Veblen & Jevon Walk Into a Data Center

Key Takeaways

• Anthropic's Mythos pricing could reach $150 per million output tokens.
• High-cost models may become Veblen goods, demand rises with price.
• Capital‑rich firms will dominate AI‑driven product development.
• Startups face higher burn rates or reduced feature parity.
• GPU scarcity amplifies advantage of firms with deep cash reserves.

Pulse Analysis

The recent leak of Anthropic’s Mythos model has sparked a reassessment of AI economics. Historically, the industry followed Jevons paradox: as compute costs fell, usage surged, driving token prices down from $5‑$25 per million to sub‑dollar levels. Mythos, however, is projected to cost $150 per million output tokens—far above current benchmarks—signaling a potential transition to Veblen‑good behavior where scarcity and prestige drive demand despite higher prices. This shift challenges the prevailing focus on cost‑efficiency and forces firms to evaluate the strategic value of premium inference capabilities.

For venture‑backed startups and mid‑size enterprises, the emerging pricing structure reshapes capital allocation. With GPUs and high‑bandwidth memory already tight, securing the hardware to run Mythos‑class models will require deep pockets or favorable financing terms. Balance sheets become a defensive moat; firms that can raise capital cheaply will invest in both the compute stack and the expensive token consumption, accelerating product development cycles. Conversely, companies locked into cheaper models like Claude Opus 4.6 may see their feature sets erode, prompting either price hikes for end‑users or a strategic pivot toward niche, cost‑sensitive markets.

The broader market impact could be pronounced. As AI‑native firms leverage Mythos to build applications ten times faster, valuation multiples may diverge sharply, rewarding capital‑rich players while penalizing those unable to match the compute spend. Investors will likely scrutinize cash burn and runway more closely, favoring businesses that treat AI capability as a core competitive asset rather than a cost line item. In the long run, the industry may see a bifurcation: a premium tier of ultra‑capable AI services and a parallel ecosystem of lightweight, cost‑optimized solutions catering to price‑sensitive segments.