CIOs are (still) closer than ever to their dream data lakehouse

The data lakehouse battle is over. And open-source Apache Iceberg has won. Not even Databricks’ 10-figure acquisition of Tabular, the startup founded by Iceberg’s creators, will change that.

Even so, the bold move has confused and distracted some CIOs. They’re at a loss to explain why Databricks – a lakehouse pioneer and the architect of Delta Lake, Apache Iceberg’s primary competitor – would spend so much to buy three-year-old Tabular, a startup with great promise but barely $1 million in annual revenue.

Some speculate that Databricks wanted to slow the cruising Iceberg ecosystem with a dose of uncertainty. Others wonder whether the company plans to pile Delta Lake projects on the Tabular crew, which continues to play an integral role in steering and developing Iceberg. That would help its own platform, the theory goes, while simultaneously sapping resources from the alternative lakehouse table format.

Another hypothesis: Databricks execs were billion-dollar stoked to stick it to Snowflake by drowning out its event with a buyout its rival reportedly sought. Or maybe they just wanted a quick way to set themselves apart in the active Iceberg space in hopes of soothing Wall Street jitters ahead of its perennially imminent IPO.

Whatever the reason, Databricks is saying all the right things about the openness and portability the acquisition will bring – albeit in terms just vague enough to keep the speculation alive.

“I do think the acquisition has been a bit of a distraction, but that’s probably true anytime that kind of money starts moving around,” David Nalley, director of open-source strategy and marketing at Amazon Web Services, told me. AWS, which has integrated Iceberg into analytics services like AWS Glue and Amazon Athena, has been actively involved in Iceberg’s development for the past three years. “That said, all the signals I’ve seen is that

more people are getting involved. The velocity is actually increasing. And we’re excited about that.”

Indeed, for all the handwringing, much of the work now being done isn’t even on the Iceberg table format, which insiders say is relatively stable. And now that it’s established as the default table format, the REST catalog layer above – that is, the APIs that help define just how far and wide Iceberg can stretch, and what management capabilities data professionals will have – is becoming the new battleground.

It’s also where Databricks can still make an impact by marrying data under its purview with information stored on competitive platforms. In fact, it’s already working toward that. In June, the week after Databricks bought Tabular, it made the Databricks Unity Catalog, its own governance tool, open source.

“The data catalog is critical because it’s where business manages its metadata,” said Venkat Rajaji, Senior Vice President of Product Management at Cloudera. Cloudera also has been investing in both Iceberg and REST catalog capability in its platform. “There’s been a ton of innovation lately around the Iceberg REST catalog because the data turf war is over. But the metadata turf war is just getting started.”

Lakehouse appeal

The pitch for data lakehouse table formats sounds almost too good to be true. The formats are basically abstraction layers that give business analysts and data scientists the ability to mix and match whatever data stores they need, wherever they may lie, with whatever processing engine they choose.

There’s a record of everything – including metadata changes – which paves the way for a host of management and governance capabilities. The data itself remains intact, uncopied and unaltered. So any number of projects can tap into the data at once. And the table formats will keep track of all of it.

CIOs give the thumbs up because the formats do away with needless data copies for individual projects that compound storage fees and swell security, reliability, and manageability headaches. And in theory, at least, it all happens without vendor lock-in.

That last part – the lack of Hotel California-style gotchas like proprietary enhancements and steep egress fees that conspire to pen enterprises into proprietary data warehouses – played a key role in shaping Iceberg by its creators, who worked for Netflix at the time. The

vendor-agnostic approach is also what helped draw large datacentric companies like Apple, Citibank, and Pinterest to the project. And it continues to fuel its rising popularity.

Delta Lake is technically open as well. Databricks donated Delta Lake to the Linux Foundation at about the same time that Netflix handed over the Iceberg project to the Apache Software Foundation. But some CIOs worry that Databricks’ outsized influence in the platform affords the company opportunity to maintain and augment proprietary hooks, like in Databricks Runtime.

“There’s definitely a feeling out there that Delta Lake is the brainchild of one company,” said Russell Spitzer, Principal Engineer at Snowflake. Spitzer, who in June joined Snowflake from Apple, is on the Iceberg project management committee (PMC) as well as the podling (incubating) PMC for Apache Polaris, a REST-compatible API that Snowflake donated to Apache in June. He also contributes code to both.

“You know, it’s open source,” Spitzer said, “But it’s really a Databricks product.”

If you can’t beat ‘em

The first wave of Iceberg adoption kicked into high gear around 2020, when it first became a top-level Apache project. In addition to AWS, more open-centric vendors like Cloudera and Dremio began building services around Iceberg. Google hopped in toward the end of the wave.

Most proprietary data platform providers sat on the sidelines during the initial wave, likely because Iceberg’s any-data-any-engine construct posed a threat to their existing business models. Snowflake was a notable exception. The data platform provider started investing in Iceberg during this period, likely because it needed a counter to Delta Lake, the lakehouse format from its most formidable competitor.

But as it became apparent that enterprises were going to combine data from competitive warehouses with Iceberg, proprietary platform providers began adding support in earnest. That put them in a better position to keep data under management – and possibly to host processing as well.

Just in the past year, Confluent, Oracle, and Salesforce all added support for Iceberg. Snowflake doubled down on Iceberg with Polaris. Microsoft, the last cloud service provider holdout – likely due to its investment in Delta Lake – joined Snowflake’s coming out party in June. And, of course, Databricks has been expanding coverage rapidly as well.

“It’s just amazing to me how far Iceberg has come,” said Snowflake’s Spitzer. “I used to have to explain why people should care about (Iceberg). And now, everyone knows. And everyone knows that everyone’s moving towards it.”

It’s all about the metadata

Iceberg creates a great foundation for combining and working with different data stores for projects. And now that the enterprise data analytics community is basically bought in, the next stage of work is happening at the catalog layer. And where AWS, Cloudera, Databricks, Snowflake, and others are all working to help Iceberg work as well as possible with as much data as possible.

“Catalogs are about more than table formats. They’re about governance as well,” said Roni Burd, Director of Open Data Analytics Engines at AWS. Burd also manages the company’s Iceberg contributions. “So there’s another really great opportunity to innovate on the catalog API, the abstraction layer above the table format. It’s what our customers are asking for. Because it’s opening up a new frontier of solving problems for them.”