The Vale plant that cut ore loss by 26% with AI is about to multiply across Brazil

Vale and ABB expanded the partnership that put artificial intelligence to track 400 variables in real time at a plant in Itabira. First-year results justified taking the technology to Brucutu and other units in Minas Gerais and Pará.

On Tuesday, Vale and ABB formalized the expansion of a partnership that had been running since 2024 inside the Conceição II Model Plant, in Itabira. The unit brings together thousands of sensors and automation systems capable of tracking more than 400 processing variables in real time, adjusting operating parameters without constant manual intervention. Officially inaugurated in June, it stopped being an isolated pilot project and became a replicable model: the agreement calls for gradually expanding the same technology to the Brucutu plant, in São Gonçalo do Rio Abaixo, with startup expected in early 2027, and later to other ore processing units in Minas Gerais and Pará.

What separates this case from so many AI pilot projects that never leave the drawing board is the decision criterion: Vale only authorized the replication after having concrete numbers from two years of operation. 25% higher productivity, 40% more premium ore destined for direct reduction and 26% less iron lost in tailings aren't projections from a technology vendor — they're results measured at a real plant, operating at industrial scale. That changes the logic of automation investment: instead of buying the promise of a new technology, the mining company bought the evidence of a technology already proven under Brazilian conditions, with the ore, the climate and the workforce it actually has. The choice of Brucutu as the next stop also teaches something about prioritization — it isn't the largest plant in the Vale system, but a unit strategic enough to justify being the second to receive the technology before a broader rollout. And there's a layer often forgotten in this kind of announcement: Vale says it trained operators, technicians and leaders with simulators and virtual reality before putting the technology into production — the bottleneck of industrial AI adoption is almost never the algorithm, it's the teams' ability to trust it and operate alongside it day to day.

Ore lost in tailings is revenue that never becomes product — and it's also an environmental liability that has to be managed for decades after a mine closes. A 26% reduction in that loss, replicated at scale, has a direct effect both on the company's margin and on the area occupied by dams and tailings piles. For the industry, the case is a proof of concept that automation and AI applied to processing — not just extraction or transport — generate measurable returns, which tends to accelerate similar investment decisions at other mining companies that still treat this kind of technology as a long-term experiment.