CNBC Commentator: AI Demand Overestimated, Only Anthropic's Pricing Strategy is Realistic
CNBC analysis points out that the AI demand indicator, token usage, shows explosive growth on the surface, but is actually severely overestimated, including internal "tokenmaxxing" competitions and opaque measurements. Anthropic has taken the lead in pricing uncertainty by shifting to token-based billing and limiting high-consumption tools.
Anthropic CEO Dario Amodei stated that data centers require 1-2 years for construction, and investments must weigh the risks of "losing customers due to insufficient capacity versus building excess capacity at a loss," while companies like OpenAI expand usage through price cuts, masking true demand signals.
Source: Public Information
ABAB AI Insight
token as a basic metric for AI usage is distorted by incentive mechanisms: companies compete to maximize consumption to prove ROI, similar to the early cloud computing "server farm" waste. This leads to distorted demand signals, with trillion-dollar expansions of data centers based on unverified assumptions.
Anthropic's shift to usage-based billing is essentially a "risk hedge": revenue directly reflects real value, avoiding overcommitment under fixed fee models. At the same time, limiting high-consumption tools forces customers to focus on high-value scenarios, obtaining cleaner data feedback.
This divergence reflects the differentiation in AI business models: OpenAI bets on scale effects and mass adoption, while Anthropic emphasizes enterprise-level authenticity. If demand retracts, the former faces the risk of idle infrastructure, while the latter's pricing is closer to sustainable levels.
In the long run, the market will reassess true demand through pricing and adoption rates. Similar to historical tech cycles, the winners after the initial bubble are those that establish credible metrics and sustainable revenue, rather than those maximizing superficial indicators.