Quick Answer:
Mexico's Coatlicue supercomputer is advancing toward a planned capacity of 314 petaflops using 14,480 GPUs across 7,500 chassis — far above the country's current 2.3 petaflops. Announced November 26, 2025 with a 6-billion-peso investment, it runs on an interim basis at the Barcelona Supercomputing Center while the permanent site is built.
Key Takeaways:
- Planned capacity: 314 petaflops via 14,480 GPUs across 7,500 chassis, housed in roughly 200 cabinets, according to Mexico Business News and Interesting Engineering.
- Regional leap: Mexico currently peaks at 2.3 petaflops; Coatlicue would also surpass Brazil (13.7 petaflops) and Argentina (12.6 petaflops), per Interesting Engineering.
- Investment and announcement: 6 billion pesos (approx. US$327 million), announced November 26, 2025, according to Mexico Business News.
- Interim operation: while the permanent Mexican site is under construction, the system runs initial programs with temporary access to the Barcelona Supercomputing Center.
- Stated use cases: climate forecasting, seismic processing for PEMEX, fiscal analysis for the SAT tax agency, agricultural satellite monitoring, and Spanish-language AI models, according to Mexico Business News.
If you run a business in Houston, Cypress, Guadalajara, or Bogota, you have probably already seen headlines about Mexico's bet to become a regional AI computing hub. Coatlicue — named after the earth mother goddess in Mexica mythology — is the project carrying that narrative, and this week it moved from political announcement to under-construction infrastructure with verifiable technical numbers behind it.
This is not a vague "we are working on AI" promise. According to Mexico Business News, the project has a public budget (6 billion pesos), an announcement date (November 26, 2025), and concrete technical specifications. Interesting Engineering confirms the same hardware figures and adds the regional context that shows just how large this leap is.
What Coatlicue Is and Why the Name Matters
Coatlicue takes its name from the Mexica goddess of earth, life, and death — a symbolic choice for a project the Mexican government presents as the engine of its next generation of digital infrastructure. According to Mexico Business News, the system was announced on November 26, 2025 and is positioned to become the most powerful computing system in Latin America once it comes online.
Unlike earlier Mexican AI announcements that lacked technical detail, Coatlicue arrives with published hardware specifications: 14,480 graphics processing units (GPUs) spread across 7,500 chassis, housed in roughly 200 cabinets, according to both sources. The infrastructure requires high-density electrical systems, water-based cooling, and high-speed connectivity — the same requirements any AI data center at this scale demands anywhere in the world.
The Numbers Behind the Compute Leap
The headline figure is 314 petaflops of designed capacity — 314,000 trillion operations per second, according to Mexico Business News. To size just how large a leap that is, Interesting Engineering provides the regional context: Mexico's current peak capacity is just 2.3 petaflops. Brazil currently operates at 13.7 petaflops and Argentina at 12.6 petaflops. Coatlicue, if completed as planned, would not just close Mexico's gap with its regional neighbors — it would surpass them by a considerable margin, more than twenty times Brazil's current capacity.
Put in global perspective, the world's leading system, El Capitan, in the United States, operates at 1.809 exaflops — nearly six times Coatlicue's planned capacity. Mexico is not competing for the global top spot; it is competing to stop depending on foreign infrastructure for its own AI workloads, climate science, and government analysis.
The number that gives context: going from 2.3 to 314 petaflops is not an incremental upgrade — it is a leap of more than 130 times the country's current computing capacity, based on the figures published by Interesting Engineering.
Timeline, Investment, and Interim Operation
The project was announced on November 26, 2025 with an investment of 6 billion pesos — around US$327 million, according to Mexico Business News — funded through a staged construction process. The same source reports that the decision on the permanent installation site was made in January 2026, and that construction is expected to take roughly 24 months.
While the physical installation in Mexico is finalized and built, the system is not idle. According to Mexico Business News and Interesting Engineering, Coatlicue operates on an interim basis with temporary access to the Barcelona Supercomputing Center, which allowed initial programs to begin running in January 2026. This collaboration with the Spanish center — part of the European Union — gives Mexico a path to start generating computing results before its domestic infrastructure is ready, rather than waiting the full 24 months with no progress at all.
