For three years the defining question of the open data stack was which lakehouse: Iceberg or Delta, Polaris or Unity, engine A or engine B. In 2026 the question changed. It is now who owns your lakehouse — because the companies that built the independent layer are being bought, and for the first time we have an auditable price on what one of them was worth.
The scorecard, with receipts
SAP → Dremio. Announced May 4, closed July 6. SAP never disclosed terms in a press release — but its Form 6-K filed with the SEC discloses cash consideration of approximately €0.5 billion, and states that the year's acquisitions dilute SAP's 2026 non-IFRS operating profit by more than €100 million (primary regulatory filing). That makes Dremio the largest disclosed-price deal in this consolidation wave, and the filing is the receipt. The strategic logic is explicit: SAP is rebuilding Business Data Cloud as an Iceberg-native lakehouse so AI agents can query SAP and non-SAP data without ETL. Dremio's open-source commitments — Iceberg, Polaris, Arrow — are maintained post-close, per both companies. That commitment is vendor-published and we will be watching whether it survives its first roadmap collision.
Databricks → Panther. Databricks' third security acquisition, announced this year, to build what it calls the "security lakehouse" — SIEM workloads restructured as lakehouse workloads with an agentic layer on top (vendor announcement). Read it structurally: Databricks is not buying customers, it is buying workload categories to run on the same table format. Security telemetry is one of the largest append-only datasets an enterprise owns, which makes it lakehouse-shaped by nature.
NetApp → DataPelago. Closed July 17. NetApp took DataPelago's Nucleus engine — built on open-source Gluten, Velox, and Substrait — to accelerate Spark and Trino against existing data stores, positioning NetApp as what its coverage called a "full-stack AI data infrastructure provider" (Blocks & Files). A storage hardware incumbent buying the query acceleration layer is the tell: the margin is migrating up-stack, and everyone can see it.
The pattern
Three different acquirers — an application giant, a data platform, a storage incumbent — and one consistent target profile: Iceberg-ecosystem assets. The analyst framing that holds up best across the year's coverage is the value-in-the-join thesis: the application company holds business context, the data platform holds analytical capability, and the acquisition captures the join between them. Dremio was bought for exactly that join. Panther extends it to security data. DataPelago extends it to the compute path.
This is also, precisely, the thing the open-format movement said would happen — and was supposed to be protection against. Open table formats made the data layer portable, so acquirers stopped buying data gravity and started buying everything around the format: the catalogs, the engines, the accelerators, the vertical workloads. The format stayed open. The companies did not.
Who is still independent
The honest census, layer by layer, as of mid-September 2026: in query engines, Starburst, ClickHouse, and MotherDuck remain independent, alongside the DuckDB project itself — whose 2.0 "Cyanoptera" release entered alpha on September 2 with GA projected for late October, carrying the client-server mode that moves it closer to the territory the acquired vendors occupied. In independent object storage, Backblaze's Q1 numbers are the cleanest public evidence the layer is worth owning: B2 up 24% year over year, AI customers up 76% and expanding three times faster than average, guidance raised — all in an SEC filing, not a pitch deck. Wasabi took its first price increase in company history this July and held its flat-rate model. The post-MinIO field we track on the sovereign storage front remains unconsolidated — and, notably, un-acquired so far.
Why 2026 is the year
The economics chose this moment, and the numbers are on the record. AI demand has reshaped storage-supplier revenue so thoroughly that the trade press now describes memory vendors' rise as "the memory super cycle made visible" — while NAND contract prices, after compounding roughly 3.5–4× in a year, moderated to +10–15% in Q3: a plateau at a permanently elevated baseline. Hardware got expensive and stayed expensive; the differentiation — and the acquirable value — moved to the software that decides how data reaches GPUs. Meanwhile the enterprise survey base keeps ratifying the destination: Forrester's Q1 landscape found 87% of organizations expect the lakehouse to be their primary analytics architecture by 2027. Incumbents are not speculating on the lakehouse. They are buying the finished migration.
We put a dated, falsifiable version of this read on the calls ledger — including exactly what would make us revise it. That is where our opinion lives; this page is the record.
Works cited
- SAP Form 6-K exhibit (SEC EDGAR) — the ~€0.5B consideration and >€100M dilution disclosures.
- SAP completes Dremio acquisition (SAP News) — close date and open-source commitments.
- Databricks agrees to acquire Panther (Databricks newsroom) — the security-lakehouse framing.
- NetApp buys DataPelago (Blocks & Files, July 17, 2026) — deal and Nucleus/Gluten/Velox/Substrait detail.
- A new era emerges: storage supplier revenue 2020–2026 (Blocks & Files, September 7, 2026) — the revenue-reshape analysis.
- Backblaze Q1 2026 8-K exhibit (SEC EDGAR) — B2 +24%, AI customers +76%, raised guidance.
- TrendForce — AI server demand supports memory prices in 3Q26 (July 3, 2026) — the Q3 +10–15% moderation.
- Try DuckDB v2.0-alpha (duckdb.org, September 2, 2026) — alpha scope and late-October GA projection.
- Dremio recognized in Forrester Data Lakehouses Landscape Q1 2026 (GlobeNewswire) — the 87%-by-2027 survey figure.