The EU-Mercosur Trade agreement, which took effect on May 1, eliminates tariffs on nearly all goods between the European Union and Argentina, Brazil, Paraguay, and Uruguay. The headlines have focused mostly on agriculture and industrial tariffs. Less well understood is how the accord reinforces or constrains Brazil's national artificial intelligence (AI) plan, the Plano Brasileiro de Inteligência Artificial (PBIA), and its ambitious technology policy launched in July 2024. The answer is mixed in a way that reflects the era in which the agreement was negotiated.
The PBIA is titled "AI for the Good of All." It aims to set Brazil up as a global leader in AI by investing in resources and data centers, training and educating Brazilians, promoting inclusivity and diversity, applying AI to public services, encouraging innovation and competitiveness, and ensuring that regulation promotes ethical and responsible use of AI, including with the creation of a National Center for Algorithmic Transparency and Trustworthy AI.
The EU-Mercosur agreement leaves out mention of AI training data, regulation, technical standards, or market access
There's not much in the EU-Mercosur agreement that would help Brazil achieve these goals. The deal says nothing about training data for large language models (LLMs) and other AI applications, regulation, technical standards, or market access for AI products and services. Brazil faces imminent policy decisions on whether AI systems can be trained on publicly available data without individual consent, whether open-source foundation models must meet the same compliance thresholds as proprietary ones, and whether the government can deploy automated decision-making in public services such as tax collection and health care.
The European Union has struggled with these same issues, guided by its own market structure, privacy concerns, and regulatory traditions. Europe's strict privacy and data-protection regulations have rankled the Trump administration and US companies, drawing complaints that they stifle competition and innovation and impose unnecessary costs on US companies seeking to operate in Europe's markets. How these concerns will affect Brazil remains unclear, since nothing in the EU-Mercosur addresses them.
EU-Mercosur's strongest contribution to Brazilian AI lies in the physical infrastructure and critical minerals that AI systems depend on. Brazil has a big advantage in this area. As I have argued in a previous piece, the agreement establishes the first binding legal framework for the supply of critical minerals between Europe and South America. The investment protections, dispute settlement mechanisms, and regulatory predictability that it establishes for refining lithium in Argentina or processing rare earths in Brazil apply equally to the energy-intensive facilities that AI requires.
Brazil's clean electricity grid, with renewables accounting for some 85 percent of generation, is a genuine comparative advantage for hosting data centers and supercomputing facilities. For European firms operating under increasingly strict environmental and corporate sustainability reporting obligations, there are substantive reasons to locate such infrastructure where the grid is largely decarbonized. Furthermore, the agreement's investment provisions reduce risk for investors and offer multi-decade return horizons.
The EU-Mercosur's legally binding framework and its investment provisions could pave the way for greater AI-related investments in Brazil, aligning somewhat with the PBIA's intended objectives. The PBIA's roadmap includes expanding national supercomputing capacity and supporting sovereign data infrastructure, which will require foreign capital and the enactment of technology partnerships. The critical minerals that the agreement encourages Brazil to process domestically, rather than ship as raw materials, are inputs to the hardware AI requires.
But the picture changes when it comes to issues raised by AI regarding data accessibility, the regulation of foundation models, technical standards, and cross-border services. The text of the EU-Mercosur agreement simply does not address these issues. There is no chapter on digital trade, no commitments on cross-border data flows, no language on source code or algorithmic disclosure, and nothing that is specific to AI systems. By contrast, the Comprehensive and Progressive Agreement for Trans-Pacific Partnership (CPTPP), the United States-Mexico-Canada Agreement (USMCA), and the European Union's own digital partnerships with Japan, Korea, and Singapore all contain such provisions.
Unaddressed regulatory issues may keep future Brazilian AI products out of the EU market
The failure of the EU-Mercosur accord to address these issues could create disagreements down the road. Brazilian firms hoping to serve the European market with AI products and services receive no locked-in market access through the agreement. Instead, they remain dependent on European unilateral decisions on data adequacy under the European Union's General Data Protection Regulation (GDPR) and on the implementation of the EU AI Act, which imposes strict rules on the handling of users' data to protect privacy and security and prevent exploitation, with heavy fines for violations. The rules are extraterritorial and apply to any system whose outputs are used in the European Union. Whatever Brussels decides about which AI applications face the heaviest regulation, what testing developers must complete before selling in the European market, and what obligations apply to the large base models that underpin most AI services will continue to determine whether Brazilian AI products can reach European customers. These requirements may not fully align with Brazil's ambitions.
