These are the 8 data warehousing vendors that show up when the paperwork says warehouse, ranked by who you contract with rather than by SQL dialect:
- Tinybird
- Snowflake
- Amazon Web Services
- Microsoft
- Databricks
- Oracle
- Teradata
Data warehousing vendors are not engines. They are companies you can sue, whose identity plane you inherit, and whose discounting motion procurement already understands. Databricks and Snowflake can both sit in a Gartner Cloud DBMS Magic Quadrant and still be opposite political choices inside 1 CIO staff meeting.
Tinybird is first because the paperwork is often for user-facing APIs, and the incumbent warehouse vendor will not sign that SLO. You still keep Snowflake, Google, AWS, or Microsoft for BI. You add a second, smaller contract for serving. The other 7 writeups are who owns the lake, the identity plane, and the year-2 discount.
The companion post best cloud data warehouse ranks meters and query shape. Use that after the economic buyer is named.
The 8 Data Warehousing Vendors
1. Tinybird for serving contracts
Tinybird is the warehousing-adjacent vendor you add when the SOW is an HTTP SLO, not a BI seat. The SKU is managed ClickHouse®: ingest (Events API or Kafka), SQL pipes as code, published HTTP endpoints, preview branches, and a 99.9% uptime SLA. Compliance on the paid path includes SOC 2 Type II, HIPAA, and GDPR. Self-managed Tinybird exists if residency forbids a SaaS control plane. You do not replace the Snowflake or Microsoft MSA with this. You stop asking that MSA to cover product APIs.
Commercial motion is product-led. Developer plans are monthly, $25 to $799 for 0.25 to 8 baseline vCPUs, with overage at $0.0002 per vCPU-second. There is no annual credit commit, no edition tax that reprices every credit in the account, and no lakehouse attach. The expansion path is more endpoints and more ingest, not Cortex, Mosaic, or an F64 that also runs Power BI. SaaS goes to 32 vCPUs on shared infrastructure. Enterprise is a dedicated cluster with named support. Who signs: engineering manager or CTO. Procurement often has no "real-time OLAP" category. Put the contract under application infrastructure, next to the API gateway, not on the Teradata or Snowflake MSA.
Lock-in that matters: pipe SQL, endpoint URLs, and resource tokens. That is lower than Unity Catalog or a Power BI semantic model, higher than "we only used S3 files." Moving to self-managed ClickHouse keeps the dialect. What you rebuild is ingest, tokens, and preview deploys. Leave cost is weeks to months, not a 18-month share unwind.
Write the SOW around a named SLO, not around "analytics platform": p95 under 100 ms on declared pipes, freshness under 5 seconds from Events API or Kafka, tenant-scoped tokens, preview branches per PR. Resend's 62 ms p90 and Audiense's "saved hiring 3-5 engineers" are the proof points that belong in the appendix. Do not ask procurement to rip out Snowflake. Ask for a second, smaller contract. The warehouse vendor keeps history, governed shares, and finance marts.
Tinybird is not a replacement motion for a 10-year Teradata estate. A CDO who only funds 1 enterprise warehouse and will not accept a second contract will still see Tinybird later as shadow IT. That is a worse way to buy it.
2. Snowflake
Snowflake reported $4.47 billion product revenue and $4.68 billion total revenue for FY2026 (year ended 31 January 2026), with remaining performance obligations of $9.77 billion. The Data Cloud motion is land a warehouse, expand with sharing, Snowpark, Horizon, Cortex. 1 contract on AWS, Azure, and GCP is the independent-vendor pitch. Mid-single-digit-billion scale is why every SI has a practice.
Discounting is aggressive in year 1 and tightens as credits become muscle memory. Edition choice is an account-level tax. Business Critical is $4/credit in AWS US East. That rate applies to every credit, not the 1 PHI schema. Teams that need it for 1 pipeline sometimes run a second cheaper account. That split is easier before the data lands.
The lock-in that actually hurts is Secure Data Sharing and the partner network. Customers may already have reader accounts. Leaving means rebuilding those shares, which is a customer-comms program, not a COPY INTO. Who signs: CDO / Head of Data. Engineering is rarely the economic buyer unless Snowpark is the app platform.
50 ms product APIs are not this vendor's economic motion. Snowflake alternatives is the product exit. The vendor exit is 6 to 18 months if shares and Snowpark jobs are real.
