Fourteen products, one page.
What each one does, and what each one charges.
The platforms a data team already owns and the point products it buys to fill their gaps, beside Data Oil: what each one is, what it does inside its own engine and what it makes you buy for the rest, what it charges at its own published list, and what a team already running it gains by putting Data Oil beside it. Every vendor figure is that vendor's own published price, and every one of them is named below with the page it came from and the day it was read.
The short version
Everything here does part of the job
A warehouse holds rows and scans them
Snowflake, Databricks, Redshift, BigQuery and Fabric are very good at that, and cheap at it. None of them indexes for relevance, embeds for meaning, transcribes, translates or writes an answer without a second product turned on beside it.
A point product does one stage well
Algolia ranks text, Pinecone and turbopuffer hold vectors, Firestore reads documents cheaply, Deepgram turns audio into text. Each is good at its stage and none of them is a database and an index and an answer.
A reporting tool draws the result
Power BI, Tableau, Qlik and Excel are not competitors and this page does not treat them as any. They connect to Data Oil over the PostgreSQL wire protocol with no adapter.
Data Oil is one engine that does all three parts — the rows, the index and the answer — over the same records, on one bill.
Side by side
What each one is, and what it charges
Data Oil stays on the page. Open any product below and it opens directly underneath ours, on the same three questions: what it is for, what the same job makes you buy beside it, and its own published list price. The last one is what decides a bill, and it is the one vendors put behind a calculator. Open as many as you like — they stack, all of them under ours.
Data OilAlways shown
- What it is
- One engine over rows, documents, relationships, full text, vectors, media and eighteen languages, with an assistant that answers against your own model
- What the same job makes you buy beside it
- Nothing. The index is the database, the graph is the same records, and the assistant is a call on them
- What it charges, at its own published list
- $0.02 per 1,000 units and $0.10 a GiB-month, on a card of 115 meters you can read before you are chargedA statement is one unit; a search is 5 to 10 by the data it examined
SnowflakeCloud data warehouseCompare
- What it is
- A cloud data warehouse: SQL over columnar storage, with elastic virtual warehouses you size and run
- What the same job makes you buy beside it
- Cortex Search for relevance, Document AI for files, a Cortex function per translation, Cortex Analyst or Agents for answers - four services, four meters
- What it charges, at its own published list
- $2.00 a credit on Standard, $3.00 Enterprise, $4.00 Business Critical; $23.00 a TB-month on demandA warehouse burns credits by size and by the hour it is awake, from one credit an hour at X-Small
DatabricksLakehouse over your own storageCompare
- What it is
- A lakehouse: Delta tables in your own object storage with SQL warehouses, notebooks and MLflow over them
- What the same job makes you buy beside it
- Mosaic AI Vector Search, an index that syncs from a table;
ai_parse_documentfor files;ai_translateper token; Agent Bricks or Genie for answers - What it charges, at its own published list
- $0.70 a DBU-hour on SQL Serverless, $0.55 Pro, $0.22 classicPlus your own object storage, and on classic compute the virtual machines underneath, which the DBU rate does not include
Amazon RedshiftColumnar MPP warehouse on AWSCompare
- What it is
- A columnar MPP warehouse on AWS, provisioned or serverless, and the rest of the estate around it
- What the same job makes you buy beside it
- OpenSearch for search and vectors, Neptune for graph, Textract, Transcribe and Rekognition for files and media, Translate for language, a Bedrock knowledge base for answers
- What it charges, at its own published list
- $0.375 an RPU-hour on Serverless, 4 RPUs at the smallest base capacity; $0.024 a GB-month of managed storageCharged by the second with a 60-second minimum
Google BigQueryServerless warehouse, charged on what it readCompare
- What it is
- A serverless warehouse charged on what each query read, with search, vector and graph indexes of its own
- What the same job makes you buy beside it
- Document AI for files, a Vertex AI connection for translation and embeddings, Vertex AI Search for answers
- What it charges, at its own published list
- $6.25 a TiB scanned on demand and $0.02 a GiB-monthThe first TiB scanned and the first 10 GiB stored are free each month
Microsoft FabricOne capacity every workload draws onCompare
- What it is
