People are finally starting to notice the weekly rate limit cuts on AI subscriptions that I've been pointing out for a while. In reality, they've been quietly trimming limits for a long time. Both Claude and GPT have done it with basically every major update. It was inevitable. Compute costs have only gone up due to DRAM bottlenecks, and subscription tiers were subsidized at nearly 100x compared to actual API pricing. The hardware bottleneck is far from solved, and physically fixing it within the next three years is extremely unlikely. Subscription prices will keep climbing, and usage limits will keep shrinking. Local AI fixes this.