The Collapse of Big Tech's Climate Promises
The latest environmental reports from digital giants reveal a troubling reality: the frantic race toward generative artificial intelligence is derailing ecological transition goals. According to data published by Le Monde, Google's greenhouse gas emissions have jumped 82% since 2019. For its part, Amazon shows a 58% increase over the same period, according to information reported by La Presse.
Each year, these trajectories push these multinationals further away from their carbon neutrality commitments, originally set for 2030 or 2040. For the first time, the growth of their carbon footprint is outpacing their revenue growth. This phenomenon is driven by the massive construction of highly energy-intensive data centres, which are essential for training and running large language models.
Why Does Artificial Intelligence Consume So Much Energy?
To understand this drift, it is helpful to distinguish between two major phases in the lifecycle of an artificial intelligence model: training and inference. Training involves feeding a model massive volumes of data so it can learn to structure its responses. This stage requires thousands of graphics processing units (GPUs) running at full capacity for weeks, or even months.
However, inference, which is the daily use of the model to answer user queries, represents the largest share of energy consumption over the long term. Every question asked of a centralized chatbot triggers a network round trip to remote servers, mobilizing colossal computing infrastructure and cooling systems. According to a study by the International Energy Agency (IEA), global electricity demand from data centres could double in the coming years, driven primarily by the rise of AI.
Given this reality, the absolute centralization of computing power among a handful of providers poses a systemic sustainability problem. An alternative approach, based on digital sustainability and decentralization, is becoming necessary.
The Answer Through Hardware Longevity: Boréal-OS
Reducing the digital carbon footprint cannot be limited to software optimization. The most effective lever lies in extending the lifespan of existing computer hardware. Indeed, manufacturing a new computer accounts for the majority of its overall environmental impact, due to rare earth extraction and the required industrial processes.
This is precisely where Quebec's sovereign ecosystem comes into play with Boréal-OS. This native operating system, designed as a lightweight and secure Linux distribution, installs directly on a machine's hard drive. It can add five to ten years of useful life to computers declared obsolete by the hardware requirements of recent commercial systems, such as the lack of a TPM 2.0 chip or latest-generation processors.
By preventing the premature disposal of entire computer fleets, particularly in schools and community organizations, Boréal-OS directly tackles the problem of electronic waste. Once the machine is refurbished, access to modern tools is achieved simply through the web browser.
Decentralization and Local Execution: The ProductivIA Platform
At the application level, the ProductivIA platform offers an architecture designed to limit dependence on the overloaded cloud infrastructure of hyperscalers. Two concrete approaches make this possible:
First, the IA Locale application leverages the WebGPU standard. This technology allows the browser to directly access the computing power of the user's machine to run intermediate-sized AI models. Inference calculations are thus performed locally, without any data transiting over the network and without soliciting remote servers. For common writing or document analysis tasks, this method eliminates the carbon footprint associated with data transmission and external processing.
Second, when greater computing power is required, the platform's orchestrator prioritizes sovereign, optimized models like Matania. Physically hosted in Quebec, this model provider relies on an electrical grid powered predominantly by hydroelectricity, a very low-carbon energy source compared to the coal- or gas-powered grids often used by large American data centres. Furthermore, using reasonably sized models, fine-tuned for specific tasks, avoids the unnecessary overconsumption associated with systematically querying underutilized giant models.
Toward a Responsible Digital Transition
The energy crisis driven by the rise of artificial intelligence shows that technological innovation can no longer bypass a reflection on its physical limits. Transitioning to more efficient models, adopting localized computing architectures, and refurbishing existing hardware are concrete ways to reconcile technical progress with planetary boundaries. Quebec's ecosystem demonstrates that a local alternative, structured around sustainability and sovereignty, is technically viable.