Chinese firms win 2.3 billion reais AI contract in Brazil; Huawei and iFlytek team up to build Latin America’s computing hub — BigGo Finance

Brazil’s government recently announced an AI investment plan totaling approximately 2.3 billion reais (approximately $444 million), of which roughly 1.3 billion reais will go toward a supercomputing infrastructure project in Rio de Janeiro, to be jointly undertaken by Chinese companies Huawei and iFlytek. This is one of the fastest-moving national-level projects since the establishment of the World AI Cooperation Organization (WAICO), marking a formal upgrade in China’s AI overseas expansion from single-product exports to systematic co-development of national-level infrastructure and large model ecosystems.

Under the plan, Huawei will be responsible for the underlying computing infrastructure, while iFlytek will lead development of speech, natural language processing, and Portuguese-language large models. The two parties will also provide supporting cooperation including technology transfer, localization adaptation, and talent development, with the goal of training 3,000 Brazilian professionals during the implementation period. Project funding comes from Brazil’s National Fund for Scientific and Technological Development (FNDCT), with launch expected in July 2027.

The Brazilian government has made clear that the project’s core objective is to enhance the country’s AI computing capabilities and data sovereignty, reducing dependence on any single country, single company, or single technology pathway.

A “second option” in the global AI landscape

The landing of this Brazil project comes at a critical juncture as Chinese AI accelerates its penetration of global markets. According to CNBC data, Chinese open-source models represented by DeepSeek have captured 46% of token usage by U.S. enterprises on the OpenRouter platform, with usage frequency of Chinese large models among overseas developers and enterprises continuing to climb.

At the same time, Brazil’s government also plans to build another AI supercomputer in the northeastern state of Rio Grande do Norte through public tender, with U.S. chipmaker Nvidia expected to be the primary supplier. Dividing the supercomputing projects between Chinese and American companies is seen as a carefully calibrated move by Brazil to seek strategic balance between major powers and avoid over-dependence on any single technology system.

iFlytek Chairman Liu Qingfeng has repeatedly emphasized that the world needs China to provide a “second option” in artificial intelligence more than ever before. The landing of this Brazil project represents the first validation of that judgment at the national infrastructure level.

The synergy logic between Huawei and iFlytek

Brazil’s selection of Huawei and iFlytek among numerous Chinese AI companies stems primarily from the deep “computing foundation + large model application” synergy the two companies have built over the past several years.

Huawei’s role is to provide the underlying computing support. Huawei has deep roots in Brazil spanning nearly three decades, having opened its first overseas representative office in Latin America there in 1999, with its equipment covering 40% of the country’s core networks. On the AI computing front, Huawei’s Ascend chips have developed complete capabilities ranging from single-node to cluster-level computing. The open-source Pangu openPangu 2.0 model released at Huawei’s 2026 Developer Conference features 505 billion total parameters, deeply integrated with Ascend computing and the HarmonyOS ecosystem.

iFlytek’s core value lies in its fully self-controlled, Chinese-made large model stack. iFlytek Spark is China’s first large model trained entirely on a domestically produced computing platform, with multilingual capabilities supporting more than 130 languages. In key languages for Latin America and ASEAN markets, some metrics have already surpassed comparable U.S. products. In the first half of 2026, iFlytek’s overseas revenue grew 160% year-over-year, forming an important new engine of revenue growth.

The critical breakthrough in the two companies’ integration of Chinese-made computing power and large models occurred in March 2025. iFlytek, working with Huawei, became the first to achieve large-scale cross-node expert-parallel cluster inference for MoE models on Chinese-made computing clusters, reducing single-card static memory usage to one-quarter of dual-machine deployment, improving efficiency by 75%, boosting inference throughput by 3.2 times, and cutting end-to-end latency by 50%. This was the industry’s first entirely new solution based on Chinese-made computing power, proving that domestic computing not only works but is getting better with use.

In September 2025, Huawei and iFlytek jointly launched the “Spark Education and Healthcare Large Model Scenario Appliance Solution,” deeply integrating Ascend computing with the iFlytek Spark large model. The Spark Appliance released in August 2023 enabled all enterprises to deploy large models privately on Chinese-made, self-innovated platforms.

From “selling tokens” to “building ecosystems”

The path of Chinese AI overseas expansion is undergoing a fundamental transformation.

In the past, Chinese AI exports sold API calls—customers used them and left, and could be replaced at any time by cheaper models. But building AI infrastructure is different. Brazil’s Folha de S.Paulo reported that the project involving Huawei and iFlytek will develop large language models similar to ChatGPT and Claude, with a focus on enhancing Portuguese-language support while also covering Spanish-language applications. The collaboration between the two parties extends beyond computing and model layers into deep local talent development and scenario penetration.

Goldman Sachs noted in a recent research report that China’s overseas expansion is no longer about goods themselves, but rather an AI-driven, technology-led industrial capability—from power infrastructure for AI data centers at the top, down to robotics and physical AI at the bottom.

The strategic backdrop for this shift is the Agreement on the Establishment of the World AI Cooperation Organization, signed in Shanghai on July 16 this year. Twenty-nine countries became founding members, spanning four continents: Asia, Africa, Latin America, and Europe. In just over a month, membership has expanded to 38 countries. The China-Brazil Rio de Janeiro Supercomputing Center project is the first officially implemented physical cooperation project under the organization’s framework.

From the agreement signing in Shanghai to the infrastructure planning in Rio, in less than 40 days, Global South countries completed a critical leap in practical AI cooperation—from consensus to implementation.

Latin America’s structural opportunity

As Latin America’s largest economy with a population exceeding 200 million, Brazil possesses natural advantages for building a regional computing hub. The country ranks among the top three globally in weekly ChatGPT usage, and ranks second worldwide in active developers building products on the OpenAI API. A Microsoft study shows that 75% of Brazilian small and medium-sized enterprises are optimistic about AI’s impact.

However, Latin America has long lacked independently developed foundational models, with public services and commercial ecosystems heavily dependent on U.S. cloud infrastructure. According to the 2023 Latin American AI Index jointly published by the UN Economic Commission for Latin America and the Caribbean and Chile’s National Center for Artificial Intelligence, intra-regional AI research collaboration accounts for only 7.7% of total activity in Latin America, with most technical cooperation relying on external markets.

Brazil’s Ministry of Science, Technology and Innovation previously issued a National AI Investment Plan, allocating approximately $4.07 billion for 2024–2028 to build a national sovereign cloud, Portuguese-language large models, and digital governance systems. But domestic investment alone cannot overcome the triple constraints of technology, capital, and regulation.

Celso Amorim, chief special advisor to Brazil’s president, said that AI cooperation with China represents a path for Global South countries to deliver a “measured and reciprocal response” to new technological challenges.

For iFlytek, Brazil’s strategic value is particularly pronounced. Brazil is one of the world’s largest Portuguese-speaking markets, and this cooperation will advance iFlytek’s positioning in Portuguese-language large models and the Latin American AI market. Leveraging this project, iFlytek can localize the full “computing foundation + private large model + industry deployment” model already proven in China, refining a standardized, replicable solution for Global South countries.

From a commercial competition standpoint, the successful landing of large-scale overseas national projects also serves as reverse validation of the engineering capabilities and security value of the fully Chinese-made large model system, feeding back into the competitive landscape of the domestic market.

When intelligence is no longer defined by a few actors, diverse technological choices can unleash greater productivity. Brazil’s choice is writing a new footnote to the narrative of Chinese AI’s globalization.