by lower tariffs on exports to the United States, close physical proximity to the rapidly expanding U.S. market for data centers (and other high-tech electronic components), the desirability of diversification away from Taiwan, and lower labor costs for competent workers. (Taiwan nearshoring to Mexico) In particular, for most AI server exports to the United States, tariff treatment is favorable, with no specific regional value content requirement under USMCA rules or origin. (Server Tariff treatment) For Mexico, the disadvantage of this treatment is that there is less incentive for foreign producers to seek local components when imported components can be used without affecting duty-free entry.

Significantly, Taiwanese enterprises such as Foxconn decided to locate manufacturing facilities in Mexico in part because American tech giants including Amazon, Google, Microsoft, and Nvidia strongly encouraged Taiwanese enterprises to establish server production there. (Encouraging Server production in Mexico)

Among the logistical advantages, supplying necessary third-country parts and components to factories in Ciudad Juárez and other border cities is relatively straightforward: a container unloaded at Los Angeles or Long Beach can reach Mexico in less than two days. Shipping the heavy servers to data centers in most of the United States by truck is also manageable. Like many other Asian enterprises that have established manufacturing operations in Mexico, Taiwanese businesses do not seem overly worried with rule of law issues in Mexico although some concerns have been expressed. The advantages of co-production with Texas are also considered important. (Texas Co-production)  Other challenges, such as electric power and water shortages, could become more of a problem for new Taiwanese and other investments in the future.

Many of these same considerations have encouraged substantial direct foreign investment in Mexico for many years, despite concerns over crime, corruption, poor infrastructure, electricity shortages and a weak judiciary and regulatory uncertainty. (Mexico's Investment Climate)

Developing Mexican Supply Chains

A major challenge for Taiwanese server enterprises and for the Mexican economy itself is developing component production in Mexico so imports can be reduced and more local jobs created. The North American automotive industry, because of integration of Mexico with the United States and Canada, has done remarkably in producing parts and components by more than twelve thousand enterprises both large and small. (North American auto parts producers) However, historically, the electronics industry in Mexico has not done nearly as well. Consequently, flat screen TV producers (e.g., Visio in Baja California) and laptop producers (e.g., Lenovo in Nuevo León) must rely on a limited if slowly growing local supply chain. (electronics supply chain) Currently, as Christine Smith and Alan Murray for the Financial Times have noted,

Ciudad Juárez, across the Rio Grande from El Paso in Texas, is at the heart of the [AI server] industry, building off the border town’s existing Taiwanese factories that have been making simpler electronics for decades. The new production lines assemble parts shipped from Asia into AI servers, which are stacked in refrigerator-sized racks that can weigh more than two tonnes each and are then trucked across the border. (server assembly)

Taiwanese producer Pegatron is reported to have five assembly plants in Ciudad Juárez. It is arguably in everyone's interest to encourage relocation of Taiwanese and other Asian parts production for servers (and other electronic products) to Mexico, to lower costs and shorten supply chains, among other considerations.

Impact on Employment in Mexico

The Mexican automotive industry lost an estimated 320,000 jobs (out of almost two million), 2024-2025, due not only to tariffs but to production of EVs (using far fewer parts than gasoline-powered vehicles), increased automation and continued reliance on vulnerable Asian suppliers. (Employment decline)  By the end of the first quarter of 2026, exports of autos and auto parts to the United States declined by $4.87 billion, or by 11% compared to 2025. (Auto exports decline)

In my view, there is no reason to believe that major auto and auto parts producers for the U. S. market will fail to respond to the Trump administration's demands that they relocate at least some operations to the United States. Should USTR Jamieson Greer convince Mexico to accept a new 50% US content requirement for autos exported to the United States, Mexico's industry employment is expected to continue to decline, as Mexico currently accounts for 42% of all U.S. auto parts. (Fifty percent U.S. Content Requirement) Thus, a key factor is the extent to which it is feasible to shift and retraining auto workers for server production or other employment (which as noted below seems unlikely).

