Why companies switch data warehouses

Companies leave older platforms such as Teradata one way, and move between Snowflake, BigQuery, Databricks and Redshift in every direction. Cost is named on both sides of the same move, the company's cloud often decides, and a larger share of the companies that mention Redshift describe leaving it. The pressures, the events, where companies go, what they gave up, how long it ran, and the moves dropped.

Based on verified interviews with the data, analytics, marketing and ecommerce leaders who use and choose data warehouses, at DTC and enterprise brands. Buyers are anonymized before publication; vendor names and views are reported as given. No vendor paid to appear or could edit this page.

Companies switch data warehouses for the cost of the platform they are leaving, because they have outgrown it, or because a decision about the company's cloud, an acquisition or a consolidation settles where the data goes — and few buyers say the old warehouse simply failed. Among companies describing a warehouse migration completed or under way — not one planned, evaluated or dropped — about two in five are leaving an older or in-house environment: Teradata, a Hadoop-era system, SAP's reporting layer, or an on-premises or homegrown warehouse. Every one of those moves leads to the cloud. About half run between cloud platforms, in more than one direction: onto Snowflake and off it, off Redshift and BigQuery, into Databricks. Cost is named on all sides, so it explains a move only against the specific bill being left. Most companies that talk about their warehouse describe no move at all, and these shares describe the moves in these interviews, not any platform's customer migration rate or market share. The companion reads are what data warehouse buyers wish they'd known, where the ratings live, Snowflake vs. BigQuery and Databricks vs. Snowflake.

2 in 5
Completed or under-way moves here that leave an older platform — Teradata, Hadoop, SAP or a homegrown system. About half run between cloud platforms.
1 in 9
Companies mentioning Redshift that describe leaving it, done or under way — against about 1 in 50 for Snowflake or BigQuery. Interview shares, not churn rates.
8.3/10
Snowflake's buyer rating — the most common destination here, and also left by several companies for another cloud platform.
Where data warehouse moves go, by the kind of platform left Moves off an older platform such as Teradata, Hadoop, SAP or a homegrown system went mostly to Google's cloud and to Snowflake, with one each to Databricks and Redshift. Moves between cloud platforms went to Snowflake most often, then to Google and Databricks in equal numbers, and a few to Azure or AWS. A small group leaving an in-house cloud build, or not saying what they left, went to Snowflake, Databricks or an internal platform. LEAVING WHERE THE MOVE WENT An older platform Teradata, Hadoop, SAP, homegrown one-way Another cloud platform Redshift, Snowflake, BigQuery, Synapse every direction In-house build, or unstated origin not named Snowflake Google Cloud, BigQuery Databricks Other: Azure, AWS, Redshift, internal OLDER PLATFORMS ARE LEFT ONE WAY; CLOUD PLATFORMS ARE LEFT IN EVERY DIRECTION
Counted moves are migrations completed or under way, one per company by its latest interview: a primary warehouse replaced, or two consolidated into one with the other retired, which counts as a move to the one kept. The platform left may still hold workloads mid-migration, and the destination is the environment kept or adopted, not necessarily where every workload went. Plans, evaluations, first warehouses, a second platform added with nothing retired, and changes to a CDP or BI tool alone are excluded. Each bar's length shows how many companies in these interviews left that kind of platform, and its segments where they went; all bars share one scale. These are observed decisions, not churn rates or market share.

The pressure that builds: why companies switch data warehouses

The reasons buyers give fall into three kinds, and a move often carries more than one: the bill being left, a platform the company has outgrown, and the reach and governance the team wants. These are the pressures; the events that settle the decision follow. Several buyers say the decision was made above them, which is the starting point of what data warehouse buyers wish they'd known.

