Four Horsemen of the AIpocalypse
6 hours ago
- #AI Bubble
- #Tech Industry Analysis
- #Startup Economics
- Anthropic's Claude services have poor uptime, sometimes below 99%, causing significant downtime and leading some enterprise clients to switch providers.
- Users report severe performance regressions in Claude's models, like Opus 4.6 and 4.7, with increased errors and token usage, while Anthropic downplays concerns.
- Anthropic's growth is fueled by subsidies and VC funding, hiding the true cost of AI services, which are unprofitable and lack operational stability.
- Anthropic's capacity issues are linked to limited global data center construction, with most new capacity reserved for OpenAI and Anthropic, raising sustainability questions.
- NVIDIA claims $1 trillion in GPU sales through 2027, but data center construction only supports $285 billion worth, suggesting oversupply and warehousing of GPUs.
- Companies are spending up to 10% of headcount costs on AI inference, with costs rising quickly, contradicting promised efficiency gains and productivity benefits.
- Microsoft is moving GitHub Copilot to token-based billing and tightening limits due to soaring costs, signaling an end to subsidies in AI services.
- AI revenues are inflated by misleading metrics like contracted ARR, with startups overstating deals, while actual benefits remain unproven and costs skyrocket.