(Bloomberg) --
Snowflake Inc. was looking for acquisitions. One startup, Tabular, seemed a potentially great asset to help the software company catch up in artificial intelligence, and Snowflake entered exclusive negotiations to buy it for more than $600 million.
Then Snowflake’s primary competitor, Databricks Inc., swooped in, and ended up paying nearly $2 billion for Tabular, according to people familiar with the deal. That price tag was an unheard-of valuation for a startup doing only about $1 million in annual recurring revenue. Adding further insult, the agreement was announced during Snowflake’s annual conference just before the keynote speech by new Chief Executive Officer Sridhar Ramaswamy.
The Tabular tussle was the latest chapter in one of the fiercest rivalries in the technology industry. Snowflake and Databricks are jockeying to be the key platform for organizing, analyzing and using huge heaps of data for AI. More than four dozen current or former employees, customers, or partners of the software vendors spoke for this story — most of whom asked to remain anonymous to avoid professional repercussions in an environment of intense and sometimes single-minded competition. The employees also describe a growing sense that massive cloud infrastructure providers, particularly Microsoft Corp., present the largest looming threats to their businesses.
Each firm was founded in the early 2010s and initially had separate niches in the wide area of data software. They referred business to one another and Databricks’ marketing team even used Snowflake for analytics.
But the relationship broke down in recent years as each company released overlapping products. Snowflake expanded quicker and in 2021 had the largest-ever software initial public offering. Databricks now is one of the world’s highest-valued startups and an IPO is highly anticipated.
Once considered the little brother, Databricks has played the public antagonist with pugnacious marketing and sales tactics. “SnowMelt” is a not-so-subtly named initiative within the firm to take business from Snowflake, especially in its home court of data warehousing, according to multiple people familiar with the matter.
Sellers can win bonuses for moving Snowflake clients to Databricks’ directly competing service. In other cases, Databricks salespeople offered to help pay off prospects’ Snowflake contracts in the form of credits if they switched vendors. Or they steeply discounted their offering to win business.
The typical pitch by Databricks is that its platform is less expensive and contains a wider suite of advanced features, such as those for building AI models for businesses from complicated unstructured data. Sellers are equipped with a program that estimates how much money potential clients would save by switching to Databricks from Snowflake.
Snowflake, too, says its offering is cheaper. “The claims on cost drive me nuts,” said longtime Snowflake product chief Christian Kleinerman in an interview. “I tell the customers — just try it and go run a representative benchmark.”
Databricks’ talking points are hammered home via advertising. When attendees to Snowflake’s 2023 user conference landed in Las Vegas, billboards were waiting for them at the airport and outside the conference venue claiming Databricks’ product was nine times cheaper.
Databricks CEO Ali Ghodsi is known for giving fiery media interviews and posting migration stories or favorable benchmarks on LinkedIn. Employees recall being asked to like or share these posts on their own accounts.
Going after Snowflake helped spread awareness about Databricks, Ghodsi said in an interview. “Two or three years ago nobody got fired for buying Snowflake as their data warehouse — I don't think that's true anymore.” Now, Ghodsi said he’s no longer orienting employees against Snowflake, focusing them instead toward goals like promoting their AI and data governance products.
While hard to know how much is due to its combative style, something is working for Databricks. Unlike most other large software companies, its revenue growth is accelerating, according to an investor presentation in June. Recurring sales were expected to hit $2.4 billion in July, with the company’s relatively new warehousing product — which most-directly competes with Snowflake — contributing more than $400 million.
Investors have rewarded Databricks with a $43 billion valuation, aligned with where the stock market currently values Snowflake, at $42.5 billion. In recent weeks, Databricks has reached out to existing investors to get detailed ownership records, often seen as a precursor to an IPO filing, according to a person familiar with the matter. Ghodsi declined to comment on a timeline for an IPO.
Some of its growth is coming at the expense of Snowflake, said Ron Gabrisko, Databricks chief revenue officer. “Eight of the top 10 Snowflake customers have moved workloads to Databricks,” he said. The rivalry helps keep sellers motivated. “Every good sales team has some competitive fire under them,” Gabrisko said. Databricks plans to add 1,500 more workers in the current fiscal year ending in January, a spokesperson said.
Snowflake reported $3 billion in annual sales over the last 12 months, growing only about half as quickly as its startup rival. Mike Scarpelli, Snowflake’s chief financial officer, said in an interview that his company has higher profitability. “The reality is we’re generating cash and they're burning cash. How long can they do that for?”
Each platform can be used for ingesting and analyzing huge sums of data — such as an airline trying to understand which customers are most likely to cancel their flights based on ticket price, destination and weather patterns. The market for this kind of software is rapidly growing and not entirely zero sum — many companies use both Databricks and Snowflake for different types of work, while countless others are still using older-generation tools that are traditional replacement targets, according to data from market research firm Enterprise Technology Research.
