Can Snowflakes Melt and Databricks Crumble?

Snowflake and Databricks are the two biggest names in the B2B data space. Over the past decade they grew from nimble startups to the two main incumbents in the data space.

While there are differences between the two companies and their products - Snowflake’s product is a cloud data warehouse, where Databricks product is a cloud data lakehouse - we ignore product differences in this post.

Instead we focus on analyzing the conditions under which both companies evolved and how this influenced their respective business models, go-to-market approach and financial profile. From there we will assess their exposure to competitive dynamics.

The circumstances

Both companies were founded early in the ZIRP (zero interest rate policy) era. Snowflake was founded in 2012 and Databricks a year later.

Both companies were built on the back of a tidal wave of VC funding. That VC funding was a direct result of ZIRP, as capital was finding its way to those investments with the best risk/reward profile and in that period many investors opted for venture capital.?

Both companies provide a managed data warehouse/lakehouse service built on top of AWS, Azure and Google Cloud.

Both companies’ ARR is roughly $2Bn the time of this article (Jan 2024).

Both companies record 70%+ gross margins.

Both companies raised over $4Bn since inception (source Crunchbase).

The ZIRP-era Startup Building Playbook

Certain key decisions founders take early on in a startup’s journey determine the DNA of the company, including but not limited to: VC funded or self-funded/bootstrapped, sales-led or product-led, paid or organic marketing.

Building a company in data infrastructure in the ZIRP era involved ample VC funding, because it was readily available. The ZIRP-era company building playbook looked somewhat like this:

1. Raise a pre-seed round.
2. Build MVP.
3. Raise a Seed round.
4. Find your first 5-10 customers. Ideally get to $1m ARR.
5. Raise your Series A.?

Usually, there are more steps after this, but let’s focus on Step 5, because all the steps that follow - Series B, C, D etc etc - are predicated upon Series A. Series A is when a startup gets a VC on the cap table and on the board. This cements the startup’s modus operandi for years to come and at least until a liquidity event.

So what was needed for a Seed startup to raise a Series A in the ZIRP era?

1. Potential for a big outcome (> 100x potential outcome).?
2. Approaching $1m in ARR.
3. At least 3-4x YOY growth rate.
4. Sticky product, i.e. high retention rate.
5. Gross Margin of > 70%.

What was not needed to raise a Series A - or any round for that matter - in the ZIRP era? Answer: profitability!?

Specifically, what did not matter? Answer: Customer acquisition costs!

The focus was on growth, only growth.?

The relationship between Gross Margin and Customer Acquisition Cost

Imagine, you are the Series A founder running Snowflake or Databricks. Imagine acquiring a new customer with a $100,000 annual contract and 70% gross margin. Would you be pleased? Yes, because you are thinking: $100,000 times at - let’s say - a 40x revenue multiple, you just added $4m to your company’s valuation!

But when you look at your annualized P&L it looks something like this:

Revenue $5m (100%)

Gross Margin $3.5m (70%)

Sales & Marketing cost -$10m (-200%)

Other expenses -$2m (-40%)

Operating loss -$8.5m (-170%)

So is this startup on the right track? It depends on (i) what we mean with “on track” and (ii) who you ask.?

Let’s consider two viewpoints:

1. The founder/CEO running Series A Snowflake or Databricks might think something like this: “I am at $5m annualized ARR now, growing 3x YOY so next year my ARR will be $15m. At a 40x revenue multiple I can raise $60m at a $600m valuation and continue my growth trajectory.”
2. A bootstrapped founder - or any CEO of a publicly listed company for that matter - would be: “We are in trouble. Our customer acquisition costs are almost 3x our gross profit. Our unit economics are totally out of whack!”

Series A Snowflake or Databricks is not worried about unit economics because the expectation is that “the show will go on”. The expectation is that there will be a Series B investor, a Series C investor etc. and eventually your random teachers pension fund that will buy the shares in the IPO.?

So the problem of the out of whack unit economics - the relationship between gross margin and customer acquisition cost - is pushed into the future.?

(note: With the benefit of hindsight we now know that "the show did not go on". The FED jacked up interest rates by 500 points in a one year timeframe and VCs that were all in on "growth at all costs" now call for “profitable growth” and financial discipline)

Network effects

There can be good reasons for growth at all costs.

