Databricks Reviews & Product Details

What I like best about Databricks is how it brings data engineering, analytics, and machine learning into one unified workspace. I find collaboration much easier with shared notebooks, and the seamless integration with big data tools saves me time. It simplifies complex workflows while still offering powerful capabilities when I need them. Review collected by and hosted on G2.com.

One thing I dislike about Databricks is that it can feel expensive, especially for smaller projects or teams. I also find cluster configuration and cost management a bit complex at times. The interface, while powerful, can be overwhelming for beginners, and debugging distributed jobs isn’t always as straightforward as I’d like. Review collected by and hosted on G2.com.

The thing I genuinely appreciate most about Databricks is how it brings everything under one roof. Before Databricks, I was juggling separate tools for data ingestion, transformation, analytics, and ML, and honestly, it was a nightmare keeping them all in sync. With the Lakehouse architecture, I can store my raw data, run heavy Spark transformations, build dashboards, and even train ML models all in the same platform. Unity Catalog on top of that gives me a single place to manage permissions, lineage, and data discovery, which used to take me days of stitching together custom scripts. It just removes so much friction from the day-to-day.

From a workflow standpoint, the notebook experience with real-time collaboration has been a game changer for my team. We used to pass around Python scripts over Slack and hope nobody overwrote each other's work now we just co-edit in the same notebook and see results instantly. The Spark Declarative Pipelines (what used to be called Delta Live Tables) let me define my entire ETL as simple SQL or Python declarations, and the platform handles retries, data quality checks, and lineage tracking automatically. I also didn't expect the Jobs orchestrator to be as flexible as it is I can wire up multi-task DAGs with dependencies, set schedules, and get alerts without needing Airflow or any external scheduler. That alone saved us weeks of infrastructure setup.

The AI and intelligence side of things has honestly surprised me the most. The AI/BI dashboards let non technical folks on my team ask questions in plain English and get actual visualizations back, which means fewer "hey can you pull this data for me" requests in my inbox. Model Serving makes deploying an ML model or an AI agent to a production endpoint almost trivially easy compared to rolling your own Flask app on Kubernetes. And the integrations are solid whether it's connecting to external tools through the Databricks SDK, using the REST API, or plugging into AI coding assistants through things like MCP servers, it plays nicely with whatever stack you're already running. The ROI has been clear for us: less time on plumbing, more time on actual data work. Review collected by and hosted on G2.com.

The pricing model is probably my biggest gripe with Databricks. It's based on DBUs (Databricks Units), and honestly, it can get confusing fast trying to figure out what's going to cost you what. You spin up a cluster for some quick testing, forget to shut it down over the weekend, and suddenly your bill looks like you were running a production workload for a Fortune 500 company. Even with autoscaling and serverless options, the costs can creep up in ways that are hard to predict, especially for smaller teams or startups that don't have a dedicated FinOps person watching the meter. I wish the pricing was more transparent and easier to estimate upfront without needing a spreadsheet and a prayer.

The learning curve is steeper than people let on. If you're coming from a traditional SQL background or you've only worked with simpler tools like basic ETL platforms, Databricks can feel overwhelming at first. There's a lot of concepts to wrap your head around workspaces, clusters, catalogs, schemas, warehouses, notebooks, jobs, pipelines and the documentation, while extensive, sometimes assumes you already know what you're looking for. Onboarding new team members takes longer than I'd like because there's no single guided path that says "start here, then go there." You kind of have to piece it together yourself or lean on someone who's already been through the pain.

Performance-wise, it's generally solid, but there are moments that test your patience. Cold starts on clusters can take a few minutes, which breaks your flow when you just want to run a quick query or test a small change. Serverless compute has improved this a lot, but it's not available for every workload type yet, and sometimes the serverless option has its own quirks with library compatibility. The notebook UI, while functional, can also feel sluggish when you're working with large outputs or long notebooks it's not as snappy as a local IDE. These aren't dealbreakers by any means, but they're the kind of paper cuts that add up over a long day of development. Review collected by and hosted on G2.com.

What I like best about Databricks is how it brings data engineering, analytics, and machine learning together in one clean workspace. It saves time, makes collaboration easier, and helps teams move faster with large data. Review collected by and hosted on G2.com.

What I dislike about Databricks is that Auto Loader can become frustrating when source data changes frequently, especially if column names or datatypes shift without warning.

For example, a field like customer_id may suddenly come in as cust_id, or a column that was previously a string may start arriving as an integer, which can cause schema drift and break downstream processing.

I also find it inconvenient when schema inference is not fully accurate, such as when nested JSON or semi-structured data is read incorrectly, because it then requires extra manual fixes and maintenance to keep pipelines running smoothly. Review collected by and hosted on G2.com.

I like that Databricks brings everything into one place, making it unnecessary to use different tools for data processing, analytics, and pipeline work. It handles large data well, and we don't have to worry about managing clusters manually. Additionally, Databricks handles collaboration and experimentation well, making it easy to try out new things. Review collected by and hosted on G2.com.

In my point of view, the one area that can be improved is cost management. If clusters aren't monitored carefully, costs can increase faster than expected. One improvement that would help is better visibility into costs at a more detailed level. More built-in alerts or recommendations when costs start increasing unexpectedly would also be helpful. Review collected by and hosted on G2.com.