On the other hand, Brazil's regulatory autonomy is under pressure from another direction. Brazil's General Data Protection Law (LGPD) was modeled on the GDPR, and the AI bill currently before Congress (PL 2338) draws heavily on the EU AI Act's risk-tier framework, which sorts AI applications into regulatory categories based on how much harm they could cause and imposes escalating compliance requirements accordingly. Regulatory alignment could yield deeper economic integration with the European Union as Brazil comes under political and commercial pressure to further converge with European rules.
Why regulatory alignment with the EU could hurt Brazil's nascent AI sector
The practical consequences of that potential convergence deserve scrutiny. Consider training data: The GDPR's consent-based framework and the EU AI Act's documentation obligations for training datasets are among the most commercially restrictive provisions in European AI regulation. Japan, by contrast, adopted a broad exception permitting the use of copyrighted material for AI training, a choice that has materially benefited its AI developers. Brazil could pursue a similarly permissive regime suited to its stage of development, but convergence with European rules would foreclose that path.
Or consider the obligations imposed by the EU AI Act on general-purpose AI models, such as transparency requirements, adversarial testing, and energy consumption reporting. These provisions are designed for firms with the resources of a large tech company, like Meta or Mistral, the French AI company. A Brazilian startup building a Portuguese-language model for the narrow purpose of agricultural extension or public health surveillance would face the same compliance architecture but would be handicapped by having only a fraction of the capacity. Even government deployment of AI is affected: Brazil might wish to move quickly on automating tax administration, environmental monitoring, or healthcare triage, but the EU AI Act's high-risk classification framework subjects many such applications to extensive conformity assessments before they can be deployed.
For an AI ecosystem at Brazil's incipient stage of development, with limited domestic supercomputing capacity, a thin private-sector AI market, and most of its leading research talent concentrated in a handful of institutions, the case for adopting Brussels' regulatory architecture wholesale is problematic. Convergence with European rules and standards may protect Brazilian consumers, but the compliance costs would be formidable. European firms are better resourced to absorb these costs than Brazilian startups. But the concern in Brazil is that this imbalance could foreclose experimentation with regulatory approaches better calibrated to local conditions.
The EU-Mercosur agreement also could also limit Brazil's ability to use industrial policies like subsidies, government procurement, and investment screening for its AI sector. The accord could also narrow the scope for local-content requirements on AI infrastructure, preferential procurement for domestic AI companies, or targeted support for sovereign data center capacity. True, Brazil's experiences with industrial policy are fraught with excesses, inefficiencies, and corruption, suggesting that such limits may be a good thing. But Brazil might also benefit from having a bit more room for maneuver. That said, the PBIA's budget of some $4 billion is small enough that none of these constraints is currently binding. However, should Brazil ever wish to scale its AI industrial policy to something resembling the US CHIPS Act or the EU Chips Act in spirit, the EU-Mercosur deal has the potential of introducing new frictions.
Brazil must look beyond potential partners—restrictive EU, extractive China, unpredictable US—and fashion its own strategy for growing its AI sector
The EU-Mercosur agreement, on its own, could constrain Brazil's AI ambitions, and the country needs a complementary strategy to nurture its AI sector. That strategy will require partnerships with jurisdictions that offer what Europe cannot, including access to advanced semiconductors, frontier model expertise, and AI services market access on terms that go beyond what unilateral European decisions allow. The countries Brazil needs to partner with to advance its AI agenda are familiar, namely, the United States, South Korea, and Japan. The difficulty is that the United States, Brazil's most plausible technology partner, is unlikely to offer a comprehensive technology agreement under the current administration. Tariff threats, executive-branch unpredictability, and the absence of a coherent strategy for Latin America leave Brazil with diminished options.
This concern parallels the one I raised regarding critical minerals. For AI specifically, the United States offers neither a binding agreement nor an exportable regulatory model benefitting Brazil or its neighbors. Latin American countries that wish to participate in AI value chains beyond the extraction of inputs—data, critical minerals, and energy—are being forced to choose between the European model, which is regulated, predictable, but restrictive, and the Chinese model, which is capital-rich, opaque, and extractive.
For Brazil, the EU-Mercosur agreement could unleash European capital flows to the region. But Brazil's ambitions, embodied by the PBIA, need an architecture that does not yet exist. Building it will challenge whoever occupies the Brazilian presidency after the elections at the end of 2026.
Data Disclosure
This publication does not include a replication package.