Slow-burn · #1

The bill being left — in every direction

Cost drives these moves in different directions. Companies moving off Teradata and Azure Synapse to Snowflake name its price; a travel company moving off Synapse cites “the cost efficiency we are seeing with Snowflake.” Another company is moving from BigQuery to Snowflake with cost saving as the stated aim, a restaurant chain is leaving Snowflake for Databricks after its Snowflake costs kept rising, and a specialty retailer chose Redshift over Snowflake when it moved to the cloud because, on its own tests, “Redshift was the most optimal one for us” on performance and cost. These are different workloads, contracts and stages. A saving expected is not a saving realised, and even a reported one is one company's experience, not a price ranking; the comparison that counts is the total cost of the same workload before and after. Each vendor meters compute in its own unit; what each one counts is on what data warehouse buyers wish they'd known.

Named in this context: Snowflake · Teradata · Azure Synapse · BigQuery · Databricks · Redshift

Slow-burn · #2

The old platform was outgrown

Behind several moves off an older platform. One buyer says of the move off Teradata, “we simply outgrew what Teradata could provide.” A specialty retailer left “an old Hadoop environment, which was very challenging to work with and very slow.” A publisher leaving a homegrown system for BigQuery says bringing on a new tool used to take years: “It would be a two-year process to bring on a new vendor.” Teradata now sells its own cloud deployment, which it says runs on AWS, Microsoft Azure and Google Cloud as a managed system (teradata.com, checked November 2026), so these accounts describe the setups these companies ran, not what Teradata offers today.

Named in this context: Teradata · Hadoop · Snowflake · BigQuery · Redshift

Slow-burn · #3

Reach, sharing and governance

Some moves between cloud platforms are about what the team can do with the data. An apparel retailer is leaving Redshift for Databricks partly because Redshift “doesn't have data governance capabilities baked in.” That is the buyer's assessment of its own environment and needs — the same buyer calls AWS “sort of like a marketplace” where “you can find everything you need,” against Databricks as one unified platform — not a finding that Redshift has no governance tools: AWS documents role-based access control, row- and column-level security, dynamic data masking and Lake Formation-managed permissions for Redshift (docs.aws.amazon.com, checked November 2026). Our suggestion, not a practice buyers describe: if governance is the reason to move, test the specific workflow on both platforms — who grants access, to which rows and columns, enforced where. A hospitality company that left Teradata for BigQuery says the new platform has “integrated a lot more data, so it gives us a lot of flexibility.” A grocery retailer moving from SAP's reporting layer to Snowflake names data protection and integrations alongside “the competitive pricing that Snowflake has to offer.”

Named in this context: Databricks · Redshift · BigQuery · Snowflake · SAP

That's the average buyer. How does your stack measure up?A free, peer-powered read of your tech stack, with a shortlist and your next steps.
Get my stack read →

The event that fires the decision

A warehouse move rarely starts in the team that uses the warehouse. In the moves buyers explain, an event outside the product often settles the timing: a cloud decision, a consolidation, a change of owner or a renewal. The interviews show these events beside the pressures above; they do not always show which one decided.

Trigger · #1

The company's cloud decides

An event that appears in moves of every kind: a decision about the company's cloud can carry the warehouse with it. In the completed and under-way moves, companies standardising on Google's cloud favour its analytics: an apparel company that sends “all the data” to “both Redshift and GCP” is retiring Redshift because Google Cloud “is going to be our final solution.” A large enterprise leaving Teradata for Google Cloud says Google's own tools “make natural connectivities to themselves.” Two evaluations show the same pull, though neither is counted as a switch: a manufacturer chose Databricks over Snowflake for a new data lake it is now piloting, “because we already use Azure as a footprint,” and a consumer app company stayed on Google after testing both.

Named in this context: Google Cloud · BigQuery · Redshift · Teradata · Databricks · Snowflake · Azure

Trigger · #2

Two warehouses into one

Several companies ran two cloud warehouses and kept one. A software company that ran Snowflake and Databricks together says “we've sunset Snowflake” and kept Databricks, though “each tool has their own pros and cons.” A media company that ran BigQuery beside Databricks kept Databricks. An apparel retailer retired both Snowflake and a Google warehouse during a wider systems change. Each counts as a move off the warehouse retired, even where the one kept was already in use — a different thing from adopting a new provider — and keeping one platform need not mean every table and job moved onto it.