- One capacity you buy in advance and every workload draws on - warehouse, lakehouse, Eventhouse, Power BI
- What the same job makes you buy beside it
- AI functions charged per call, Copilot charged against the same capacity, and a Power BI seat for everyone who reads the result
- What it charges, at its own published list
- $0.18 a capacity-unit hour pay-as-you-go, $0.54 for overage; OneLake $0.026 a GB-month hotAn F2, the smallest SKU, is two capacity units
Elastic CloudA search engine with a store under itCompare
- What it is
- A search engine with a document store under it: BM25 and HNSW over the same index, and ES|QL to query it
- What the same job makes you buy beside it
- A database for the rows, the transactions and the joins, and the job that keeps it and the index in step
- What it charges, at its own published list
- $0.14 an ingest VCU-hour and $0.09 a search VCU-hour; $0.047 a GB-month retained, $0.05 a GB out
MongoDB AtlasDocument database, search on its own nodesCompare
- What it is
- A document database with Atlas Search and Atlas Vector Search running on their own search nodes
- What the same job makes you buy beside it
- Search nodes as a tier of their own, and a separate product for anything to do with media, documents, translation or answers
- What it charges, at its own published list
- M10 $0.08, M30 $0.54, M50 $2.00 an hour; search nodes from $0.24 an hourDedicated clusters in AWS US East
AlgoliaA hosted search API over a copy of your recordsCompare
- What it is
- A hosted search API. You send it a copy of your records and it ranks them well
- What the same job makes you buy beside it
- The database the copy came from, and the job that keeps the two in step on every write
- What it charges, at its own published list
- $0.50 per 1,000 searches after the first 10,000 a month, and $0.40 per 1,000 records held after the first 100,000The records line is charged every month for as long as you hold them, not once
PineconeManaged vector databaseCompare
- What it is
- A managed vector database: embeddings in, nearest neighbours out
- What the same job makes you buy beside it
- The records themselves, a full-text index for anything matched by word, and the pipeline that embeds and syncs into it
- What it charges, at its own published list
- $4.00 to $4.50 per million write units, $16.00 to $18.00 per million read units, $0.33 a GB-monthStandard serverless, at the cheapest cloud and region
turbopufferA vector index, and only thatCompare
- What it is
- The cheapest place to keep a vector index, and only that
- What the same job makes you buy beside it
- The rows, the relationships and the transactions, which live elsewhere and are kept in step by code your team owns
- What it charges, at its own published list
- $0.33 a GB-month, up to $2.00 a GB written, $1.00 a PB queried with a 1.28 GB minimum per queryA $16.00 monthly minimum on Launch
FirestoreServerless document store, no full-text at allCompare
- What it is
- A serverless document database, very cheap per document read, with no full-text search in the product at all
- What the same job makes you buy beside it
- A search engine beside it - Google's own guidance - and a separate product for vectors at scale, media, documents, translation and answers
- What it charges, at its own published list
- $0.03 per 100,000 reads and $0.09 per 100,000 writes; $0.18 a GiB-month, $0.12 a GB outA daily free quota of 50,000 reads and 20,000 writes, in us-central1
DeepgramTranscription. Audio in, text outCompare
- What it is
- A transcription API. Audio in, text out, and nothing else
- What the same job makes you buy beside it
- Everywhere the text then has to go: the store, the index, the embeddings, the translation, and whatever answers questions about it
- What it charges, at its own published list
- $0.0043 a minute on Nova, batch - $0.26 an hourThe cheapest of the four transcription APIs on this card; ours is $0.24 an hour
Power BI and TableauReporting front-ends. Not a databaseCompare
- What it is
- Reporting front-ends. They draw what a database hands them, and they are not a database
- What the same job makes you buy beside it
- The database underneath, and the pipeline that carries data into the semantic model or the extract
- What it charges, at its own published list
- Power BI $14.00 a user a month on Pro and $24.00 Premium Per User; Tableau from $15.00 a user a month on Standard and $35.00 on EnterpriseData Oil does not replace either of these. It connects to them — see below
List prices for a US region; enterprise discounts, reserved capacity and regional differences are not modelled. Every figure, the assumption its conversion made and the page it came from is in the sources at the foot of this page.
