Itaú agrees to buy Brazilian AI model startup NeoSpace for $589 million

What Itaú is buying

Brazilian bank Itaú Unibanco has agreed to buy all of NeoSpace, a São Paulo AI startup that builds foundation models for business data, for 2.925 billion reais. Reuters, as published by CNA, put the price at about $589 million and said Itaú disclosed the deal in a securities filing on Monday, 26 October. Part of the money is due at closing and the rest is deferred over five years.

The bank is not folding NeoSpace into its own technology team. According to Valor Econômico, the startup will keep autonomous management and operations under co-founder Bruno Pierobon, who stays as chief executive, and the agreement provides for keeping its team and continuing its research and development. That is an unusual choice for a bank buying an AI company, and it suggests Itaú wants the lab to keep behaving like a lab.

The deal is a big step up in value. In January 2025 NeoSpace raised $18 million at a $100 million valuation in a round led by Itaú, NeoFeed reported. At the reported dollar price, the bank is now paying nearly six times that valuation, less than two years later, for a company founded at the end of 2023.

What NeoSpace builds

NeoSpace calls its technology Large Data Models, or LDMs. NeoFeed describes them as foundation models that, unlike large language models, are built to work with any kind of data and document rather than only language. The aim is to take in large volumes of information from different sources so companies can spot and anticipate patterns in customer behaviour.

The company does not try to compete with OpenAI and Anthropic for the same public text. Instead, NeoFeed says, it builds platforms that let companies build and train models on their own data. For a bank, that data is transaction histories, credit records, account activity and customer service logs, the kind of structured information that general chatbots are not trained on and that banks are wary of sending to an outside provider.

NeoSpace has customers outside finance, in telecoms, aviation and energy, according to NeoFeed. Pierobon recently told the publication the startup was in advanced talks over two opportunities in the United States, where it is setting up an office, and one in Europe. The filing as reported does not say whether those outside customers and talks will continue under Itaú's ownership, although the promise of autonomy suggests they could.

A two-year partnership and an older connection

According to Itaú's statement, the two companies have worked together for almost two years. During that time they developed generative AI tools for credit, financial management, customer relationships and fraud prevention, Valor reported. The bank told NeoFeed the startup's technology was already being used in projects the two built jointly.

The relationship goes back further than NeoSpace itself. Pierobon founded the startup with Felipe Almeida, Gustavo Debs and Thiago Teixeira. In 2019, Itaú paid more than 500 million reais for Zup, a software and technology services company that Pierobon, Almeida and Debs had founded in 2011. With this deal, NeoFeed notes, the bank is bringing that trio back into its group for a second time.

Other backers in NeoSpace's 2025 round included Martin Escobari of General Atlantic and Micky Malka of Ribbit Capital, according to NeoFeed. The reported terms do not say how much of the purchase price goes to the founders versus those investors, or what conditions apply to the deferred payments.

Where the technology goes at Itaú

Itaú says NeoSpace's technology should feed into ia.i, the bank's generative AI platform. NeoFeed reports that ia.i launched at the end of July and lets customers use the bank's products and services through natural language, by text or voice. Itaú aims to make it available to all individual customers of its superapp by the end of 2026.

Valor says the bank sees room for more personalised services, financial guidance and credit offers tailored to each customer's profile. Those are the uses where models trained on a bank's own records could matter most, since a credit offer depends on data no public model has seen.

Fraud prevention is the other obvious use. Latin American banks have become targets for AI-assisted attacks, as AI Lookout reported in our story on a cybercartel using AI agents against banks in the region. A model that learns normal account behaviour from the bank's own data is one way to catch unusual activity faster.

Why a bank wants its own foundation models

In the statement, Itaú chief executive Milton Maluhy Filho said Brazil had already set international benchmarks in banking innovation and now had the chance to raise them with frontier AI. He called the acquisition a strategic investment and said that by combining foundation technology, financial knowledge and scale, the bank could build products that set new market standards in Brazil and around the world.

Pierobon said the partnership with Itaú had let NeoSpace apply foundation models to real problems in the financial system and show they could work at scale, and that the new stage would let it deepen its research with the same team and vision.

Big banks elsewhere are reorganising around AI, but mostly by hiring executives and signing deals with model providers. Barclays, for example, named its first chief AI officer while cutting thousands of jobs, as AI Lookout reported in our Barclays story. Buying the model maker outright is a different bet: it gives Itaú control over the models, the researchers and the roadmap, rather than renting general-purpose systems.

NeoSpace is one of many Brazilian groups building their own models. NeoFeed cites a study from the Federal University of Goiás that found more than 50 language models developed in Brazil, including from WideLabs, NeuralMind and Clarice.ai, but says NeoSpace stood out as one of the few to raise significant money to build them.

What still has to happen

The deal still needs approval from Cade, Brazil's competition authority, and the Banco Central do Brasil, and has to meet other conditions in the contract, Valor reported. No closing date was given in the coverage AI Lookout read.

Owning a model company also brings model risk inside the bank. Regulators have been setting out what they expect, including the Financial Stability Board's final guidance for banks covered in our report on the FSB's 12 AI sound practices. Any credit or fraud decisions shaped by NeoSpace's models will have to meet those expectations.

For the wider market, the price is the headline. A bank paying close to $600 million for a two-year-old model startup it already knew well is a strong signal that, at least in finance, models trained on a company's own data are seen as worth owning rather than buying as a service.