Named in this context: Databricks · Snowflake · BigQuery

Trigger · #3

A new owner, a merger or an ERP move

An ownership change is moving one company off BigQuery; a merger left another running a second warehouse it is now deciding whether to keep; a distributor plans to retire a Teradata setup that came with an acquisition. A grocery retailer's move to Snowflake followed the company's move to a new SAP ERP, which it saw as the moment to put analytics on its own platform.

Named in this context: Snowflake · BigQuery · Teradata · SAP

Trigger · #4

The renewal

Only one company here names the renewal as the moment, and it is the clearest case of the platform not being the problem. A large healthcare company leaving Teradata for BigQuery rates the old technology well and blames how it was bought and rolled out — the story is below. Low ratings that are really about the build are the subject of lesson 04 on what data warehouse buyers wish they'd known.

Named in this context: Teradata · BigQuery

Where do companies switch to?

Among the completed and under-way moves here, counting each company once, Snowflake is the most common destination, in about two in five moves, with Google's cloud and BigQuery close behind at about three in ten and Databricks after them. Moves off older platforms split between Google and Snowflake: the Teradata exits here went to both. Moves between cloud platforms run in every direction, and Snowflake sits on both sides of them — several companies moved onto it from Redshift, BigQuery, Synapse or a lakehouse on AWS, and several left it for BigQuery, Databricks, Azure or AWS. Within these moves, Snowflake's net gain rests substantially on companies leaving older environments, since several also moved off it. Relative to the companies that mention each platform, a higher share describe a completed or under-way departure from Redshift: about one in nine, against about one in fifty for Snowflake and for BigQuery. Only one company here moved onto Redshift, years ago, from Hadoop. The groups are companies that named each product in an interview, not verified customers — some mention a product they used before or only evaluated — so these are not churn rates or a current customer's odds of leaving. The platforms side by side are on Snowflake vs. BigQuery and Databricks vs. Snowflake, where the moves described shift workloads more often than whole platforms.

What do companies give up when they switch warehouses?

Buyers who moved describe trades, not only gains. A consumer subscription company that moved onto Snowflake from Cloudera's platform misses the looser SQL it was used to — “Snowflake is so rigid and structured that from a coding standpoint, it's really annoying” — and disputes the speed claim, “it's not faster,” while granting that a data engineer would prefer the rigidity. A media company that kept Databricks over BigQuery finds Databricks “a bit slower than BigQuery was” on complex queries. A specialty retailer that chose Redshift over Snowflake gave up the cross-company data sharing it now wants: “that capability does not exist in Redshift.” That is the retailer's experience and its requirement at the time — it calls both of its Redshift pain points “more configuration” — not a universal limit: AWS documents live data sharing between Redshift warehouses across AWS accounts and Regions, and Lake Formation-managed sharing within or across accounts, though not across Regions (docs.aws.amazon.com, checked November 2026). Our recommendation: test the exact arrangement — which accounts receive the data, with what permissions, in which regions and formats, and whether the partner sharing it runs on the same platform. What companies keep matters as much. A large healthcare company leaving Teradata is keeping its table structures and field names, because changing them would mean rework for every team that reads the data. These are individual accounts, and they describe each company's setup at the time.

How long does a switch take?

Few buyers here give a start and an end. One large retailer describes its move off Snowflake: “it began, like early 2024 and ended around November.” More often the two platforms run together. An apparel company sends data to both Redshift and Google Cloud and cannot date the retirement: “I don't know what that soon looks like.” A large enterprise leaving Teradata is “right smack dab in the middle of the migration. So it's a little 50-50 right now.” Others move workloads a campaign or a table at a time, and some keep a little of the old platform after the rest has gone. A consumer products company shows how long the decision itself can take: interviewed in September 2025, it called Redshift “good enough for our size and our needs”; the next August it was “trying to get sign-off to migrate to Snowflake” — an interval between two interviews, not a measured duration. Go-live is not the end either: a company can be sending data to the new platform while still running, and paying for, the old one, and being ready for production, moving the workloads and retiring the old platform each take their own time. The interviews do not establish a typical duration for any of them. How a migration in progress holds up other purchases is lesson 06 on what data warehouse buyers wish they'd known.