Unfortunately, reliable employment data are not available as of this writing. Fernando Alba, deputy economy minister in Chihuahua state, where Ciudad Juárez is located, is quoted in the Financial Times as observing, "One truck [of servers] is the equivalent of one thousand cars.” Unfortunately for the Mexican labor force, one thousand cars require far more workers than a truckload of servers produced in highly automated plants. Data provided by experts to the Financial Times’ reporters indicates that under current conditions that server production could realistically aspire to reach 3-7 per cent Mexican content, primarily regional direct labor, facility overhead, and localized logistics, compared to 39 per cent for autos. Nor does there appear to be any significant Mexican impact on product design. (Mexican content in servers) The director of the Claudio X. Gonzalez Center for the U.S. and Mexico, Tony Payan, recently shared with me his on-site observations that manufacturing in Ciudad Juárez began a decline before the beginning of the second Trump administration and is continuing despite expanding AI data server production.

Still, some auto industry employers might find employment in the data server industry. Workers with experience in assembly, production, quality control, maintenance, and supervisory roles could be transferrable, according to Luis Ricardo Rodriguez, managing director of Monarch Global Industries in Monterrey who counsels multiple foreign investors in Mexico. He notes that the capacity to absorb displaced automotive workers depends heavily on the type of operation. Server assembly, final-device manufacturing, and some back-end semiconductor activities are more labor-intensive, while semiconductor fabrication and other advanced manufacturing processes are significantly more capital- and skill-intensive. Many positions in data server production positions would require technical retraining, specialized skills, and, in some cases, English proficiency. As a result, these investments could absorb part of the displaced workforce, but probably not automatically or on a one-for-one basis.

Nevertheless, in time it seems reasonable to expect that higher Mexican component content could gradually be achieved, some of it much more labor-intensive than AI server assembly. A substantial share of the value incorporated into Mexican electronics exports is still generated abroad, with China, Taiwan, Korea, Malaysia, and Singapore remaining key suppliers, particularly of high-value components. But the domestic supplier ecosystem is gradually expanding, according to Rodriguez. Some multinational electronics companies already source a meaningful portion of their supply chain locally, in areas such as printed circuit boards, substrates, wiring, packaging materials, metal components, and other back-end manufacturing inputs.

As far as broader North American sourcing is concerned, it also seems reasonably possible that TSMC, a giant Taiwanese semiconductor producer with more than $265 billion of actual and planned investment in Arizona, will eventually be supplying some of the chips necessary for server production in Mexico from the United States, although currently most imports are from Asia. (https://www.taipeitimes.com/News/editorials/archives/2026/07/29/2003861544). Other American chip input to server assembly in Mexico could also be significant as eight of the world’s top chipmakers—Nividia, AMD, Intel, Alphabet, IBM, Meta, Broadcom and Qualcomm—are U.S. companies, although many of the chips they make are not currently produced in the United States. (U.S. computer chip producers) One estimate suggests that a single AI server rack contains at least 4,500 chips accounting for as much as 95% of the content value of a server ranging from inexpensive foundational chips to unique integrated circuits. (Chip value in servers)  (Chip varieties) Among other important considerations, sourcing more of the chips from the United States may be feasible if increased incentives for domestic production and for upstream materials are implemented. (U.S. Production Incentives) An added political and economic benefit from more U.S. sourcing would be reducing Mexico’s trade surplus with the United States.

Nor are U.S. policies regarding automotive industry imports likely to be extensively reversed in 2029, regardless of which political party gains the presidency. The U.S. auto unions, which have broad influence on many Republican as well as Democratic politicians in the United States, oppose Mexican imports in significant part because of much lower Mexican labor costs (e.g., at General Motors’ Mexican plants workers reportedly earn $25 per day compared to $18-$32 per hour in the United States. (Auto worker wages)  U.S. unions, which have opposed freer trade for decades, support higher tariffs on autos and auto parts imported from Mexico and Canada. (Union support for tariffs) Consequently, while server production can at least narrow or bridge the gap in the overall value of Mexican exports to the United States, it will probably not have a major impact on longer-term unemployment in the Mexican auto industry.