Capability
In the engine, or a separate service
Almost every product here can do full-text or vectors somehow, so a column of ticks would say yes fourteen times and tell you nothing. What decides the architecture is whether a capability is one index over the records the SQL already reads, or a second product with its own meter, its own permissions and its own sync to fall behind. Open the one you run and it lands under ours, capability by capability.
Data OilAlways shown
- SQL over your rows
- In the engine
- Full-text ranking
- In the engineBM25
- Vector search
- In the engineHNSW
- Graph
- In the engineVertex and edge collections, traversal and pattern matching
- Files, media and documents in
- In the engine31 formats; a document's tables become rows
- Translated on write
- In the engine18 languages, charged once however many you ask for
- Answers with citations
- In the engineRewritten, planned, retrieved four ways, reranked, cited
- Your model, your key
- In the engineYou register the endpoint; the key is sealed and never leaves
SnowflakeCloud data warehouseCompare
- SQL over your rows
- In the engine
- Full-text ranking
- Separate serviceCortex Search
- Vector search
- In the engineVECTOR type; embeddings are a Cortex function
- Graph
- –recursive SQL, not traversal
- Files, media and documents in
- Separate serviceDocument AI and Cortex parsing
- Translated on write
- Separate servicea CORTEX.TRANSLATE call per row, per token
- Answers with citations
- Separate serviceCortex Analyst and Cortex Agents
- Your model, your key
- –Cortex runs Snowflake's own models
DatabricksLakehouse over your own storageCompare
- SQL over your rows
- In the engine
- Full-text ranking
- Separate servicehybrid keyword in Mosaic AI Vector Search
- Vector search
- Separate serviceMosaic AI Vector Search, an index synced from a table
- Graph
- –GraphFrames, a library over Spark
- Files, media and documents in
- Separate serviceai_parse_document and Auto Loader
- Translated on write
- Separate serviceai_translate, per token
- Answers with citations
- Separate serviceAgent Bricks and Genie
- Your model, your key
- In the enginea model serving endpoint
Amazon RedshiftColumnar MPP warehouse on AWSCompare
- SQL over your rows
- In the engine
- Full-text ranking
- –add Amazon OpenSearch
- Vector search
- –add OpenSearch or Aurora pgvector
- Graph
- –add Amazon Neptune
- Files, media and documents in
- –add Textract, Transcribe, Rekognition
- Translated on write
- –add Amazon Translate
- Answers with citations
- –add a Bedrock knowledge base
- Your model, your key
- Separate serviceRedshift ML through Bedrock
Google BigQueryServerless warehouse, charged on what it readCompare
- SQL over your rows
- In the engine
- Full-text ranking
- In the engineSEARCH() over a search index
- Vector search
- In the engineVECTOR_SEARCH over a vector index
- Graph
- In the engineBigQuery Graph, ISO GQL
- Files, media and documents in
- Separate serviceDocument AI and object tables
- Translated on write
- Separate serviceML.TRANSLATE through a Vertex connection
- Answers with citations
- Separate serviceVertex AI Search
- Your model, your key
- Separate serviceremote models through Vertex AI
Microsoft FabricOne capacity every workload draws onCompare
- SQL over your rows
- In the engine
- Full-text ranking
- –full-text in the SQL database item only
- Vector search
- Separate serviceper item: Eventhouse or the SQL database
- Graph
- –
- Files, media and documents in
- Separate serviceAI functions and Data Factory
- Translated on write
- Separate servicean AI function per call
- Answers with citations
- Separate serviceCopilot, against the same capacity
- Your model, your key
- Separate serviceyour own Azure OpenAI endpoint
Elastic CloudA search engine with a store under itCompare
- SQL over your rows
- –documents and ES|QL, not transactions and joins
- Full-text ranking
- In the engineBM25
- Vector search
- In the engineHNSW
- Graph
- –a visualisation, not traversal
- Files, media and documents in
- Separate serviceingest pipelines and connectors
- Translated on write
- –
- Answers with citations
- Separate servicethe AI Assistant and Playground
