Theses›Datacentre & AI compute›Data infrastructure

Data infrastructure

Durable value in the AI data stack accrues to whoever owns the WRITE path and the data gravity — storage, governance, pipelines, streaming, observability — not the read/retrieval path; retrieval commoditises into Postgres extensions and model-API context windows while lakehouse…

  • Bear
  • Closing
  • $505B by 2035

The callAgents need a governed place to write and prove their work, but Databricks and Snowflake are building it; the route is public markets, not a startup.

D1D2R1R2R3C1foragainst D1D2R1R2R3C1foragainst
Fig. 1The balance. Drivers and catalysts against restraints, each block sized by its weight; catalysts hang dashed because they have not happened yet, sized by their odds. It tips toward the restraints.
  1. D1Fleets of AI agents reading, writing and acting on company data all day need lineage, memory and governance that warehouses built for analysts do not track.
  2. D2EU AI Act lineage and governance duties put dated regulatory spend on the write path.
  3. R1Incumbents hold the data and ship every new sub-layer as a bolt-on, so the residue a startup can own shrinks toward the retrieval sliver.
  4. R2The write-path view is the published consensus, with vector databases still the fastest-growing category, so the view is no longer early or different.
  5. R3The startup-reachable market fell from $22B to about $9B when checked, and the 2035 share depends on 2020-21 companies compounding at 50% for a decade on undisclosed revenue.
  6. C1The EU AI Act's Annex III high-risk duties become enforceable, putting lineage and governance spend on the write path.2 Dec 2027
Rests onthe beliefs underneath
Supported by · confidence
What would prove itdated tests, open and resolved
20272028today1234predictionswatch 20272028today1234predictionswatch
Fig. 2Each ring is a dated test: dashed while open, struck through when overdue, ticked or crossed once resolved.
  1. 1Whether a standalone vector database still grows above 30% with expanding margins while long-context adoption rises; if so, this view was wrong.Mar 2027 · open
  2. 2AI Act high-risk duties in force and the governance spend that follows.Dec 2027 · open
  3. 3By end-2027 value in the AI data stack concentrates in the WRITE path while standalone vector/retrieval erodes.Dec 2027 · open
  4. 4Databricks' IPO, the public-markets route into the write path.Dec 2027 · open
Expressionslisted, private, technologies
  • DDOGDatadog
    • cheap
  • IBMIBM
    • full
     Superconducting qubits roadmap (Heron, Condor, Kingfisher).
  • MDBMongoDB
    • expensive
  • SNOWSnowflake
    • expensive
  • AMZNAmazon Web Services
    • fair
     Sells managed cloud databases and the Redshift warehouse to enterprises and developers (Amazon subsidiary; parent ticker).
  • MSFTMicrosoft
    • full
     Sells SQL Server licences and Azure managed databases to enterprises.
  • ORCLOracle
    • not screened
     Sells Oracle Database and MySQL licences and cloud database services to enterprises.
  • GOOGLGoogle Cloud
    • cheap
     Sells managed cloud databases and the BigQuery warehouse to enterprises (Alphabet subsidiary; parent ticker).
Private · 14 private companies, 10 of them in the US; none in continental Europe.

ElectricSQL (Electric DB Limited) · Xorq (LETSQL Inc.) · Embucket, Inc. · Bruin Data Limited · ClickHouse · Supabase · dbt Labs · Neon · Pinecone · Airbyte · ElectricSQL · Embucket · Bruin · Xorq

Technologies
Who has already bet on itprivate rounds, from public sources

$2.8B raised by 10 companies, $1.3B of it in the last three years. Wave under way: Large rounds in the last three years; the category is funded and getting crowded.

Backed by a top-tier venture firm: ClickHouse, Supabase, dbt Labs, Airbyte, Pinecone, Neon, ElectricSQL. Backed by a strategic investor: Supabase, dbt Labs, Airbyte, ElectricSQL.

2 more funded companies in the market.

What changedthe record, newest first
  • 26 Jul 2026The write-path call became the crowded trade while I was still calling it contrarian.