OBJETIVO PLANETA | Los CENTROS DE DATOS y su CONSUMO DE ENERGÍA Y AGUA | RTVE Noticias

Summary of the RTVE “Objetivo Planeta” segment: Data centers, energy/water use, and the (limits of) sustainable AI

The program discusses how Spain’s rapid expansion of data centers—often driven by public-sector contracts and big-tech investment—has created growing environmental concerns, particularly around electricity demand, water consumption for cooling, and the lack of transparency from companies operating these facilities.

Main arguments and analyses

• AI and “the cloud” aren’t just technical systems; they depend on a full global supply chain. Lucía Ortiz Áte argues that even if data centers improved efficiency, “sustainable AI” is far more complex because it involves mining raw materials, manufacturing hardware (often in polluting conditions), transporting equipment, operating data centers with electricity, and eventually disposing of devices.

• Energy efficiency improvements alone won’t solve the problem if consumption keeps growing. The guests emphasize a “rebound” dynamic: technology may become more efficient, but overall usage expands faster than efficiency gains—driven by lifestyle and production models that increase dependence on digital services.

• Transparency and regulation are insufficient. Ana Valdivia highlights that data centers’ and algorithms’ real environmental footprints are difficult to assess due to opacity and incomplete regulation. She also notes that current legislation does not adequately address sustainability impacts of AI/data centers.

• Spain is becoming attractive for data-center investment for cost and grid reasons—but that can shift regional burdens. Cefe López and Valdivia explain that investors seek cheaper land and access to renewable electricity. Spain’s relatively decarbonized grid lowers operational costs, but the expansion of hyperscale data centers can also turn regions into net energy importers, creating conflicts over energy and water resources.

Concrete figures and examples mentioned

• The program cites that in Spain there are “tens, not close to hundreds” of data centers, and that in Aragon at least 23 projects have been planned—expected to generate major territorial conflict.
• It is claimed that by 2030 Aragon’s electricity consumption could multiply by 5 to 7 due to planned data centers.
• An example given: a Meta data center in Talavera de la Reina (after reduced projections) is estimated to use about 504,000 liters of water annually, potentially consuming a very large share of available non-human water resources in that context.

How the panel explains energy and water use

• Energy: Data centers run computation on silicon chips; most energy becomes heat, requiring large-scale dissipation. The panel uses analogies and examples such as:

  • Training large language models can cost around 1–5 GWh (as stated in the discussion).
  • Inference/queries also add up massively at scale (e.g., “ChatGPT queries” compared to a city’s energy use over a day).

• Water: Water is used to cool systems through largely open-cycle cooling towers. Even if server rooms use closed circuits internally, the overall cooling towers require fresh water inputs because evaporation removes heat and a portion of water is lost, requiring replenishment.

Proposed solutions and recommendations

• Better legislation and enforcement (EU-level gaps were specifically noted).
• Critical user and institutional engagement: Ana and others call for approaching AI with skepticism rather than assuming it is inherently “good” or “saving” society.
• Efficiency plus consumption restraint: Lucía stresses responsibility as consumers—reducing unnecessary AI interactions and questioning the purposes of AI use.
• Technological directions: Cefe mentions longer-term alternatives such as photonics/photonic computing (using light/photons to reduce frictional losses), framed as future-looking rather than immediate.
• Algorithmic improvements: The panel also references reducing computation via techniques like pruning to reduce operations for similar outputs.

Overall takeaway

The episode concludes that AI and data centers are not automatically sustainable. While improvements and better governance are possible, the dominant problem is that environmental impacts scale with demand, and the incentives of private operators plus weak transparency/regulation make sustainability harder to achieve.

Presenters or contributors

• Lorenzo (host/presenter)
• Lucía Ortiz Áte
• Cefe López
• Ana Valdivia