- Your model, your key
- In the engineinference endpoints
MongoDB AtlasDocument database, search on its own nodesCompare
- SQL over your rows
- –documents with transactions; SQL through the Atlas SQL interface
- Full-text ranking
- Separate serviceAtlas Search, on its own nodes
- Vector search
- Separate serviceAtlas Vector Search, on its own nodes
- Graph
- –$graphLookup, within one collection
- Files, media and documents in
- –
- Translated on write
- –
- Answers with citations
- –
- Your model, your key
- –you write the retrieval
AlgoliaA hosted search API over a copy of your recordsCompare
- SQL over your rows
- –a copy of your records, not the records
- Full-text ranking
- In the engine
- Vector search
- In the engine
- Graph
- –
- Files, media and documents in
- –
- Translated on write
- –one index per language, each charged again
- Answers with citations
- Separate serviceAsk AI, a product of its own
- Your model, your key
- –
PineconeManaged vector databaseCompare
- SQL over your rows
- –
- Full-text ranking
- In the enginesparse vectors, not BM25 over your rows
- Vector search
- In the engine
- Graph
- –
- Files, media and documents in
- –
- Translated on write
- –
- Answers with citations
- Separate servicePinecone Assistant
- Your model, your key
- –you write the retrieval
turbopufferA vector index, and only thatCompare
- SQL over your rows
- –
- Full-text ranking
- In the engineBM25
- Vector search
- In the engine
- Graph
- –
- Files, media and documents in
- –
- Translated on write
- –
- Answers with citations
- –
- Your model, your key
- –
FirestoreServerless document store, no full-text at allCompare
- SQL over your rows
- –documents; no joins
- Full-text ranking
- –Google's guidance is a third-party engine
- Vector search
- In the engineKNN, no index tuning
- Graph
- –
- Files, media and documents in
- –
- Translated on write
- –
- Answers with citations
- –
- Your model, your key
- –
DeepgramTranscription. Audio in, text outCompare
- SQL over your rows
- –
- Full-text ranking
- –
- Vector search
- –
- Graph
- –
- Files, media and documents in
- In the engineaudio only, and the text comes back as a file
- Translated on write
- –
- Answers with citations
- –
- Your model, your key
- –
Capabilities as each vendor describes its own product on its pricing and documentation pages. Separate service means the vendor offers it and meters it apart from the engine; a dash means you add somebody else's product. Nothing on this table is a criticism of any of these products doing the job it was built for — a vector store that only holds vectors is not failing at anything.
Price, one slice at a time
Holding half a terabyte and answering 60,000 statements
The same slice of work on every side: hold 500 GiB and answer 60,000 statements a month, at each vendor's smallest unit, 8 hours a day for 22 working days. Nothing in this table is indexed for search, embedded, translated or answered — that is two tables down.
| Holding it and querying it, and nothing else | A month, USD |
|---|---|
| Microsoft Fabrican F2 capacity, paused outside those hours, with OneLake storage | $77.32 |
| Amazon RedshiftServerless at four RPUs, plus managed storage | $276.88 |
| Snowflakean X-Small warehouse on Standard, plus storage at $23 a TB-month | $364.35 |
| Google BigQueryon demand: no hours at all, charged on what each query scanned | $369.76 |
| Databricks SQLa 2X-Small Serverless warehouse; your own object storage not counted | $492.80 |
| Data Oil, with no size at allthe storage and the statements, and nothing for capacity kept warm | $55.93 |
| Data Oil, with the Large class held every hour of the monthall 730 hours, not 176 | $593.21 |
Between $77.32 and $492.80 a month, and they are all doing the same thing: holding rows and scanning them. Ours is shown twice because the honest answer is two numbers — with no size at all you pay for the storage and the statements and nothing for capacity kept warm, and with a machine held every hour of the month you pay for the machine, which is what every row above also does for 176 of the month's 730 hours. This is the comparison a warehouse wins, and it is on this page for that reason.
