Data lake.
A governed lakehouse over object storage — open table formats, cataloged and query-ready.
A lakehouse layer over object storage: open table formats, a unified catalog and ACID transactions, so the same data is query-ready for analytics, BI and the model without copies.
- Category
- Storage
- Deployment
- Managed → air-gapped
- Governance
- IAM · encryption · audit
The lake promised one copy. The warehouse quietly made another.
Data lakes were meant to be the single place all data lands — cheap, open, scalable. In practice the warehouse needed its own copy to serve SQL fast, the ML team extracted features into a third store, and BI cached a fourth. The lake became a staging area for systems that each kept their own truth, and governance fractured across all of them. The lakehouse closes that gap by making the lake itself transactional and queryable — adding the ACID tables and the catalog the warehouse used to provide, directly on the open files everyone already reads.
Open table formats: a database's guarantees on object storage.
Aether's lakehouse stores data as open table formats over object storage — ACID transactions, snapshot isolation, time travel and schema evolution, on files anyone can read. A failed job doesn't leave half-written data; a reader sees a consistent snapshot while a writer commits; you can query the table as of last Tuesday; and you can add or rename a column without rewriting petabytes. These are the guarantees that made the warehouse trustworthy, now available on the open storage the rest of the stack already uses.
One catalog, every engine — including the model.
The point of the lakehouse is that the SQL warehouse, the Spark pipeline, the BI tool and the model all read the same tables, in place, under one catalog and one permission model. There is no nightly copy into the warehouse so the dashboards can see the data, and no export to a separate AI stack so the model can train on it. Row- and column-level security and column-level lineage are defined once and inherited everywhere — a governed table is governed no matter which engine touches it, and the model is simply another engine reading governed data.
The lock-in is a format you can read, not a vendor you can't leave.
Because the storage is open table formats, the data is portable by construction — another engine can read it, and you can move it without a proprietary export. That is the difference between a lakehouse and a warehouse that happens to sit on cloud storage. And like every Aether service, it deploys where your data is allowed to live: managed, in your VPC, on-prem, or fully air-gapped — not only where a vendor offers a region.
Time travel and multi-engine, in one governed place: read a table as of a point in time from SQL — the same table a Spark job and the model also read, with no copy and no export.
-- Read the orders table as it was before the bad batch load
SELECT count(*), sum(total)
FROM lake.orders
FOR SYSTEM_TIME AS OF '2026-06-18 09:00:00';
-- The very same governed table — no copy — is where:
-- • the warehouse serves the dashboard
-- • a Spark job computes features
-- • the model trains and answers questionsOn a classic lake you'd reconstruct that point-in-time from raw files by hand, and each downstream system would hold its own copy with its own permissions. Here the table is transactional and time-travellable in place, and every engine — including the model — reads the one governed copy under one identity.
Three steps to running.
Write open table formats over object storage with ACID transactions.
One catalog and permission model across analytics, BI and ML.
Read from warehouse, Spark and the model with no extra copy.
Data lake, in full.
Iceberg/Delta-style tables with ACID transactions and time travel.
One catalog and permission model across analytics, BI and ML.
Add and change columns without rewriting the lake.
Read directly from warehouse, Spark and the model — no extra copy.
Provision it in a few lines.
Every service is reachable from the same SDK, CLI and infrastructure-as-code — one identity, one bill, one audit trail across the whole catalog.
import { aether } from "@aether/sdk";
const store = aether.storage.data_lake("training-data");
await store.put("run-42/weights.safetensors", file);
const url = await store.presign("run-42/weights.safetensors", { ttl: "1h" });At a glance.
- Formats
- Iceberg / Delta-style, ACID
- Catalog
- Unified, one permission model
- Time travel
- Yes
- Schema
- Evolution without rewrite
- Readers
- Warehouse · Spark · model
Built for real work.
Single source of truth for analytics
Feature stores for ML
Governed self-serve data
On one model, not stitched together.
The usual stack runs data lake in one product, the model in another and the data in a third — and the seams between them are the cost. Aether Cloud runs it on the same platform that serves the model, governs your identity and deploys into your boundary, with the rest of the catalog one hop away.
No stitching a vector DB to one place, a warehouse to another and a model to a third — data lake sits next to the rest of the catalog, one identity, one bill.
The provider that runs Aether runs your data lake — so the data and the model never leave the same governed boundary to talk to each other.
Need a capability that isn’t here yet? The model writes and deploys it into the same boundary — the catalog is a starting point, not a ceiling.
Good to know.
An open lakehouse with ACID tables and one catalog — without lock-in, and on the platform that serves the model.
No — query the lake in place, or pair it with the warehouse for the hottest workloads.
Yes — open table formats, so your data is portable and not trapped.
Your boundary, your choice.
Pairs well with.
Exabyte-scale, eleven-nines-durable object store with lifecycle tiering and presigned access.
High-IOPS network block storage that attaches to any instance, with snapshots and encryption.
Managed shared file systems — POSIX and high-throughput parallel options for training data.
Low-cost archival tiers for compliance and long-term retention, with fast restore when you need it.
Run Data lake on Aether Cloud.
A governed lakehouse over object storage — open table formats, cataloged and query-ready. Deployable managed, in your VPC, on-prem or fully air-gapped — talk to us about the configuration your workloads and your boundary require.