3. Google
Google sells BigQuery as the analytics SKU inside Google Cloud. Looker, Pub/Sub, Vertex, and GCS are the attach. Commercial motion is a cloud commit (CUD / EDP-style) plus BigQuery editions. Finance already has Google even when the data team wanted Snowflake.
Lock-in that matters: Looker modeling and authorized views. Slot culture (once reserved, overprovisioned). CMEK and VPC-SC designs that assume GCP. Who signs: the GCP account owner, often Infra or a Google-centric CDO.
If the rest of production is AWS, the year goes to interconnects. That is a vendor problem, not a SQL problem. BigQuery alternatives is the product-level exit, which is easier than the vendor-level exit.
4. Amazon Web Services
AWS sells Redshift (and Redshift Serverless) plus the argument that S3 is already the warehouse. Spectrum, Glue, Lake Formation, and IAM are the real products. Commercial motion is the Enterprise Discount Program. Redshift is rarely the reason for the AWS commit. It is the SKU that appears because the data is already in the account.
Gartner putting AWS highest on execution in Cloud DBMS reflects breadth (RDS, DynamoDB, Aurora, Redshift). It does not mean Redshift is the best warehouse. IAM policies and Lake Formation grants are the lock-in. So is the political cost of a second warehouse vendor on top of AWS. Snowflake-on-AWS is still 2 vendors. Who signs: Cloud CoE or FinOps, not always the analytics lead.
Amazon Redshift alternatives if the SKU is the problem. Redshift vs ClickHouse if the workload already left batch reporting.
5. Microsoft
Microsoft sells Fabric (OneLake, SQL endpoint, Data Factory) with Synapse still on many MSAs. Power BI is the gravity. E5 bundles and Azure consume the conversation. Warehouse capacity is an F-SKU next to Office. ISG and other 2026 platform reports keep Microsoft in the estate conversation even when engineers prefer Snowflake.
Lock-in that matters: Power BI datasets and semantic models. Direct Lake sounds portable until the report estate is the company. F64 is also the SKU that drops per-viewer Pro licenses, which is how Fabric gets pulled into the EA even when the warehouse engine is not the reason. Who signs: CIO / Microsoft account team. Data engineering often inherits Fabric after the EA is signed.
Sub-second multi-tenant APIs are not this vendor's economic motion. The purchase is Power BI capacity.
6. Databricks
Databricks sells a lakehouse platform. Unity Catalog, Delta/Iceberg, Spark, MLflow, Mosaic. SQL Warehouses are 1 storefront. Commercial motion: land with engineering (Spark jobs), expand to SQL and AI. A $5.4 billion revenue run-rate (early 2026, company-cited) is why boards treat Databricks as strategic, not a warehouse line item. Gartner has placed them high on lakehouse vision. That is a platform score, and 2025 was the first year they also participated on the operational axis (Lakebase).
Lock-in that matters: Unity Catalog as the grant plane. Notebook culture. Job compute that only the platform team understands. Who signs: VP Eng or Chief Data/AI Officer. Rarely a BI-only buyer.
A governed warehouse for finance is not the same purchase as a research lab. Databricks alternatives when the lakehouse is the overbuy. ClickHouse vs Databricks if the fight is already serving vs Spark.
7. Oracle
Oracle sells Autonomous Data Warehouse and the broader Oracle Cloud database family. Exadata heritage, Autonomous tuning, and the existing Oracle apps estate. Commercial motion: ULA renewals and OCI commits. ISG's 2026 analytic platforms research has ranked Oracle at the top of some categories. That tracks incumbent Oracle shops, not greenfield SaaS.
Lock-in that matters: PL/SQL, licensing audits, and the operational knowledge in the DBA org. Dumping to Parquet underestimates the stored procedures. Who signs: a CIO with an Oracle relationship older than the data team.
Greenfield product analytics with no Oracle footprint pays for a universe the team will not enter. 18 to 36 months is a normal ADW-to-cloud-native program, and most of that time is people, not pipes.
8. Teradata
Teradata sells VantageCloud, workload management, and industry models (telco, finance, retail) that took decades to encode. Commercial motion: long enterprise renewals, often hybrid (on-prem plus cloud). Analyst buyers-guides still list Teradata as exemplary for governed analytic platforms. New logo velocity is not Snowflake-class. Installed-base gravity is real. Recent fiscal prints have sat in the mid-$1.6B range with declining revenue.
