# Wh · Warehouse & Notebooks — element 45 of 58 > Where questions meet data. Turns raw tables into answered questions. - **Group:** 10 · Data & Analytics - **Necessity:** Optional - **Price band:** $ · under $30/mo - **Maturity:** Emerging - **Edition:** v2026.Q3 · verified 2026-09-13 ## Leading tools (v2026.Q3) - **Hex** — ai-native analytics workspace - **DuckDB / MotherDuck** — small data, big speed - **BigQuery** — serverless, generous free tier - **ClickHouse** — real-time speed at scale - **Marimo** — the reactive python notebook ## Our take DuckDB-class tools made 'small data' respectable. Most startups never need more — and AI writes the SQL now. ## Combines with An, Bi ## The top 5 — deep dossier (verified 2026-09-13) Hex for the notebook-and-answers layer, DuckDB/MotherDuck for the warehouse under it — most startups want exactly that pair and nothing heavier. Hex became the AI analytics workspace: notebook, threads, and semantic-model agents on your own context, shipped at a 2026 cadence (Agent Tasks, MCP, Evals) nobody else matches. MotherDuck made the small-data thesis a product — a free 10GB tier and a $250/mo org plan cover data volumes 90% of startups never exceed. BigQuery wins when you're GCP-native and want a 1 TiB/month free serverless warehouse plus up to $350k in startup credits; ClickHouse when user-facing, real-time analytics at billions of rows is the product; Marimo when you want the notebook itself to be open-source, reactive, and git-native for $0. 1. **Hex** (Hex Technologies) — Community free · Professional $36/editor/mo · Team $75/editor/mo · Enterprise custom · compute add-ons $0.32–$4.06/hr. Best for: One workspace where founders and data folk ask questions in SQL, Python, or plain English — and the answers become shareable apps, not screenshots. Why: Repositioned from 'collaborative notebook' to 'The AI Analytics Platform' and shipped like it through 2026: Agent Tasks (Jun 16), MCP client connectivity (May 28), web search for agents (Jun 11), and agent Evals (Aug). $70M Series C (May 28, 2025; Avra, a16z, Sequoia, Snowflake) with 1,500+ teams including Notion, Figma, Brex, Ramp, Reddit, and Anthropic. The Hashboard acqui-hire folded a BI team directly into the agent push. Watch: $36–75/editor/mo stacks up fast for small teams, and the free Community tier is genuinely limited (5 published apps, 7-day history). No 2026 raise verified — the war chest is the May 2025 round while Snowflake, Databricks, and OpenAI all build competing analyst agents. [https://hex.tech](https://hex.tech) 2. **DuckDB / MotherDuck** (DuckDB Labs (OSS) / MotherDuck (cloud)) — DuckDB free (MIT) · MotherDuck Lite free (10GB, 10 compute-hrs/mo, 3 users) · Business $250/org/mo + usage · compute $0.60–$24/hr per-second billed. Best for: The default startup warehouse: laptop-fast analytics on the gigabytes-to-low-terabytes most companies actually have, with a serverless cloud when you need to share. Why: The small-data movement's flagship, and it won the argument: DuckDB crossed 40k GitHub stars (Aug 5, 2026) with a mature 1.5.x line plus an LTS channel, and DuckLake v1.0 (Apr 2026) gave it a credible lakehouse format simpler than Iceberg. MotherDuck's Lite tier is free forever at 10GB, and 2026 launches — Flights (agent-native Python pipelines, Jun) and Guides (a context layer for AI agents, Jul) — aim it squarely at agent-driven analytics. Watch: Single-node scale-up has a ceiling — at tens of TB or high-concurrency BI you're shopping again. MotherDuck's last verified raise is the $52.5M Series B (Sep 2023); no round since surfaced on vendor or press pages, so runway is a fair diligence question. [https://motherduck.com](https://motherduck.com) 3. **BigQuery** (Google Cloud) — Free: 1 TiB queries + 10 GiB storage/mo · on-demand $6.25/TiB scanned · slots from $0.04/slot-hr · startup credits $2k–$350k. Best for: GCP-native startups that want a zero-ops, petabyte-capable serverless warehouse that costs approximately nothing until real scale. Why: The best startup deal among the big-cloud warehouses: the first 1 TiB of queries and 10 GiB of storage each month are permanently free, on-demand is a