NeoSpace: the startup from Uberlândia that bets on data instead of language — and was bought by Itaú

⏱️ Reading time: ~5 minutes

Introduction

When people talk about the global artificial intelligence race, the names that come up usually come from Silicon Valley or Asia. On October 26, 2026, a Brazilian name entered that conversation through an unusual route: Itaú Unibanco, Brazil's largest private bank, signed an agreement to buy 100% of NeoSpace, an AI startup founded in late 2023 in Uberlândia, in the state of Minas Gerais, for R$ 2.925 billion (about US$ 589 million).

The price tag stands out, but what makes the case interesting is the company's technical thesis. NeoSpace does not build a chatbot or adapt foreign models. It trains its own foundation models, aimed not at text but at corporate data — what the company calls Large Data Models (LDMs).

Who NeoSpace is

The founders include Bruno Pierobon, who remains CEO, Felipe Almeida and Gustavo Debs. The trio already had a history with the same buyer: they created Zup, a software company whose control Itaú acquired in 2020.

NeoSpace's trajectory was fast:

• January 2025: an US$ 18 million round led by Itaú, which put in US$ 15 million and took a 15% stake. Investors linked to General Atlantic and Ribbit Capital also took part. The startup was valued at around US$ 100 million.
• March 2026: the company appeared on the main stage at GTC, NVIDIA's annual conference.
• April 2026: it projected more than R$ 300 million in revenue for the year, expecting revenue to double annually.
• September 2026: it was negotiating a new round of up to US$ 250 million at a US$ 1.25 billion valuation.
• October 2026: instead of the round, came the full sale to Itaú.

The company says it is the largest user of NVIDIA B200 GPUs in Latin America — about 1,200 units, hosted in Australia — and plans to quintuple its computing capacity within two years.

What a Large Data Model is

Large language models (LLMs), like the ones behind chat assistants, were trained to understand and produce text. NeoSpace's proposal is different: a model trained to interpret large volumes of operational data — financial transactions, consumption history, operational records, industrial information — and documents in any format.

According to the company, the LDM maps relationships between data points, identifies complex patterns and generates recommendations or actions in near real time. One method it has described: the model uses 24 months of user data, with the first 18 months used to learn behavior patterns and the last 6 to validate the predictions.

In practice, NeoSpace offers a platform for each company to train its own model on its own data.

Where the technology is applied

Cases cited by the company and the press include:

• Banks: credit, customer relationships and fraud prevention. The models were already being trained on Itaú's data on these three fronts.
• Telecommunications: predicting customer churn.
• Aviation: optimizing routes and fuel consumption.
• Energy, oil and gas: supporting exploration decisions.
• Insurance, retail and mining: sectors described as potential markets.

On results, the company cites performance gains of 8% to 10% in some cases and up to 30% in others, compared with traditional machine learning models. These figures were released by the company itself and have not been independently audited.

The Itaú acquisition

The deal was disclosed to the market through a material fact filing. Part of the amount is paid at closing and the rest over up to five years, subject to conditions — the bank did not disclose the size of the first installment. The transaction still requires approval from CADE, Brazil's antitrust authority, and the Central Bank of Brazil, with no set deadline.

According to the announcement, NeoSpace will keep its management and team, but is expected to serve only Itaú once approvals are granted. The technology is set to power the bank's AI assistant, ia.i. Itaú's CEO, Milton Maluhy Filho, said the purchase aims to place the bank "among the institutions that will define the next chapter of the global financial industry."

What is still unknown

• What happens to current clients in telecommunications, aviation and energy once the company serves the bank exclusively.
• The fate of the office NeoSpace was setting up in the United States and of the negotiations under way there and in Europe.
• The size of the initial installment and the conditions for the remaining payments.
• The financial gain Itaú expects from the deal, which was not disclosed.

Conclusion

The NeoSpace case highlights a front of artificial intelligence that gets less attention than chat assistants: foundation models trained on companies' operational data rather than on language. It is a bet that, in the corporate world, the value lies in predicting behavior and risk from a company's own data.

It is also a case of foundational technology developed in Brazil, outside the Rio–São Paulo axis, that in under three years went from an US$ 18 million round to a sale of nearly R$ 3 billion. With exclusivity expected for Itaú, this technology will no longer be offered to the market and becomes an asset of a single bank.