Lock-in that matters: tactical workload rules and semantic models. Migrations are programs with systems integrators, not weekend cutovers. Union jobs and runbooks that only exist in 1 CoE's heads are part of the critical path. Who signs: line-of-business analytics plus the incumbent Teradata CoE.
A 30-person startup cannot staff this vendor. Do not put Teradata on a greenfield SaaS shortlist to be complete.
What the buying committee argues about
A typical shortlist meeting has 4 people who think they want the same thing.
The CDO wants 1 platform, 1 grant model, 1 invoice they can defend at the board. The VP Eng wants p95 and a team that will not page a warehouse SRE. The CIO wants the vendor already on the EA or the EDP. FinOps wants a meter they can forecast. Those 4 goals do not fit in 1 SKU. Pretending they do is how Databricks gets bought to please engineering while finance still lives in a Power BI dataset on Fabric.
Write the economic buyer on the whiteboard before the first demo. If that person is a CDO who already has a Snowflake relationship, this is a renewal, not a bake-off. If that person is a CTO who needs HTTP endpoints, this is an application-infrastructure RFP that someone mislabeled.
4 objects inside every warehouse MSA
A meter that will be fought quarterly. Credits, slots, RPUs, DBUs, DWUs, CUs, vCPU. The vendor's professional services org is trained to expand that meter, not to shrink it. Year-1 discounts exist to create muscle memory. Year-3 true-ups exist to collect it.
An identity and governance plane. Snowflake roles, GCP IAM plus VPC Service Controls, AWS IAM plus Lake Formation, Entra ID plus Purview, Unity Catalog, Oracle IAM, Teradata Viewpoint. Switching vendors is switching this plane, not exporting Parquet.
A gravitational product. Power BI, Looker, SageMaker, Snowpark, Databricks ML, Oracle apps, Teradata industry models. The warehouse discount exists to pull the account into that gravity. If ML, Office, or apps are not wanted next year, score that now.
A support ladder. Named TAM, severity-1 comms, regional residency, FedRAMP / IL4 / IRAP if required. Independent vendors (Snowflake, Databricks, Tinybird, Teradata) live or die on this. Hyperscalers bundle it with the cloud account, which is convenient until the Sev-1 is a ticket in the AWS queue.
If the RFP only scores ANSI SQL and SOC 2, the logo the SI already staffs will win.
Clauses that decide year 2
Rollover. Snowflake consumption contracts often let unused capacity roll at renewal if more is purchased. Some do not. A team that saved 30% of committed credits in year 1 can discover those credits vanish. Ask for the rollover sentence in writing. Hyperscaler commits (EDP, Azure EA, GCP CUD) roll differently: they are cloud-wide, so unused Redshift or BigQuery spend can be burned on EC2 or GCE instead.
Egress and sharing. Exporting Parquet to leave is rarely the expensive line. Egress to a customer, a second cloud, or a reader account is. Snowflake sharing avoids a copy and creates a relationship that must be unwound. AWS egress out of the region is a known tax. Score leave cost on shares and reader accounts, not on UNLOAD.
Termination assistance. 30 days of read-only access is not a migration. Ask for 90 days of production-equivalent read plus export bandwidth that does not require a new SOW. Independent vendors will negotiate this. Hyperscalers will point at the objects already in the bucket, which only helps if the semantics (views, row policies, metric defs) are in that bucket too.
Residency and support hours. FedRAMP, IL4, IRAP, and Sev-1 in the team's working hours are vendor facts, not feature rows. A US-only TAM on a Sydney-traded product is a no. Named TAM vs pooled queue is why independents win regulated deals even when the hyperscaler is cheaper.
Audit and most-favored. Public-sector and some banks will require audit rights on the meter. If the vendor cannot explain how a credit, slot, or DBU is constructed, FinOps cannot forecast it. That is a disqualification.
Systems integrators are a hidden 5th object. Accenture, Deloitte, Slalom, and the boutique Snowflake or Databricks shops staff what they can bill. An SI with 80 Snowflake architects and 4 Redshift people will not write a fair shortlist. Ask who is on the bench for this dialect before they write the RFP.
Analyst reports score a broader market than warehouses
Gartner's Cloud DBMS Magic Quadrant (18 November 2025) ranks a broader market than warehouses: AWS, Microsoft, Google, Oracle, Snowflake, Databricks, Teradata, and others in a category that includes operational databases. AWS's high execution score reflects the portfolio, not Redshift winning a warehouse bake-off.
