06-reference

Stratechery — Google Earnings, Meta Earnings (Update 5-4-2026)

·source: Stratechery (Ben Thompson)
stratecherybig-tech-earningsgoogle-cloudmeta-aianthropiccapexagent-infra

Why this is in the vault

Quarterly Big Tech earnings shape the cost and availability of the inference and training capacity that RDCO’s agent stack rides on. Thompson’s read on the Google vs Meta divergence (Google +10% on Cloud, Meta -9% on capex) clarifies where the hyperscaler money is actually going and how much of Google Cloud’s growth is just Anthropic recycling capital. That feeds directly into RDCO’s mental model for who to build on, what to expect from inference pricing, and where the consumer-AI lane is opening up.

Core argument

Wall Street rewarded Google and punished Meta even though Meta’s core business arguably outperformed. The split reads as: Google is monetizing AI infrastructure now (Cloud at $20B in Q1, +63% YoY, $460B backlog), Meta is asking investors to trust capex into a future personal-agent business with no current monetization story. Thompson surfaces two underappreciated points:

  1. Anthropic is plausibly the dominant input on both sides of Google’s beat. Google’s “Other income” jumped 237% to $37.7B largely on a $73.6B remeasurement of non-marketable equity, almost certainly driven by Anthropic’s Feb 2026 raise at a $380B valuation (Google owned ~14% as of March 2025). Separately, Anthropic’s commitment to up to 1M TPUs (“tens of billions of dollars” worth) is likely a major contributor to Cloud revenue acceleration too. The “Enterprise AI revenue grew nearly 800% YoY” quote is almost certainly Anthropic-shaped.
  2. Meta’s lack of a cloud business is a feature, not a bug. Because Meta has no enterprise customers competing for inference, all compute can flow to consumer products. Zuckerberg framed AI investment as terminal value: without recursive self-improvement and frontier models in-house, Meta loses its grip on consumer attention, which kills the whole business. The shopping-agent angle is the wedge.

Thompson stakes a position: there’s a real opening for AI built for individuals (vs. “the machine god”), and Meta’s advertising engine plus consumer focus may make it the right shape to capture it - if it can ship a product that resonates.

Mapping against Ray Data Co

Medium relevance. Two operational implications:

What this does NOT change: physical-AI thesis, MAC roadmap, or any near-term shipping decision. It is calibration, not redirection.

The contrarian read worth noting: if Anthropic is actually 50%+ of Google Cloud’s growth story, that’s concentration risk for both sides - and a reason to keep watching Anthropic’s own infra diversification (AWS Trainium, in-house). 2026-04-30-stratechery-amazon-earnings-trainium-commodity is the other half of this picture.