For businesses with network effects it makes sense to focus on growth at all costs because of ‘winner-takes-all”-dynamics. Examples are: AirBnB (marketplace), Uber (marketplace) and of course Google (search). Once the winner is established, it can rely heavily on the network effect and lean on their established reputation at which point customer acquisition cost can drop significantly and profitability surges.

I argue that this is not - or to a much lesser extent - the case for Snowflake and Databricks. There are plenty of substitutes. To name a few: Amazon EMR, Google’s BigQuery, Dremio, HPE Greenlake, IOMETE , MS Azure Data lake, OneHouse.ai and Starburst. There is no “winner-takes-all”-dynamic and there is no point at which customer acquisition cost drops significantly. At least not to the same extent as for a company with significant network effects like AirBnB (that stopped paid marketing altogether and enjoys a 85% gross margin).?

Snowflake recognizes the importance of network effects and has been investing heavily in their “Data Cloud”, where customers can securely access each other’s data. The jury is out there on whether this will work. Data infrastructure is a > $100 Bn+ industry. Snowflake and Databricks have a small percentage of that market. Will they be able to get to AirBnB, Google search, Uber-like market share? I think it is unlikely. Also, because data exchange can be accomplished in different ways, avoiding vendor lock-in.

Unsustainable gross margins

Can a business with limited network effects and no real moat maintain a 70+% gross margin?

If there are no network effects and no real moat, new entrants will try to take away business from the incumbents by focusing on price or product and price. This is a fundamental free enterprise dynamic and also applies to enterprise software.?

Apart from the competitive dynamics, there is an additional impact. The impact of the availability of financing/funding on price levels. The emergence of student loans, inflated college tuition. The emergence of mortgages inflated house prices. The emergence of ZIRP-inflicted VC funding inflated prices and gross margins in the B2B data space.

What happens if the financing resource diminishes? Prices drop. Gross margins drop.

The balloon pops.

A clash between two generations

I predict a clash between the ZIRP-generation and the post-ZIRP generation of companies (the "Challengers"). These two groups grew up under wildly different circumstances. One grew up on too many calories, the other one is growing up with calorie restriction.?

What is a likely business model for a disruptor that wants to take business from Snowflake or Databricks? Let’s call this the post-ZIRP?or “Challenger Playbook”:

1. Raise a small seed round or bootstrap.
2. Build a similar or better product as Snowflake and Databricks leveraging the best open source projects with focus on amazing UX and flexible deployment options.
3. Follow a Freemium strategy: Create a free Community Plan and a paid Enterprise Plan.?
4. The Community Plan is free (for ever) and extremely generous.?
5. Enterprise Plan: Offer > 100% of the features for < 50% of Snowflake/Databricks costs..
6. Don’t do paid advertising. Fully rely on the freemium model, organic/SEO and word-of-mouth.?
7. Only raise significant funding once you understand your unit economics so you can be certain that you can combine growth and profitability and that the funding is accretive to accomplishing your goals.

Will Snowflake melt and Databricks crumble?

Nah. They'll be around, both are solid companies but their margins will compress due to competition. It might play out like this:

Some risk-averse customers that don't mind overpaying, will opt for the trusted Snowflake or Databricks brandname that was built on the back of billions of cheap ZIRP funding.?

Intelligent customers will see past the mascots, the sponsored sailboats, the conferences, the paid ads and try the challenger product at no (free community plan) or way lower cost (enterprise plan).

The intelligent customer may run the challenger product next to Snowflake or Databricks. Just to kick the wheels and check if they’re for real.

The intelligent customer starts with running one workload on the challenger data lakehouse.? Then add another workload. Then the customer realizes that the challenger product not only saves hundreds of thousands of dollars, but is actually pretty good.

This dynamic will force Snowflake and Databricks to compete on price. This will compress gross margins and it will force a hard look at their unit economics and cost structure.

“Your fantasy is running wild”, you may say. “Certainly a startup can not develop a product that is on par with Snowflake and Databricks? Surely you are aware that Snowflake and Databricks have hundreds of engineers.”

We already did. Start on our Community Plan today. It’s free. Forever.