Use the quadrant to see who analysts consider durable. Do not use it as a warehouse shortlist. ISG's 2026 Analytic Data Platforms buyers guide, for example, puts weight on Oracle, Databricks, and Teradata in ways a 40-person SaaS RFP never will.
Money is a clearer signal than a dot on a chart. Snowflake's FY2026 print is audited consumption revenue. Databricks's $5.4B figure is annualized run-rate. Boards treat both as strategic. Teradata's installed-base gravity is the game, not new-logo velocity.
How a switch actually runs
Open table formats (Iceberg, Delta) reduce file lock-in. They do not reduce semantic lock-in: metrics definitions, row access, shares, BI models. Budget the second.
Snowflake to BigQuery is often 6 to 18 months because of shares, roles, Snowpark, and partner tools. BigQuery to Snowflake is usually faster on files (6 to 12 months) and slower on Looker modeling and slot culture. Redshift to anything is 6 to 24 months of distkeys, Spectrum assumptions, and IAM sprawl. Fabric or Synapse to anything is 12 to 24 months because Power BI datasets are bound to the capacity. Databricks to a warehouse-only vendor is 12 to 24 months of Unity Catalog and Delta-as-system-of-record. Oracle ADW to cloud native is 18 to 36 months of PL/SQL and licensing. Teradata to cloud is 18 to 36 months of workload management, industry models, and the people who run them. Tinybird to self-managed ClickHouse is weeks to months: MergeTree SQL stays, APIs and ingest are rebuilt.
Those ranges assume an SI, a named executive sponsor, and a freeze on new shares or semantic models 6 weeks before cutover. Without those, add a year.
An RFP meeting agenda that scores the contract
90 minutes. 8 scores, 1 to 5, weights in parentheses. No 200-row feature spreadsheet.
- Economic buyer match (15). Does this vendor already have a relationship with the person who signs?
- Meter fit (20). Credits vs scan vs capacity vs vCPU against this team's QPS, not a vendor demo.
- Identity plane (10). Can IdP, SCIM, and row/column security be reused without a new religion?
- Leave cost (15). Files plus semantics plus shares plus BI models. Iceberg alone is not a 5.
- Attach gravity (10). Is the account okay being sold ML / Office / apps next year?
- Serving path (15). HTTP, tokens, preview deploys. Warehouses score 1 to 2 unless a gateway is staffed.
- Support that matches incidents (10). Sev-1 in working hours, residency, compliance pack.
- Staffing reality (5). Is the dialect (T-SQL, BQ SQL, Spark, MergeTree) already in-house?
A vendor that scores 5 on (1) and 1 on (6) is a fine BI contract and a bad product-analytics contract. Buy both if the weights say so. Do not pretend 1 MSA covers (6).
When 1 MSA cannot cover BI and APIs
Snowflake the product from Snowflake the vendor is a clean buy. Databricks SQL from Databricks is a clean buy. Redshift from AWS because the EDP forced it is common and often messy. Tinybird can be purchased without touching the warehouse vendor at all. That last pattern is the usual winning one for APIs.
If procurement asks for a single winner across BI and user-facing analytics, that winner is a pair of vendors. The warehouse vendor keeps history and governed shares. The serving vendor keeps p95. Anything else is a discount repaid in an incident.
Frequently Asked Questions (FAQs)
What is the difference between a warehousing vendor and a warehouse product?
The vendor is the legal entity on the MSA (Google, AWS, Microsoft, Snowflake Inc.). The product is the SKU (BigQuery, Redshift, Fabric, Snowflake the service). Redshift from AWS because of an EDP is a vendor decision. Snowflake on AWS is 2 vendors. Score both.
Can 1 vendor cover BI and user-facing APIs?
Warehouses score 1 to 2 on serving path unless a gateway is staffed. A CDO who wants 1 invoice can still fund a second, smaller contract under application infrastructure. Tinybird is that second contract when the SLO is p95 under 100 ms and tenant-scoped tokens.
How long does a warehouse vendor switch take?
File export is weeks. Semantic lock-in (shares, Looker models, Unity Catalog, Power BI datasets, PL/SQL) is 6 to 36 months depending on the pair. Iceberg reduces file lock-in only. Budget the second.
Do analyst Magic Quadrants pick the warehouse?
Gartner Cloud DBMS includes operational databases. AWS's execution score reflects RDS, DynamoDB, Aurora, and Redshift together. Use the quadrant for durability. Use the 8-point RFP agenda above for the warehouse contract.