Want your own stack read the same way?

Get mine →

The stories behind the switches

A large healthcare company is leaving Teradata for BigQuery, and its data strategy leader does not blame the product: “Teradata itself was like an eight.” The problem was the purchase: “It wasn't the software; it was that we had one of their data models as part of what we bought.” A renewal prompted an analysis, and the company decided the data model “wasn’t worth the cost for the value we were getting,” building its own instead. The move itself copies the old structure: “From day one, we're doing much more of a lift and shift,” keeping field names because otherwise “we would have had a tremendous amount of editing on consumption.” Executive director, data strategy · large healthcare company
A consumer subscription company had just moved onto Snowflake from Cloudera's platform — “We still have some things on CDP Impala” — and its growth analytics lead preferred the old one: “CDP Impala was a lot nicer in my view.” The old engine forgave sloppy SQL; on Snowflake, “it's not faster,” and “Snowflake is so rigid and structured that from a coding standpoint, it's really annoying.” The same buyer grants the other side: a data engineer would prefer Snowflake “because that rigidity enforces better data structures.” Associate director, growth marketing analytics · consumer subscription company
A consumer app company tested Databricks and Snowflake and stayed with the cloud it had. “We evaluated them for about six to eight weeks,” its engineering leader says, and the reasons not to move were total cost of ownership, a steeper learning curve, extra tooling and “the number of vendors we had to onboard from a security standpoint.” On cost: “these platforms start off at a smaller rate, but as you integrate more aspects of your data into it, the cost adds up quickly.” It built a lakehouse on Google instead. Vice president, engineering · consumer app company

The counter-current: the moves considered and dropped

Most companies that talk about their warehouse describe no move, and several that weighed one stopped. A real-estate company looked at Microsoft Fabric for a lakehouse and set it aside: “We've now shifted our focus to actually looking for a new CMS instead.” A food retailer paused its move from an on-premises warehouse to Azure while it launched a new website and app. A healthcare brand explored Databricks and did not go ahead, because “our data is not currently in Databricks.” A retailer will not follow its sister brands from Snowflake to Databricks, because “we've so recently transitioned onto Snowflake ourself.” Others remain open: a financial services company is weighing a move off Redshift but has “not yet made that decision,” an online retailer interviewed this month is evaluating Snowflake and Databricks, and an apparel retailer plans to move from BigQuery to Fabric because its IT team already runs Fabric. Considered, paused and planned moves are not counted as switches here.

What this means if you're weighing a switch

Buyers report that warehouse moves follow the bill being left, a platform outgrown, and decisions about the company's cloud, owner or consolidation, and that the old and new platforms often run together for a long time. Our recommendations, not practices buyers describe. Price the move against your actual bill and workload rather than a list price: run representative workloads — large transformations, interactive queries, scheduled reports, peak demand — through each platform, and compare compute, storage, data movement, concurrency, workload management and commitments in each vendor's own units. Make two estimates, the ongoing cost of running and the one-time cost of migrating: moving the data, converting SQL and pipelines, testing, engineering time, training, partners, rebuilding security, and the months of paying for both. Estimate the payback period and the assumptions that move it, then check projected savings against actual consumption after each phase.

Before blaming the platform, separate its limits from the build. A data model, inefficient transformations, unclear ownership, duplicated data and reports wired to the old structures move with you unless you change them deliberately, as one healthcare company found with Teradata. Then decide what must stay compatible — table structures, field names, key SQL, permissions, pipelines, scheduled jobs and the reports downstream. A lift and shift limits disruption but does not guarantee identical behaviour or performance; a rebuild can improve the design but costs the people who read the data more work. Test representative production workloads first: query results, data types, missing and duplicate records, update timing, access controls, concurrency and latency, and the SQL, orchestration changes and skills each platform needs. If the company's cloud is likely to decide, ask the owners of that decision before the shortlist forms.

