Impossibl. is building its own AI infrastructure to cut inference costs. The startup, founded by Fábio Roma, is targeting the stage where language models answer user commands. Its plan moves from aggregating AI models to operating data centers in Brazil.
Three expansion phases
The company will first centralize access to various AI models on one platform. It then plans to lease computing capacity for that user base. The final phase includes building proprietary data centers.
Brazil as a priority market
Brazil receives priority attention in the startup's strategy. The country ranks as the third largest global consumer of AI tokens, according to the company. It also shows fast adoption of new digital tools.
Lowering inference costs
The startup concentrates on the inference stage of AI systems. This is the phase where models execute commands and produce responses. Its goal is to reduce token prices and computing expenses.
The company emerged after Roma moved to the United States in 2022. He won a hackathon with about 250 competitors and a US$15,000 prize. The work implemented a recent study on recursive language models.
Read more: startupi.com.br