Perspective: AI demand is inflated, and only Anthropic is being realistic

Read the full articlePerspective: AI demand is inflated, and only Anthropic is being realistic on CNBC Tech

What Happened

The main usage metric for artificial intelligence, called tokens, looks explosive on paper, but it may be significantly overstated.

Our Take

Token volume reported by most AI firms includes filler text, repeated prompts, and system noise—Anthropic is the only vendor breaking out net effective tokens used in production.

High token counts don’t translate to real usage; many RAG pipelines waste 40% of LLM calls on redundant context regeneration. Teams optimizing for cost using GPT-4 usage dashboards are flying blind—stop measuring raw output volume and start tracking task completion rates with Haiku.

Teams running production agents should audit their actual task throughput with Claude Opus logs; everyone else can ignore this until their API bill exceeds $50K/month. Do audit token utility instead of chasing throughput benchmarks because most 'usage' is noise.

What To Do

Do audit token utility instead of chasing throughput benchmarks because most 'usage' is noise

Builder's Brief

Who

teams running production agents

What changes

cost tracking and usage validation workflow

When

weeks

Watch for

enterprise AI spend exceeding $50K/month on a single model

What Skeptics Say

Companies have incentive to inflate usage metrics to attract investment; reported growth may reflect padding, not product-market fit.

Cited By

CNBC Tech Perspective: AI demand is inflated, and only Anthropic is being realistic