Jobs was right. Gates was wrong

3 hours ago

Running an AI model locally on top-tier hardware (MacBook Pro M5 Max) for 3.5 hours produced a mediocre result (6/10 quality) at an estimated cost of $1, while state-of-the-art models like Claude or GPT deliver better quality (9/10) in 20 minutes at a lower cost (~$0.28) under subscription models.
Current subscription prices for frontier AI services appear artificially low due to competition, but a structural paradox exists: leading providers must raise prices to cover costs, driving users to cheaper alternatives, which then invest to reach the frontier, perpetuating a cycle that hinders sustainable profit.
The future of AI is predicted to shift from software-based models to hardware-integrated systems, where AI logic is embedded directly into chip architecture (e.g., by NVIDIA), offering advantages in speed, cost, latency, and energy efficiency.
Companies like OpenAI and Anthropic may lose up to 90% of their market value, repositioning as data center operators or providers of orchestration layers, while Microsoft could follow a similar path, with traditional software interfaces losing relevance in favor of intentional, execution-driven systems.
The real value in computing will lie in physical barriers like silicon, semiconductor fabrication, and supply chains, which are hard to replicate, marking the end of software as the protagonist and a return to hardware-centric innovation.

Hasty Briefsbeta