Chile Embodies A.I.'S No-Win Politics (www.nytimes.com)

Chile has become a flashpoint in the global debate over A.I., as the government tries to attract cloud investment while communities and environmentalists push back against the resource costs and foreign control of compute. Officials are proposing rules to steer data centers away from water-stressed Santiago toward the solar-rich north (Antofagasta) and to secure local access to foreign compute—an “astronomy model” where a share of capacity benefits domestic researchers. The country already hosts about 33 data centers (expected to double by 2030) and faced plans for roughly 30 more; community fights have halted at least one Google expansion and spotlighted large cooling-water draws (estimates cited include ~50 L/sec at one Google site and a proposed Cerrillos facility that would have used ~228 L/sec). For the AI/ML community the stakes are practical and technical: where data centers are built affects latency-sensitive applications (firms warn 680 miles from Santiago could hurt voice A.I.), availability of GPU-scale training resources for local model-building, and environmental sustainability of compute-intensive workloads. Chilean researchers are already training regional models to avoid cultural blind spots (e.g., early ChatGPT misattribution of Chilean literature), but tensions over transparency, resource allocation and dependence on U.S. cloud providers highlight a broader trade-off—accelerating A.I. capacity without repeating extractive infrastructure models that harm ecosystems and limit local control.