Briefing: Google's Moat Is Wide and Grand (2023)
Written: May 6, 2023 | Language: Chinese | Author: E. J. Original article: https://www.ateasthillside.com/p/40 (Substack — paywall) Briefing by: ejsays.com Read with: https://posts.ejsays.com/googles-grand-moat-is-drying-up-nobody-is-attacking-it/
Core claim (2023): The leaked Google memo arguing "Google has no moat" was wrong. Google's moat is wide, multi-layered, and not primarily technical. It is commercial and infrastructural. Open source is not the threat the memo imagined.
Context: In May 2023, an internal Google memo leaked arguing that open-source AI would erode Google's competitive position. The author disagreed immediately and publicly. This briefing documents that original argument. For the 2026 update — written after the author watched 18 of her own pages disappear from Google's index in silence — see the companion piece above.
Why the memo was wrong on open source: The author distinguishes two types of open-source innovation. The first is disruptive — scale innovations like LoRA that allow models to run on consumer hardware, comparable to the microcomputer replacing the mainframe. The second is incremental fine-tuning on existing models, which becomes a red ocean quickly. The memo conflated the two. More importantly: without a foundation model to fine-tune, open source has nothing to attach to. LLaMA "accidentally" leaked in March 2023. Future leaks are not guaranteed.
Google's three moats (2023 framework):
| Moat | Structure | Why it holds |
|---|---|---|
| Consumer platform | Android + Chrome + app stores | Developers go where users are; users go where apps are. Google collects platform fees without building anything new. Microsoft has no equivalent consumer moat. |
| Enterprise | Google Workspace | Deeply embedded in SMB and non-profit. Not glamorous. Very sticky. |
| Data quality | Metadata on billions of users | Not just data — metadata. Describes behavior, intent, pattern. OpenAI trained on scraped web data of poor quality. Google's data is orders of magnitude better. |
On the consumer platform moat: Who builds the AI apps? Largely Apple and Google developers. Where do they distribute? Apple App Store and Google Play. Google can sit still and collect platform revenue while the AI ecosystem builds on top of it. This moat does not require Google to win the model race. It requires Google to remain the platform. As of 2023, no new device category threatened Android or Chrome's position.
On data: Two companies know their users better than the users know themselves — Google and Apple. Even as data privacy regulations tighten, metadata remains. Metadata reveals more than raw data. Any company wanting high-quality training data in 2023 would eventually need to come to Google. Google could simply decline.
The author's dissent from the memo's conclusion: "得程序员者得市场" (win the developers, win the market) misses the point. Apple's developers did not come for open source. They came because they could make money. Platform economics, not technical openness, drives developer behavior. Open source has never been the deciding factor in platform competition.
What the author did not predict (2026 update): The moat could evaporate not from external attack but from internal neglect. The 2023 analysis assumed Google would maintain its infrastructure position. The 2026 data suggests that assumption deserves scrutiny.