+What every one of these five figures assumedFive choices
| Platform | The choice this figure made |
|---|---|
| Snowflake | an X-Small virtual warehouse, the smallest Snowflake sells, at one credit an hour on the Standard edition in AWS US East (Northern Virginia), and storage at the on-demand rate with no compression benefit claimed |
| Databricks | a 2X-Small SQL Serverless warehouse, the smallest Databricks SQL offers, at four DBU an hour. Table storage stays in your own object storage and is not counted here at all, which makes this line cheaper than the bill would be |
| Amazon Redshift | Redshift Serverless at the smallest base capacity the pricing page allows, four RPUs, in US East (Northern Virginia); managed storage on the same data |
| Google BigQuery | active logical storage and on-demand analysis in the US multi-region, with the monthly free tier deducted - 10 GiB of storage and the first TiB scanned; batch loading is free, because nothing is done to the data on the way in. Google lists active logical storage hourly, and the hourly figure works out at $0.023 a GiB-month rather than the $0.02 the headline states |
| Microsoft Fabric | an F2 capacity - two capacity units, the smallest Fabric SKU - paid as you go and paused outside the hours the comparison assumes, with OneLake storage at the hot rate. A real half-terabyte workload wants a larger capacity; F2 is the published floor |
| Data Oil | the same 500 GiB at $0.10 a GiB-month and the same 60,000 statements, each returning 20 rows and about 50 KB; then the same again with the Large compute class held all month |
Where a choice could go either way it went the way that makes the other platform cheaper, which is why Databricks is priced with its table storage left out entirely and Fabric with the smallest capacity Microsoft sells. A comparison that flatters itself is one the reader can take apart.
Price, rate for rate
The five point products, against the meters they replace
A point product is bought for one line of the card, so it can be compared on that line and nowhere else. Here are the seven rates where Algolia, Pinecone, turbopuffer, Firestore and the transcription APIs price the same thing we do.
| What you are buying | Data Oil | The point product | Ours, as a share |
|---|---|---|---|
| Reading a thousand records | $0.0185 per 1,000 | Firestore, $0.03 per 100,000 document reads$0.0003 per 1,000 | 6167% |
| Writing a thousand records | $0.0239 per 1,000 | Firestore, $0.09 per 100,000 document writes$0.0009 per 1,000 | 2656% |
| A search | $0.15 per 1,000 | Algolia, $0.50 per 1,000 searches, and $0.40 per 1,000 records held every month$0.50 | 30% |
| A row made retrievable | $0.0005 per 1,000 | Pinecone, $4.00 to $4.50 per million write units$0.004 per 1,000 | 12% |
| A gigabyte taken in | $0.50 per GiB | turbopuffer, $2.00 per GB written$2.15 | 23% |
| A gigabyte held | $0.10 per GiB-month | Neon $0.35, turbopuffer $0.33, Supabase $0.125$0.38 | 27% |
| An hour of audio transcribed | $0.24 an hour | Deepgram $0.26, Whisper $0.36, Google $0.96, AWS $1.44$0.26 | 93% |
Two of these rows go the other way, and they are on the page for that reason. Firestore charges per document read and is genuinely cheaper at it than we are — that is its whole model, and a key-value read is the one thing it is built to sell. What it does not do is rank text at all: Google's own documentation tells you to put a search engine beside it, and the moment you do, the comparison is the one two sections down rather than this one. Everywhere else on this table we are between 3% and 93% of the product being replaced.
Price, the whole job
333 times, for the same work
A warehouse holding rows is not the purchase anybody actually makes, and neither is a vector store on its own. The real job is the one this pack prices line by line: 500 GiB loaded once, a gigabyte of its free-text column translated into Japanese, and 100 people querying it for a month. Bought as products, that is three of them.
| The same job, whole, either way | A month, USD |
|---|---|
| A warehouse, a search product and a translation APIBigQuery to hold and query it, Algolia to make it findable, AWS Translate for the Japanese, all at list | $1,358,653.17 |
| Data Oil, one billheld, indexed, embedded, translated, and every statement answered | $4,083.79 |
The line that makes 2.7 billion rows findable is $1,342,177.28 of that total, every month, forever — which is why swapping the warehouse barely moves it: the whole spread between the cheapest and the dearest warehouse above is 0.03% of this figure. The assistant is out of both columns, because none of those three products answers a question at all; here it is $299.73 more on the same bill.
