Upstream, a ShovelSense XRF pilot at Centinela in 2025 characterized copper and gold grades at the point of loading, enabling material routing decisions before ore reaches the primary crusher

Decision Lens

Most operations deploy point solutions that optimize in functional silos. According to industry reporting from March 2026, Antofagasta is running a different architecture: SIRO process control, autonomous haulage dispatch, and real-time in-bucket ore characterization feeding one another across three Chilean operations. Reported 2025 outputs include approximately 4% more unconstrained SAG operating hours at Los Pelambres and simultaneous throughput and recovery gains at Centinela. The operational question this raises is not whether individual AI modules produce ROI, but whether the integration layer between them compounds returns in a way that isolated deployments cannot replicate.

90-Second Brief

As the week closes, antofagasta has built an integrated AI and autonomous operations platform across Los Pelambres, Centinela, and Antucoya in Chile, connecting SAG mill optimization, milling, flotation, leaching, and fleet management into a shared data layer. Centinela began running six autonomous trucks at Encuentro Sulfuros pit in February 2026, the first stage of a planned 53-unit district fleet. The company is simultaneously investing $4.4 billion in a second concentrator at Centinela, with a 120 MVA renewable energy substation already energized to support precommissioning and eventual full operations.

What’s Actually Happening

The SIRO platform is modular but connected: SIRO SAG targets grinding constraints, SIRO MINCO addresses milling and flotation, and SIRO Ácido manages leaching with real-time ore characterization inputs. Recommendations are prescriptive, drawn from aggregated cross-site data rather than site-specific rule sets. At Los Pelambres, machine learning reportedly identified dynamic SAG mill constraints that conventional monitoring had not detected, producing roughly 4% additional unconstrained operating hours. At Centinela, SIRO MINCO combined historical process data with real-time mineralogy to tune operational parameters concurrently, with reported results of a 1% throughput increase and a 0.75 percentage point recovery improvement in 2025.

The autonomous haulage program at Centinela has been developing since 2021. The current phase—six trucks active at Encuentro Sulfuros as of February 2026—is the first production deployment, with plans to reach 26 units at that pit through 2026 and 53 units across Esperanza, Esperanza Sur, and Encuentro Sulfuros post-2026. Upstream, a ShovelSense XRF pilot at Centinela in 2025 characterized copper and gold grades at the point of loading, enabling material routing decisions before ore reaches the primary crusher. OREPro 3D blast optimization at Los Pelambres reportedly delivered roughly 5% improvement in fragmentation metrics in 2025, with downstream effects on grinding energy consumption and plant throughput.

Why It Matters for Mining Operations Directors?

The compounding structure is the operationally significant point. A 4% gain in unconstrained SAG hours produces additional contained metal without new mill capacity. A 0.75 percentage point recovery improvement moves directly to payable copper. When those two outcomes coexist with a concurrent throughput increase, the implication is that the AI layer is resolving competing constraints simultaneously—not trading one metric for another, which is the common failure mode of conventional advanced process control.

For Directors managing established processing circuits, the ShovelSense pilot signals a directional shift in where grade management occurs: upstream from the plant to the shovel. If in-bucket XRF characterization scales from pilot to production, it changes the information available to fleet dispatch and mill feed blending before ore enters the circuit—shifting the throughput and recovery problem from reactive adjustment inside the plant to proactive routing at extraction. The blast optimization layer adds a further upstream intervention point: better fragmentation at the bench reduces specific energy in SAG and ball milling and supports more stable feed presentation, which is precisely where SIRO SAG’s recommendations operate. The architecture treats each stage as an input to the next rather than an isolated optimization target.

The Forward View

Centinela is the clearest leading indicator of what this integrated model looks like at district scale. The ramp from 6 to 53 autonomous trucks, combined with the $4.4 billion Nueva Centinela second concentrator, sets up an operation where autonomous haulage, real-time ore characterization, and AI process control are designed in from the start rather than retrofitted to existing infrastructure. The 120 MVA renewable energy substation—already energized and supplied by Siemens Energy—removes one energy supply constraint from the commissioning sequence and positions the expanded operation for lower carbon intensity from day one.

Operations watching this build should track the integration point between autonomous dispatch and ShovelSense grade data. That is where the closed-loop concept becomes operationally real: material routing decisions made at extraction, confirmed by AI-driven process targets downstream. Whether this architecture performs at full 53-truck district scale remains to be demonstrated. The sequencing of technology investments, however, indicates a deliberate build toward it rather than opportunistic adoption.

What We’re Uncertain About?

  • ShovelSense XRF maturity. The Centinela deployment in 2025 was a pilot. Whether grade-routing accuracy met the threshold for production-scale rollout, and what the integration pathway with autonomous dispatch looks like, has not been confirmed in available reporting. What would resolve this: a formal production deployment announcement from Antofagasta with accuracy and routing outcome data.

  • Autonomous fleet ramp execution. Six trucks had been operating for roughly six weeks at the time of reporting. Scaling to 26 units across a single pit and then to 53 across three pits introduces dispatch, geotechnical interface, and maintenance coordination variables that cannot be assessed from early-phase data. What would resolve this: mid-year 2026 operational reporting from Centinela on fleet availability and tonnes-moved metrics.

  • SIRO gain sustainability. The reported 4% SAG improvement and 0.75 percentage point recovery gain reflect 2025 performance periods. Whether these represent sustained annual averages or shorter optimization windows is not confirmed. What would resolve this: multi-year performance data or audited annual production reporting with technology attribution.

  • Nueva Centinela commissioning timeline. The second concentrator is under active construction. Project execution risk—including schedule, integration of advanced process control into new facilities, and technology performance at commissioning—remains unresolved as of March 2026.

One Question to Bring to Your Team

If we mapped our current process data architecture against this model, where would a shared feedback loop between grade characterization at the shovel, fleet dispatch, and process control expose the largest constraint we are currently managing reactively—and what would it take to close that loop?

Sources

  • Com — Antofagasta’s AI-Driven Copper Growth Strategy Through Advanced Innovation (Link)

Tags