The unified platform approach is what I appreciate most. Having notebooks, data engineering pipelines, ML workflows, and SQL analytics all in one place saves a ton of time instead of juggling multiple tools. The collaborative notebooks make it easy to share work with teammates, and the cluster management has gotten a lot smoother over time. Delta Lake integration is also a huge plus for keeping our data reliable and consistent. Review collected by and hosted on G2.com.

The cost can get out of hand pretty quickly if you're not careful with cluster sizing and uptime. It's not always obvious how to optimize spending, and the pricing model feels complex. The learning curve for new team members is also steeper than I'd like, especially for people who aren't already familiar with Spark. Sometimes the UI can feel sluggish when working with larger notebooks, and debugging job failures could be more straightforward. Review collected by and hosted on G2.com.

I like that Databricks is an all-in-one powerhouse where I can do multiple works in one place. It's powerful to manage data from multiple sources and have it in a single UC to manage permissions with row-level security. I also appreciate that I can create experiments, run multiple models, and select the best one from logs, which was difficult on other platforms. Once I learned the setup, it's been easy and comfy to work with. Review collected by and hosted on G2.com.

I find it difficult to use the calculator to determine CPU serving endpoint prices because the documentation doesn't explicitly explain this. It only mentions 1 concurrency equals 1 DBU on the Azure page, which isn't clear. The pricing calculator has a single option for serving endpoints, labeled as medium with four DBU, but lacks separate options for GPU or CPU and their concurrency, making it hard to understand how it works properly. Initially, I also felt it was very tough to learn Databricks and manage deployments of workspaces, although it became easier over time. Review collected by and hosted on G2.com.

I really value how the platform brings data lakes and warehouses together into one place. It makes managing data much easier, and the SQL performance is very fast thanks to the Photon engine. I also like the collaborative notebooks because they allow me to work with both SQL and Python seamlessly in a single environment. Review collected by and hosted on G2.com.

The cost can be high, and the DBU billing system is quite complex to track. I also found that there is a significant learning curve when it comes to Spark and configuring clusters. For smaller, quick tasks, the setup time and technical overhead can sometimes feel like a bit too much. Review collected by and hosted on G2.com.

What I appreciate most about Databricks is its ability to unify data engineering, data science, and analytics on a single platform. The collaborative environment—especially the notebooks and integrated workflows—makes it much easier for teams with different skill levels to work together without constant context-switching.

Another highlight is the integration with popular tools and cloud services that are widely used in the market today, which makes it easier to move data between them. The performance monitoring and job scheduling features help maintain visibility over pipelines, and the Delta Lake support for reliable data management has also been very useful. Review collected by and hosted on G2.com.

Cost management is one area that could be improved. While Databricks offers autoscaling and flexible cluster options, it’s easy for resource usage to escalate unexpectedly, especially with large datasets and long-running jobs. Keeping costs predictable often requires careful oversight and a solid understanding of the platform’s pricing model.

Additionally, some of the more advanced features—such as fine-grained access controls and more complex job orchestration—can feel less intuitive. The documentation is extensive, but it occasionally leaves gaps that end up requiring trial and error. Review collected by and hosted on G2.com.

We use Databricks daily as our core data platform for building and running pipelines across a medallion architecture, from extracting data out of SAP and Arkieva all the way to reporting-ready datasets. The notebook experience is intuitive, the feature set is massive, and Asset Bundles have made our CI/CD story with Azure DevOps really solid. Integration with cloud services was smooth, and once things are set up they just work. The learning curve can be steep for newer team members, especially around things like Unity Catalog and DABs, and costs can creep up if you're not staying on top of cluster configurations. Support is decent and the docs are strong enough that we rarely need to open a ticket. Overall, it's a powerful platform that does a lot under one roof, and it's hard to imagine our data engineering workflow without it. Review collected by and hosted on G2.com.

The cost can creep up fast if you're not careful with cluster sizing and job configurations, so it takes some effort to keep things optimized. Also, the learning curve for newer team members can be steep, especially around things like Asset Bundles, Unity Catalog, and getting the CI/CD pieces wired up properly. Review collected by and hosted on G2.com.

What I like most about Databricks is how it streamlines the entire data workflow by bringing processing, analysis, and machine learning into one platform. The collaborative notebook environment makes it easy to share code, context, and reasoning with teammates, which helps everyone stay aligned. It also performs strongly on large datasets while abstracting away most of the cluster management, so I can focus on solving the problem rather than dealing with infrastructure. On top of that, centralized access control and clear visibility into data usage support responsible data governance, offering a solid balance between power and ease of use. Review collected by and hosted on G2.com.

Databricks has a few downsides, although many of them feel more like trade-offs than outright negatives. My biggest concern is cost: if clusters aren’t managed carefully, expenses can climb quickly, even though the platform can scale very efficiently when it’s tuned properly. There’s also a real learning curve with Spark and distributed computing concepts, and debugging or performance tuning can be more involved than with simpler tools. Lastly, because it’s a managed service, you give up some low-level control compared with self-hosted systems, but the upside is that it takes a lot of the operational and infrastructure work off your plate. Review collected by and hosted on G2.com.