transparent $6.25/TiB, and the Google for Startups Cloud Program covers $2k (pre-funded) to $200k–$350k (funded / AI-first) of whatever you do spend. Nothing to size, patch, or resume — a real advantage over credit-metered rivals for teams with no data engineer. Watch: On-demand billing punishes careless SELECT * on wide tables, and costs get opaque once you graduate to slot editions. Gemini-in-BigQuery assistance trails the dedicated notebook agents, and lock-in to GCP is the price of the convenience. [https://cloud.google.com/bigquery](https://cloud.google.com/bigquery) 4. **ClickHouse** (ClickHouse, Inc.) — Open source free (Apache-2.0) · Cloud usage-based (scaled compute + storage, trial credits) · no public per-unit list prices. Best for: When analytics is the product — user-facing dashboards, observability, or agent telemetry needing sub-second queries over billions of rows. Why: The breakout warehouse company of the cycle: $400M Series D at a $15B valuation (Jan 2026), annualized revenue tripled to $250M by May 2026 on an IPO path, and the Langfuse acquisition (announced with the Series D) planted it in LLM observability — a workload startups actually have in 2026. Open-source ClickHouse remains free and self-hostable when the cloud bill stings. Watch: Overkill below billions of rows — DuckDB covers that for free. Cloud pricing has no public per-unit rates (calculator-only), self-hosting is real ops work, and joins/mutations still trail general-purpose warehouses. [https://clickhouse.com](https://clickhouse.com) 5. **Marimo** (marimo (CoreWeave since Oct 2025)) — Free, Apache-2.0 · notebooks are pure .py files · molab cloud free tier (GPU-backed, expanding under CoreWeave). Best for: Python-first builders who want notebooks that behave like software: reactive execution, git-diffable files, runnable as scripts, deployable as apps. Why: The open-source notebook with the momentum: 22.2k GitHub stars, users from OpenAI and Cloudflare to Stanford, and an AI-native editor with built-in SQL. Reactive cells kill the hidden-state bugs that make Jupyter notebooks unreproducible. CoreWeave's acquisition (Oct 30, 2025) funded the team from 2 to 8+ with a public pledge that the notebook stays free and permissively licensed, plus GPU compute behind the molab cloud. Watch: Ecosystem still a fraction of Jupyter's, and .py-file format means existing .ipynb workflows need conversion. Cloud roadmap (molab) now rides CoreWeave's strategy — fine today, worth watching as the parent's priorities shift. [https://marimo.io](https://marimo.io) ### How to choose - If Your working data fits in memory or on one SSD — which is 90% of startups, per the small-data thesis → DuckDB locally, MotherDuck Lite (free, 10GB) when you need sharing. Skip the distributed warehouse entirely; revisit only past low-terabyte scale. - If Non-analysts need answers, not just analysts needing queries → Hex — Threads for plain-English self-serve on a governed semantic model beats mailing notebook links, and Agent Tasks handles the recurring questions. - If You're on GCP or can get Google startup credits → BigQuery — 1 TiB/month free queries plus $2k–$350k program credits means your warehouse is effectively free until product-market fit. - If Analytics is user-facing product — dashboards in your app, observability, agent telemetry → ClickHouse (Cloud, or Tinybird for the managed-API path); DuckDB-class engines aren't built for that concurrency. - If You want notebooks in git, reviewed like code, at zero cost → Marimo — pure .py files with reactive execution; fall back to Jupyter only when a library or workflow demands .ipynb. ### The field (21 more) Snowflake, Databricks, Deepnote, Jupyter / JupyterLab, Google Colab, molab, Observable, Evidence, Count, Tinybird, DuckLake, Firebolt (fading), Bauplan, Fabi.ai, Querybook (fading), Apache Zeppelin (fading), Mode (acquired), Hashboard (acquired), Hyperquery (acquired), Noteable (dead), Zepl (dead) Full dossier data: https://elems.ai/e/wh.json --- Source: [elems.ai](https://elems.ai/e/wh.html) — the periodic table of the AI-led startup. Data: https://elems.ai/elements.json (CC BY 4.0, cite elems.ai).