AI.CUBE signs $50 million deals with AWS and Oracle to boost AI capacity in Brazil and Latin America

2–3 minutes

Contracted infrastructure

Brazilian AI startup AI.CUBE, founded in 2024 and based in Silicon Valley, signed two agreements in November worth a total of $50 million with Amazon Web Services (AWS) and Oracle. The contracts provide cloud processing power to run AI agents, with most of it intended for operations in Brazil and some reserved for expansion in Latin America.

Details of the agreements

The AWS agreement provides three years of access to Nvidia‘s B200 GPUs. The Oracle contract is for five years and includes Nvidia‘s GB200 systems. The startup did not disclose the individual value of each contract or the number of chips it will have available.

Objectives and use of capacity

The infrastructure will be used to serve federal agencies, financial institutions, and companies in regulated sectors that need to strictly control access to sensitive data. Additionally, part of the capacity will support business expansion in Latin America.

“These contracts expand our processing capacity and allow us to distribute workloads between AWS and Oracle infrastructures,” said Garry Dias, CEO and co-founder of AI.CUBE.

What the contracts enable processing of

The contracts provide inference capacity, the stage where already trained models process information and generate responses. In an agent that performs tasks across multiple systems, this step can repeat with each decision: the model can interpret a request, examine what it found in an application, and decide what action to perform in another.

Processing capacity and tokens

AI.CUBE estimates that the infrastructure will be able to process an average of 7.6 billion tokens per hour. Tokens are units of information, such as words or parts of words, processed by models. The volume varies depending on the chosen model, the type of request, and performance requirements. The number of tasks agents can complete also depends on how much processing each task requires.

AI Factory and permission management

The startup calls its operation that combines contracted chips with its own software and services AI Factory. The goal is to reserve resources for each operation and configure permissions according to customer requirements, a critical aspect when agents perform corporate tasks involving consulting information or making changes to a company’s systems.

Qilbee platform and automation

AI.CUBE developed Qilbee, a platform that coordinates agents capable of executing process steps in different programs and maintaining the context of an action for the next one. Each project defines what agents are allowed to do and when human supervision should intervene.

Customers, expansion, and additional negotiations

The company has already developed projects with PicPay, Banco BMG, BRF, SESI, and Anvisa. In 2025, it was selected among 40 startups for AWS’s global generative AI accelerator, alongside two other Brazilian companies. The goal now is to double the customer base by the second half of 2027 and expand operations in Latin America and the United States. The company did not disclose its current number of customers.

In addition to the $50 million already contracted, AI.CUBE is negotiating an additional $63 million agreement with Oracle to contract more capacity starting in 2027. “We are in advanced negotiations to contract this capacity, with a focus on 2027. We expect to conclude the agreement by the end of the year,” said Dias.