Home › Insights › The customer's data is deeper than any vendor's — so the winning vendors sell not models but structured ignorance: the willingness to know nothing about customer data and still produce useful output.
Insight 5

The customer's data is deeper than any vendor's — so the winning vendors sell not models but structured ignorance: the willingness to know nothing about customer data and still produce useful output.

Architect the whole offering around "we deliver value without ever touching your proprietary data." Pre-screening models on public + Matter Loop wet-lab data → ranked candidate set → customer's internal team validates against their data.

The "data is the moat" thesis is broken

The thesis breaks the moment you realize: the data belongs to the customer, not the startup. BASF doesn't buy from startups; they partner with academia and keep data internal. Customer keeps data → vendor flywheel never spins.

Vendor-side strategy: structured ignorance

Matter Loop should: take target spec + constraints from customer → return a ranked candidate set with synthesizability + qualification-path scores → customer's internal team validates against their data.

Evidence

"BASF generates 70M+ battery characterization data points/day and won't share. Compressed an 18-month project to 3 weeks using 150 years of proprietary data."
— BASF case study
"The data is often messy, handwritten, unstructured, or physically stored in cabinets."
— Killebrew (Citrine COO)
"MatterGen MIT-licensed, MACE-MP-0 open, PLaID++ ~50% better at SUN materials Sept 2025 — foundation-model moat is collapsing."
— MACE/NequIP benchmarks
All insights →