Risks for Mexico

The likelihood that the Trump administration would limit Mexican-sourced server imports seems low, given the importance of major AI enterprises such as Meta, Microsoft, Google, OpenAI and Nvidia to the administration and to the American economy. (AI providers' power)  However, it is not out of the question given U.S. concerns with the increasing trade deficit with Mexico and Mexican reliance on third-country inputs. U.S. actions detrimental to Mexican electronics exports, rational or not, could occur because of their continued reliance on parts and components imported from China and other Asian countries. (trade circumvention actions)

Another risk in my view is a decline in U.S. demand, either because of the bursting of an AI "bubble" warned against by some observers, or simply a decrease in demand in future years as new data centers are no longer being built. (AI Bubble) It is difficult to assess the chances of a bubble because of overbuilding, but overall, the installation of new AI data centers in the United States seems more likely to continue increasing than decreasing in the foreseeable future. It has been suggested that the AI facilities buildup “is the largest scale infrastructure build-out in the history of humanity,” compared in scope to the railroads in the 1800s and Roosevelt’s New Deal in the 1930s. AI investment in infrastructure is forecast to increase globally from $318 billion in 2025 to over $1 trillion in 2029. (AI Build-out)

Still, a temporary slowdown in U.S. AI data center construction could occur because of two factors. First, growing citizen opposition, reflecting NIMBY along with rising consumer electric bills and high center water usage, has led to moratoriums on new construction. New York was recently the first state to order a state-wide one-year construction moratorium. (Data Center Moratoriums) Secondly, it has also been predicted that data center power demand will increase from 31 GW in 2025 to 66 GW in 2027. (Data Center power demands)

Increasing AI server demand may also occur because of continuing upgrading of servers, for example to reduce electric power requirements or speed up computing operations. Thus, the situation is again quite different from the mature automotive industry, where US demand for new cars and small trucks is relatively stable at 15-17 million in recent years (rather than increasing), despite a small decline 2026 over 2025. (U.S. new car sales)

Any of these scenarios for future AI data server demand could have a negative impact on all players in the AI industry in the United States and on suppliers of the servers as well as the chips and other components. The impact on Mexico's exports could be material. The risk suggests that it is urgent for the Mexican government, working with Taiwanese investors, to take steps once again to try to develop a broader group of Mexican-based components suppliers, which presumably could supply electronic parts and components not only to server producers but to other producers of electronic products in Mexico. The fact that Taiwanese companies are investing in other electronic sectors, encouraging a "strategic axis in advanced manufacturing" between the two countries, is a positive factor for Mexico's economy. (https://mexicobusiness.news/trade-and-investment/news/mexico-taiwan-deepen-strategic-axis-advanced-manufacturing) This not only results in technology transfer and job creation, but provides an important alternative to mainland Chinese investment that is no longer politically feasible. (Mexico's China Problem)

Conclusions and Recommendations

The boom in AI server production and exports is not a solution to Mexico’s job losses because of Mr. Trump’s trade policies, even though it may replace much of the export revenue lost in the automotive sector. However, in my view it presents Mexico with a golden opportunity to move a large segment of Mexican manufacturing up the supply chain. As noted earlier, only 3-7% of the value of the AI data servers represents Mexican inputs; the rest are Taiwanese, Chinese, other Asian, or American. No one expects Mexico to become a manufacturer of chips in the foreseeable future; Mexico lacks the engineering talent and legal stability among other factors. Still, Taiwanese companies, working with the Mexican government, could in time raise the Mexican content of the servers, perhaps significantly if they believe the investment climate is sufficiently favorable. Mexico could also strongly encourage Taiwanese producers in Mexico to source their server chips from the United States rather than Taiwan or elsewhere in Asia, to the extent feasible. In this respect at least, American, and Mexican objectives are consistent: it is in the United States’ as well as Mexico’s interests to continue to encourage increased AI chip production in the United States, whether for use in AI servers or otherwise.

Whether the Sheinbaum administration has the willingness and the political ability to improve the business climate to the point where substantial additional direct foreign investment in component production, from Taiwan and elsewhere, occurs is problematic. In the view of some observers, President Sheinbaum’s predecessor, Andres Manuel Lopez Obrador, squandered many of the potential benefits of the nearshoring boom that began in 2017 or before by actively discouraging new direct foreign investment, particularly in the energy sectors. (Undermining Mexico's Favorable Investment Climate) After nearly two years as president, Sheinbaum in my view has mostly made things worse, replacing Mexico’s appointed judges with popularly elected ones, emasculating Mexico’s independent regulatory agencies, and accelerating Mexico’s movement toward becoming a one-party state, as it was before 2000. (Single-Party governance) Thus one can hope but not really expect the current Mexican government to be proactive in improving the investment climate, in this instance giving the AI server industry no viable alternative to importing most of the components from outside Mexico, as is the case today.