What the Government Plans to Use This Capacity For
The use cases reported by Mexico Business News are not limited to abstract AI research. They include concrete applications within the Mexican government itself:
Reported use cases for Coatlicue:
- PEMEX: processing seismic and geological datasets
- SAT (tax agency): tax declarations, audits, trade flow analysis, and customs operations
- Climate science: extreme weather forecasting and disaster-risk modeling
- Agriculture: processing more than two million satellite images for soil, drought, and irrigation monitoring
- Public health: epidemiological modeling and genomic analysis
- AI development: language models and virtual assistants in Spanish and Indigenous languages
This list matters because it reveals the project's real intent: it is not just about bragging rights for the region's largest supercomputer, but about building compute capacity that Mexican federal institutions — from the tax agency to the state oil company — can use directly for their day-to-day operations.
A National Cluster and International Collaboration
Coatlicue is not arriving into an infrastructure vacuum. According to Mexico Business News, Mexico already operates a National Supercomputing Cluster with a combined capacity of 9.45 petaflops and more than 10,000 terabytes of storage, made up of systems at several institutions: Yuca at Universidad de Sonora, Leo Atrox at Universidad de Guadalajara, Xiuhcoatl and Abacus at the Instituto Politecnico Nacional, and Miztli at UNAM, with geographic presence spread across Baja California, Sonora, Mexico City, Puebla, Jalisco, Nuevo Leon, and Chiapas.
The project also draws on international collaboration beyond the Barcelona Supercomputing Center. Mexico Business News reports an agreement with India's Centre for Development of Advanced Computing for technical workforce training — a necessary component for Mexico to operate and maintain a system at this scale once it comes online domestically.
What This Means for Your Business in the US and LATAM
Here is the part that often gets lost in coverage of state infrastructure: when a government announces billions of pesos in AI compute to train Spanish-language models, that announcement is a signal of where consumer behavior is heading — not a guarantee that your business benefits automatically.
The Spanish-language models Coatlicue will help train will feed, directly or indirectly, into the same AI systems your customers already use to search for where to buy, hire, or get a quote: ChatGPT, Perplexity, Google AI Overviews. If your business is not structured for those systems to find and confidently cite you, the expansion of AI infrastructure across the region does not help you — it helps your competitor, who did put in the work to be visible to those answer engines.
This is exactly the gap MerchandisePROS's AI Search Optimization (AEO) service closes: structuring your site, your schema data, and your authoritative citations so ChatGPT, Perplexity, and Google AI Overviews name you when a customer asks for your service in your city — instead of naming a competitor who already did the visibility work.
"Mexico is investing billions in the infrastructure that will decide how your customers find you. The question is not whether that infrastructure is coming — it is whether your business is already visible to it."
- Diego Medina F, Founder of MerchandisePROS
Frequently Asked Questions
What is Mexico's Coatlicue supercomputer?
Coatlicue is Mexico's federal public supercomputer, named after the Mexica earth mother goddess. According to Mexico Business News and Interesting Engineering, it is designed for a capacity of 314 petaflops via 14,480 GPUs spread across 7,500 chassis, which would position it as the most powerful system in Latin America.
How much computing capacity will Coatlicue have compared to the rest of the region?
According to Interesting Engineering, Mexico's current peak capacity is 2.3 petaflops, compared to Brazil's current system at 13.7 petaflops and Argentina's at 12.6 petaflops. Coatlicue's planned 314 petaflops would far exceed all three. As a global reference point, El Capitan, the leading US system, operates at 1.809 exaflops.
How much will Coatlicue cost and when was it announced?
According to Mexico Business News, the project was announced on November 26, 2025 with an investment of 6 billion pesos (approximately US$327 million), funded through a staged construction process.
Where does Coatlicue operate while the permanent site in Mexico is being built?
While the permanent installation on Mexican soil is finalized and built, the system operates on an interim basis with temporary access to the Barcelona Supercomputing Center, according to Mexico Business News and Interesting Engineering.
What does Mexico plan to use Coatlicue's capacity for?
According to Mexico Business News, the stated use cases include climate forecasting and disaster-risk modeling, seismic and geological data processing for PEMEX, fiscal and trade analysis for the tax agency SAT, agricultural monitoring using more than two million satellite images, epidemiological and genomic analysis in public health, and the development of Spanish- and Indigenous-language AI models and virtual assistants.