Finally, define how the move ends before it starts: the workloads that go first, who validates them, and the conditions for retiring the old platform. While both run, reconcile key datasets and metrics, confirm that jobs and downstream applications still work, test permissions and recovery, and give each workload a cutover plan with a way back. Have data engineering, security and the teams that rely on the reports sign off the retirement, rather than taking data arriving on the new platform as proof. Then date the end: when the old platform stops receiving data, when its dependencies are removed, how historical data stays reachable, and when the contract ends. Buyers here who could not date the end were still paying for both.

Common questions

Why do companies switch data warehouses?

In these interviews companies switch for three kinds of reason, often together: the cost of the platform they are leaving, a platform they have outgrown — slow, hard to share, short on governance or hard to reach — and a decision about the company's cloud, an acquisition or a consolidation that settles where the data goes. Cost is named on every side: as a reason to move onto Snowflake, to leave it, and to stay put. Few buyers say the old warehouse simply failed.

Which data warehouse do companies switch to?

Among completed and under-way moves in these interviews, counting each company once, Snowflake is the most common destination, in about two in five moves, with Google's cloud and BigQuery close behind and Databricks after them. Moves off older platforms such as Teradata split between Snowflake and Google. Moves between cloud platforms run in every direction, and Snowflake appears on both sides of them. These are observed decisions, not market share.

Why do companies leave Amazon Redshift?

About one in nine companies that mention Redshift in these interviews describe a move off it completed or under way, against about one in fifty for Snowflake and for BigQuery — a comparison across companies that named each platform, some of which used it before or only evaluated it, not a churn rate. The reasons given by companies leaving or weighing a move off it are processing power, governance built into a single platform, easier reporting for end users, the maintenance the warehouse needs, and a company move to Google's cloud. The destinations are Snowflake, Databricks and BigQuery.

How long does a data warehouse migration take?

Few buyers in these interviews give both a start and an end. One large retailer describes a move that began early in the year and ended around November. More often the old and new warehouses run side by side for months or years, with data sent to both, workloads moved one at a time, and a retirement date the buyer cannot name. Treat a planned date as a plan until the old platform is switched off.

Why do companies decide not to switch data warehouses?

In these interviews companies dropped or paused moves because the total cost grew as more data and tools were added, because the team would need new skills and extra vendors, because another project such as a new CMS or website took priority, or because they had only just finished a migration. Some chose the platform they were already running after testing the alternatives.

Your stack, read by the buyers who've run it

Peer-powered: built from hundreds of verified buyer interviews, and fitted to the tools you already run. You get a shortlist for your stack and your next steps.

Get my stack read

Free · no password · your answers are anonymized before they inform research like this.

Want to come back to this?We'll email you a link to this page, so it's easy to find later or forward to your team.
Methodology. Alium conducts verified interviews with software buyers — the data, analytics, marketing and ecommerce leaders who use and select these platforms. This page aggregates what buyers in that corpus, through November 2026, say about replacing, consolidating or keeping a data warehouse. A move counts when the company's main warehouse or lakehouse changes, including two warehouses consolidated into one and a move off an on-premises, Hadoop-based or homegrown system; moves between a warehouse and a CDP, BI tools, first warehouses, moves a buyer describes at a previous employer and moves by a company onto its own products are excluded. Moves are confirmed in the buyer's own words, counted once per company by its latest interview, and kept apart by status: completed, under way, planned or under evaluation, and considered and dropped. Departure ratios compare companies describing a completed or under-way move off a platform with the companies that mention that platform in interviews. Reasons are editorial codings of what buyers said, and a move can carry several. Ratings are buyer-satisfaction averages from published transcripts; the full table is on the data warehouse lessons page. Buyer identities are verified at interview time and anonymized before publication; vendor names are reported as given. No vendor paid to appear or was able to edit this page.