I'm a data ontologist by trade. Here's why I think the market has the AI moat story backwards on beaten-down SaaS names - and why I own FDS into earnings tomorrow.

The consensus AI infrastructure trade is Snowflake, Databricks - raw data platforms. Everyone agrees on this. The market has priced it accordingly.

My day job is literally taking messy, ungoverned data and building the semantic structure - the ontology - that makes it usable. Twenty years doing this at places like Nike and SurveyMonkey, plus building my own 1,700-ticker XBRL screener as a side project.

That experience tells me the consensus trade is missing the actual moat layer.

Any AI agent trying to replace FactSet, Veeva, Roper, or SPGI will ironically need the data those companies have spent decades validating and structuring. You can't scrape it. You can't synthesize it from the open web. A general purpose model trained on Reddit and public data will know nothing about pharmaceutical regulatory submissions (Veeva), county tax administration (Roper), or financial model construction (FactSet) — it'll produce plausible-sounding fluff instead of the real thing.

Meanwhile these names are down 20-30% in the SaaSpocalypse selloff, getting lumped in with companies that actually do have weak moats.

Full framework - four categories of AI-relevant data, why raw infrastructure is overpriced relative to validated context, and where the true FCF yields actually sit - in the piece. I own FDS and ADBE; FDS reports earnings tomorrow morning so we'll get a live data point on whether the thesis holds. RDDT is also extremely interesting from this angle.

[https://cavemanscreener.substack.com/p/context-is-50-iq-points-part-ii-data](https://cavemanscreener.substack.com/p/context-is-50-iq-points-part-ii-data)