The End of Compute Scarcity? Not So Fast

9 hours ago

AI compute scarcity may not be ending; SpaceX (xAI) and Meta renting out capacity likely reflects idiosyncratic issues rather than systemic excess, with high prices and short-term deals indicating continued scarcity.
The rise of open-source models and end of token maxing may bifurcate AI usage: low-cost models could handle most tokens, while frontier models from labs like Anthropic and OpenAI capture premium revenue from mission-critical work.
AI spending is highly concentrated; data shows top firms spend significantly more per employee, suggesting frontier tokens will dominate revenue despite a smaller share of total tokens.
Overall median EV/NTM revenue multiple for SaaS companies is 3.5x, with high-growth firms at 19.7x, mid-growth at 5.3x, and low-growth at 2.8x, reflecting valuation based on growth expectations.
The author anticipates AI's growth trajectory will outweigh short-term optimizations like token maxing, but notes a risk from the 'own your weights' trend for large AI labs.

Hasty Briefsbeta