In a July ETR survey of companies that use both software tools, more customers said they were shifting workloads to Databricks than vice-versa, but the vast majority said they would keep both. Still, Databricks’ Gabrisko said he expects in the long term customers will pick one or the other.
“I have no idea why he is so obsessed with Snowflake, because I am not obsessed with Databricks,” said Scarpelli, the Snowflake CFO, in reference to Ghodsi.
Snowflake, the more established software vendor with higher revenue, has been less vocal about rivalry. But it, too, has bragged about poaching customers and launched products to compete with Databricks, such as for data engineering and machine learning. “SparkAttack” is an initiative within the company to snatch machine learning workflows from Databricks, a reference to the “Spark” technology created by Ghodsi and other members of Databricks’ executive team.
“They put up slides about all these customers they have — well the reality is virtually every one of those customers is a Snowflake core data warehousing customer and growing with us,” Scarpelli said. “As a private company you can say whatever you want.”
Kleinerman, the Snowflake product chief, said Databricks’ negative marketing makes claims, particularly on cost and product breadth, that haven’t been true in years. Snowflake has made huge strides in products for machine learning, unstructured data and cost management, he said.
Ease of use is one of the main pitches by Snowflake, including in sharing data between companies and users. Salespeople often argue the platform is more akin to an easy-to-grasp iPhone while Databricks is the highly customizable, but sometimes glitchy, Android. Both companies charge customers based on how much they use the product like a utility bill, rather than the flat-rate subscriptions dominant in application software. Cloud companies that rent computing power and storage, such as Amazon.com Inc.’s AWS, also charge this way.
Bond Brand Loyalty, a Canada-based customer experience firm, saved money and increased performance by standardizing most of its data work onto Snowflake, said Francis Silva, who leads its data practice. That’s because less-technical users were able to use Snowflake in a way they weren’t able to use Databricks, he said.
Still, Snowflake has been on the defensive. Its stock price is down 36% this year as Wall Street worries that customers are trimming software spending, dragging down sales growth rates. A recent hacking campaign aimed at customers without multi-factor authentication has yielded a storm of tough headlines.
Ramaswamy, a former Google advertising leader who took over as Snowflake CEO in February, has focused on launching new products. His tenure has brought “a lot more energy in the office — better sense of urgency,” Scarpelli said.
Ramaswamy originally joined Snowflake via the acquisition of AI search startup Neeva in 2023. Databricks also bid at the time, according to people familiar with the process. When Ramaswamy became CEO, Databricks’ Ghodsi took some credit for the executive turnover. “I think we put a lot of pressure on them,” Ghodsi said in a March Bloomberg TV interview. “Snowflake was basically not doing AI whatsoever.”
While the rivalry has drawn comparisons to earlier tech feuds, the biggest competitive risk for each company may be the mammoth cloud providers — Microsoft, Amazon and Alphabet Inc.’s Google. These giants are already required as a base computing layer and now they’re working to improve their data products, giving them the ability to bundle different tools and potentially undercut the smaller vendors on price.
“We actually see our biggest competitor today being Google with BigQuery,” Snowflake’s Scarpelli said, referring to Google’s data platform. He also noted Microsoft as an emerging threat.
Within Databricks, “coopetition” is a becoming a more common way to refer to the relationship with Microsoft. The two have long had close ties — a massive chunk of Databricks business comes from customers using Microsoft’s cloud infrastructure owing to a highly successful joint service sold through its Azure cloud services. Microsoft even discussed an acquisition with Databricks in late 2017, according to people familiar with the matter.
But Microsoft has been investing in its own competing data analytics service, which rebranded last year as “Fabric.” The software is sold alongside Power BI, a common data visualization tool that many customers of Databricks and Snowflake are already using. Microsoft, the world’s largest software maker, listed Databricks as a competitor for the first time in a July filing.
"It's a relatively new product," Adam Conway, senior vice president of product at Databricks, said about Microsoft’s Fabric. "On the record, I'll leave it at that." He added that he's "not worried about another company out-innovating us."
Partnership is still the official line. During an all-hands meeting earlier this month, Ghodsi warned against criticizing Fabric or other Microsoft products on social media, telling attendees that “our message to customers should always be that we are better together,” according to a presentation slide seen by Bloomberg.
But Databricks appears to be toughening its positioning against Microsoft. Earlier this year, it released its own visualization tool which could compete against Power BI. Some sellers have received new anti-Fabric talking point guidance while others have been workshopping a competitive codename for Fabric in the mold of “SnowMelt.” And just as Databricks employees threw stones at Snowflake on social media, they are now chucking them at Microsoft. Last month, one director wrote a LinkedIn post about Fabric reminiscent of the classic anti-Snowflake playbook.
“PSA,” wrote a Databricks worker on LinkedIn, “if you or someone you know is considering Fabric for your data platform or to build GenAI applications make sure you read the fine print or you may find yourself paying 3X for some workloads.”
--With assistance from Dina Bass and Sarah McBride.
©2024 Bloomberg L.P.