What you gain
If you already run one of these
Nobody replaces a warehouse on a Tuesday, and this page does not ask you to. Every row below starts from what you keep.
| If you already run | What stays exactly as it is | What Data Oil puts beside it | What stops being a thing you run |
|---|---|---|---|
| Snowflake | Your warehouse, your models, your BI connections, your SQL. | One store where the row a transaction wrote is the row the search ranks, eighteen languages written on the way in, documents whose tables become rows, and answers against your model rather than the platform's. | Cortex Search, Document AI, a translate call per row and an embedding job: four separately metered services and the code that orders them. |
| Databricks | Delta tables in your own object storage, notebooks, MLflow, the SQL warehouse. | Transactions and CRUD beside the analytics, a graph over the same records, and retrieval that does not wait for a sync to finish. | A vector index that syncs from a table, and the freshness question that comes with it: how far behind is the index right now? |
| Amazon Redshift | The cluster, the Glue catalogue, the IAM around it. | Search, vectors, graph, media, documents, translation and cited answers on one bill and one credential set. | OpenSearch, Neptune, Textract, Transcribe, Translate and a Bedrock knowledge base: six services, six bills, six sets of permissions. |
| Google BigQuery | On-demand SQL, and the indexes BigQuery already has for search, vectors and graph. | A price charged on the statements you ran rather than the terabytes a query happened to scan, and enrichment on the way in rather than a job afterwards. | Document AI, a Vertex connection for every translation and embedding, and Vertex AI Search as the product that answers. |
| Microsoft Fabric and Power BI | Your capacity, your reports, your semantic models, your seats. | A PostgreSQL connection Power BI opens with no adapter, over data that is already indexed, embedded, translated and answerable. | The pipeline between the source and the semantic model, and the capacity that has to be awake for it to run. |
| Elasticsearch | Your indices, your analyzers, your relevance tuning. | The transactions, the joins and the graph over the same records the index ranks - so the search reads what the write committed. | The database on the other side of the sync, and the connector that carries rows between the two. |
| MongoDB Atlas | Your documents, your drivers, your aggregation pipelines. | SQL joins and window functions, a real graph, media and document ingestion, eighteen languages and answers with citations. | Search nodes as a tier you size and pay for separately from the cluster. |
| Algolia | Your relevance rules, your synonyms, your front-end. | The database underneath the index, so a search reads the record the write committed rather than the copy that was pushed after it. | The second copy of every record, the sync on every write, and $0.40 per 1,000 records a month for holding it twice. |
| Pinecone | Your embedding model and the vectors you have already paid to compute. | The rows, the joins, the full text and the graph beside the vectors, and a filter that reads a real column rather than a metadata blob. | The second store, and the pipeline that embeds, upserts and reconciles into it. |
| turbopuffer | The cheapest vector storage you found, and you found correctly. | Everything a vector index cannot answer: transactions, joins, relationships, media, eighteen languages and a cited answer. | The code your team owns that keeps the index and the database in step, and the question of which of them is right. |
| Firestore | The document model, the client SDKs, the offline sync. | SQL joins, full text, vectors and a graph over the same documents, with residency enforced per write and a database placed per region. | The third-party search engine Google's own documentation tells you to put beside it. |
| Deepgram or another transcription API | Every transcript you have already produced. | The transcript landing already chunked, embedded, indexed, translated and answerable - at $0.24 an hour against $0.26. | The four jobs between we have the text and somebody can find it. |
| Tableau, Qlik, Looker or Excel | Every report and every seat. | One PostgreSQL connection to all of it - the rows, the full text, the vectors and the graph - with no adapter and no extract. | The ETL that fed the extract, and the second copy of the data it left behind. |
Not a competitor
Your reporting tools connect to it, unchanged
The PostgreSQL wire protocol is built into the engine, so Power BI, Tableau, Qlik, Looker, Excel, DataGrip, DBeaver and every ODBC or JDBC client connect with no adapter and no driver to install. The seats you already pay for keep working, against data that is already indexed, embedded, translated and answerable.
