Enterprise AI Analysis

AI preparation involves the creation and refinement of models, primarily through extensive training on large datasets. This process includes dataset production—often from vast internet scrapes and sometimes copyrighted material—and model design and development. The cost of training foundation models like ChatGPT is immense, leading to a fierce competition for top AI talent. Human data workers, often outsourced to the Global South for low wages, play a crucial role in annotating data and providing feedback to 'fine-tune' models.

AI deployment represents the downstream phase where developed models are made accessible for real-world applications. This can be through subscription models, APIs, or AI-as-a-service platforms offered by large tech companies. Developers can build features on top of these foundation models, creating a wide array of AI-powered products and services. Deployment is reliant on the preceding infrastructure and preparation phases and often involves continuous development and refinement.

E-waste is a critical, often invisibilized, component of the AI supply chain. When hardware and infrastructure become obsolete or non-functional, they contribute to a growing volume of electronic waste. The computing-intensive nature of generative AI accelerates this cycle, increasing the demand for new hardware and subsequently, the generation of e-waste. This final stage highlights the significant environmental costs associated with the entire lifecycle of AI systems.