Analysis ·
The desperate search for the "green" algorithm: AI looks to software to cut its enormous costs
Original title, in Spanish: “La búsqueda desesperada del algoritmo "verde": la IA está buscando en el software abaratar sus descomunales costes”
- Publication
- Xataka (Spain)
- Author
- Pablo G. Bejerano
- Published on
- 3 October 2026 · read in the page structured data
Summary
The article explains how optimising software, and not only hardware, can reduce the energy consumption of enterprise AI. According to a McKinsey report, 89% of companies worldwide use AI and 44% are in the scaling phase, but around 20% restrict its use because of high operating costs. Spanish researchers Coral Calero and Verónica Bolón Canedo describe techniques such as reducing calculation precision or mixture of experts (MoE), popularised by DeepSeek, to cut training and inference costs. The article also covers the rise of cheaper Chinese models, widely adopted including by American companies, and cites the International Energy Agency's projections for data centre electricity consumption.
Summary based on the full text of the article.
Read the article on Xataka →Figures cited
-
89 %
share of companies worldwide using AI
companies worldwide, 2026
« el 89% de las empresas a nivel global usan la IA »
Source: McKinsey, The State of AI in 2026: On the road to ROI
-
44 %
share of companies in the AI scaling phase
companies worldwide, 2026
« un 44% se encuentra en fase de escalar la tecnología »
Source: McKinsey, The State of AI in 2026: On the road to ROI
-
20 %
share of companies restricting AI use because of costs
companies worldwide, 2026
« alrededor del 20% de las empresas encuestadas restringe el uso de la IA por sus elevados costes operativos »
Source: McKinsey, The State of AI in 2026: On the road to ROI
-
415 TWh
global electricity consumption of data centres in 2024
global, 2024
« En 2024 el consumo global de electricidad de los centros de datos se situaba en 415 TWh, según la Agencia Internacional de la Energía »
Source: Agence internationale de l'énergie
-
945 TWh
projected global electricity consumption of data centres in 2030
global, 2030 projection
« las estimaciones de la entidad apuntan a los 945 TWh para 2030 »
Source: Agence internationale de l'énergie
-
45.5 %
market share of Chinese models on OpenRouter in August
OpenRouter platform, August
« las compañías chinas tuvieron un 45,5% de cuota y superaron la suma de Google, OpenAI, Anthropic y Meta »
Source: OpenRouter
Classification
- Country
- Worldwide scope
- Company size
- All sizes
- Industries
- All industries, or industry not specified · Software vendors, IT services companies, telecoms · Energy, utilities, environment
- Companies
- DeepSeek (cited), Z.ai (cited), Meta (cited), McKinsey (study author), Moonshot (cited), Alibaba (cited), OpenAI (cited), Anthropic (cited), OpenRouter (cited), Google (cited)
- Keywords
- energy costs, software efficiency, mixture of experts, AI adoption, data centres, Chinese models
Entry written by Marsanne Studio from the cited article, with the help of an AI model (claude-sonnet-5). The rights to the article and its illustrations belong to Xataka and its author. Entry translated from French by Claude Sonnet 5.5.