What you point at it
The same host, port, database and credentials you would give any PostgreSQL server. Your reports open, your extracts refresh, your analysts browse the catalogue. What they find on the other side is not a copy: it is the store itself, with the full-text and vector indexes over the same records.
Measured, tool by tool
Since September 2026 the PostgreSQL port is measured against the tools people actually own — DataGrip, DBeaver, Power BI, psycopg, Npgsql and the JDBC driver — each one connecting, browsing and querying, with a compatibility table per tool: connects, browses, queries, parameters, COPY, and what is refused and why. The roadmap has what comes next.
Your numbers, not ours
Send the invoices and we will do this table for your estate
Three months of bills from every vendor that holds, searches, translates or answers questions about your data. Back comes a fixed monthly price for the same workload on Data Oil, held for twelve months, with the line-by-line workings and every assumption named.
Sources
Where every figure on this page came from
+Every vendor price, its assumption, and the page it was read fromOpen the sources
| Vendor | Figures used | Read from |
|---|---|---|
| Snowflake | $2.00 / $3.00 / $4.00 a credit by edition; $23.00 a TB-month on demandAssumes an X-Small virtual warehouse, the smallest Snowflake sells, at one credit an hour on the Standard edition in AWS US East (Northern Virginia), and storage at the on-demand rate with no compression benefit claimed | the Snowflake Service Consumption Table effective 16 September 2026, tables 2(a) and 3(a): $2.00 / $3.00 / $4.00 a credit by edition, $23.00 per TB-month on demand |
| Databricks | $0.70 / $0.55 / $0.22 a DBU-hourAssumes a 2X-Small SQL Serverless warehouse, the smallest Databricks SQL offers, at four DBU an hour. Table storage stays in your own object storage and is not counted here at all, which makes this line cheaper than the bill would be | Microsoft's Azure retail price list for Azure Databricks, East US, read 2026-09-20: Premium Serverless SQL $0.70 a DBU-hour, SQL Compute Pro $0.55, SQL Analytics $0.22, Jobs $0.30, All-purpose $0.55. Databricks' own pricing page renders its rates in JavaScript and could not be read |
| Amazon Redshift | $0.375 an RPU-hour; $0.024 a GB-monthAssumes Redshift Serverless at the smallest base capacity the pricing page allows, four RPUs, in US East (Northern Virginia); managed storage on the same data | aws.amazon.com/redshift/pricing, read 2026-09-20: $0.375 per RPU-hour with a 60-second minimum charge, capacity adjustable from 4 to 1024 RPUs, Redshift Managed Storage $0.024 per GB-month |
| Google BigQuery | $6.25 a TiB scanned; $0.02 a GiB-monthAssumes active logical storage and on-demand analysis in the US multi-region, with the monthly free tier deducted - 10 GiB of storage and the first TiB scanned; batch loading is free, because nothing is done to the data on the way in. Google lists active logical storage hourly, and the hourly figure works out at $0.023 a GiB-month rather than the $0.02 the headline states | cloud.google.com/bigquery/pricing, read 2026-09-19: $6.25 per TiB on-demand after the first TiB, $0.02 per GiB-month active logical after the first 10 GiB |
| Microsoft Fabric | $0.18 a CU-hour; OneLake $0.026 a GB-month hotAssumes an F2 capacity - two capacity units, the smallest Fabric SKU - paid as you go and paused outside the hours the comparison assumes, with OneLake storage at the hot rate. A real half-terabyte workload wants a larger capacity; F2 is the published floor | Microsoft's Azure retail price list for Microsoft Fabric, East US, read 2026-09-20: $0.18 per capacity-unit hour pay-as-you-go, $0.54 for overage, OneLake storage $0.026, $0.019 and $0.004 per GB-month hot, cool and cold |
| Power BI | $14.00 and $24.00 a user a monthAssumes list price paid yearly, per user per month | microsoft.com/power-platform/products/power-bi/pricing, read 2026-09-20: Power BI Pro $14.00 and Premium Per User $24.00 a user a month, paid yearly |
