Adoption, and why no one is talking about it

• LLMs do not disclose active user growth or retention. (Analysis of each company suggests why, below)
• Independent reporting by five data sources on web traffic and app downloads indicates adoption flattened months ago.
• Chip sales are reported, but end-user adoption is not.
• Media coverage focuses on new features rather than disclosed trends in persistent users.
• Salesforce executives have publicly stated that trust in LLMs has declined.
• Satya Nadella has publicly acknowledged that Copilot integrations “for the most part don’t really work,” and Microsoft has resorted to actively paying third-party trainers to encourage usage.
• An MIT study found that 95% of enterprise AI pilots fail.
• ChatGPT is the dominant consumer LLM, yet Menlo Ventures reports that roughly 95% of users rely on free access rather than sustained paid usage, indicating shallow engagement rather than durable adoption. Data suggests even free usage has plateaued.

If adoption were accelerating, it would be the first metric disclosed.
So why is it missing?

Analysis

Microsoft

Would reprice if they disclosed:

• Active Copilot users as a % of eligible seats
• Retention after 90 / 180 days
• Revenue per Copilot user vs non-Copilot user

Why they don’t:

• Weak persistence would be immediately visible
• Copilot is bundled, masking true demand

Amazon

Would reprice if they disclosed:

• % of AWS customers running AI workloads in production (not pilots)
• Incremental AWS spend attributable to AI per customer
• Churn or expansion differences between AI vs non-AI customers

Why they don’t:

• Pilots don’t equal durable revenue
• AI cost curve hits margins before benefits appear

Alphabet

Already partially repriced because:

• Expectations were low
• AI framed as defensive to search

Would further reprice if they disclosed:

• AI-assisted search usage as % of total queries
• Ad revenue lift or dilution from AI answers

Why disclosure is cautious:

• Any cannibalization risk spooks the core business

Meta

Would reprice if they disclosed:

• Revenue lift per user attributable to AI ranking
• Incremental infra cost per AI-optimized user session

Why they don’t:

• AI here is optimization, not a product
• Hard to separate signal from baseline engagement

Apple

Would reprice if they disclosed:

• On-device AI usage frequency
• Retention or upgrade lift tied to AI features

Why they don’t:

• AI is defensive, not additive
• Disclosure only happens when benefits are undeniable