Claude and ChatGPT are getting worse. It's not your imagination (www.artificialstudio.ai)

Recent observations indicate that popular AI models like Claude and ChatGPT are experiencing slower response times and reduced performance, a situation more structural than coincidental. The underlying infrastructure for these AI systems is nearing capacity, leading companies to make trade-offs that compromise quality. As the demand for complex tasks has surged, the compute requirements have skyrocketed—an hour-long interaction with agentic AI can consume up to 500,000 tokens compared to 500-1,000 tokens for simple queries. As a result, companies like Anthropic are instituting token limits during peak hours to manage resource constraints, while OpenAI faces significant financial losses driven by a model that subsidizes free users. This infrastructure strain is causing financial and operational challenges for AI firms. With a projected loss of $14 billion for OpenAI in 2026 and a similar loss of $3 billion for Anthropic, reliance on free user bases remains unsustainable. As firms push to triple electrical capacity amidst a backdrop of material shortages and permit delays, the AI landscape is entering a critical transitional phase. The quality of AI-generated outputs is increasingly inconsistent, compelling users—especially creatives—to adapt their workflows and leverage platforms that allow for testing across different models to ensure reliable performance. Understanding these dynamics is essential for AI users to navigate the current landscape effectively.