| Tableau | $15.00 and $35.00 a user a monthAssumes Tableau Cloud list, billed annually, per user per month; Tableau Cloud+ is quoted on application and is not priced here | tableau.com/pricing, read 2026-09-24 in a browser: Tableau Standard $15 and Tableau Enterprise $35 a user a month billed annually, Tableau Cloud+ on application. The Creator / Explorer / Viewer roles this pack previously quoted are no longer published |
| Elastic Cloud | $0.14 and $0.09 a VCU-hour; $0.047 a GB-month retainedAssumes Elastic Cloud Serverless, the lowest rate each meter publishes | elastic.co/pricing/serverless-search, read 2026-09-20: ingest from $0.14 a VCU-hour, search from $0.09, machine learning from $0.07, storage and retention from $0.047 a GB-month, egress from $0.05 a GB |
| MongoDB Atlas | M10 $0.08 to M50 $2.00 an hour; search nodes from $0.24Assumes Atlas dedicated clusters in AWS US East, with Atlas Search on its own search nodes, which is how Atlas Search is sold beyond the shared tiers | mongodb.com/pricing, read 2026-09-20: M10 $0.08, M30 $0.54 and M50 $2.00 an hour; Search nodes S30 $0.24 and S50 $0.99 an hour |
| Algolia | $0.50 per 1,000 searches, $0.40 per 1,000 records a monthAssumes the free 10,000 searches and 100,000 records a month deducted; one index, and no replicas for a second ranking or a second region | algolia.com/pricing, Grow plan, read 14 September 2026 |
| Pinecone | $4.00–$4.50 per million write units, $16.00–$18.00 per million read units, $0.33 a GB-monthAssumes the Standard plan's serverless rates, the low end of each range, which is the cheapest cloud and region; one write unit to a record of a kilobyte or less, which is what a row of a tabular ingest is. Corrected: was carried at $2.00 per million write units when D53 was decided - half the real figure. D53's direction held and its landing point did not; see D55 | pinecone.io/pricing, Standard, read 2026-09-19: write units $4-$4.50 per million, read units $16-$18 per million, storage $0.33/GB/mo, all varying by cloud and region |
| turbopuffer | $0.33 a GB-month, $2.00 a GB written, $1.00 a PB queriedAssumes the whole collection written once a month, every query charged at the 1.28 GB minimum, and the Launch minimum applied where usage falls under it - which it does at all three of these volumes | turbopuffer.com/pricing and docs/pricing-log; unit figures as quoted by usagepricing.com, checked 2026-08-28 |
| Firestore | $0.03 per 100k reads, $0.09 per 100k writes, $0.18 a GiB-monthAssumes the daily free quota applied at 30 days to the month, and no full-text search in the product at all; Google's own guidance is a third-party engine beside it | cloud.google.com/firestore/pricing (us-central1) via search; firebase.google.com/pricing for the free quota |
| Deepgram and the transcription APIs | $0.0043 a minute batch on Deepgram Nova, $0.26 an hour; Amazon Transcribe $1.44Assumes batch transcription of one channel, with no diarisation, redaction or custom vocabulary; streaming is dearer at every vendor | deepgram.com/pricing and aws.amazon.com/transcribe/pricing, with the per-hour conversions as compared by deepgram.com/learn/best-speech-to-text-apis-2026 |
| Data Oil | the deployment's own published rate card, 115 meters, USD, read 4 October 2026 | GET /api/v2/pricing on the reference deployment |
One of these cannot be read by machine and says so rather than looking checked: Databricks renders its own rates in JavaScript, so the DBU figures come from Microsoft's Azure retail price list, which publishes the same SKUs. Two were read again and both changed something. Pinecone's write units had been carried at half the real figure. And Tableau, which refuses automated reads, was opened by hand in September 2026 — its per-role Creator, Explorer and Viewer prices are no longer published at all, and the editions above are what the vendor lists now. A figure nobody can re-read is a figure that is only ever right by accident.