- The popular pay-as-you-go model for cloud computing is unexpectedly now a risky proposition.
- Some of the largest cloud data providers, like Snowflake, use consumption-based pricing.
- Experts say the model hasn’t been tested in a downturn and may be risky for companies relying on it.
One of Snowflake’s biggest appeals is its pricing model: You pay for as much computing as you use. This simplicity helped rocket the data-warehousing giant — and a class of other cloud-based companies — to enormous success.
Now, as the market crashes, that model is suddenly a massive liability. Companies trying to cut costs are looking to dial back their cloud operations through their own businesses slowing or increasing demand for efficiency internally. That leads to fewer uses — and fewer dollars for the cloud providers.
Snowflake’s stock tumbled after it reported earnings on Wednesday. The downturn had already wiped out more than $70 billion in market value for the pandemic darling, which held one of the largest software initial public offerings in 2020. On its earnings call on Wednesday, eight of the 20 questions in one way or another requested more information about consumption-based pricing.
While Snowflake said it had seen an upswing in consumption in recent weeks, it also signaled that some of its largest customers had pulled back, igniting fears that a wildly successful model would expose an entire industry to additional pressure in a
“Everyone loves a consumption model on an upswing,” said Derrick Wood, a managing director at the investment bank Cowen & Company. “I don’t think we’ve been tested in terms of what it looks like in a tightening environment. In a tightening environment, contracts, where you have ratable revenue contracted in your P&L, look pretty good.”
Cloud players say consumption-based pricing with companies like Snowflake is an unproven model in a downturn
Rather than sign hefty contracts with specific usage amounts — typical of old-school cloud vendors — Snowflake customers enjoy the flexibility of paying for as much computing as they need. Their spending can dial up and down as their businesses demand.
That’s been a big appeal of many modern cloud companies. Mission-critical orchestration tools like Prefect rely on consumption-based pricing. So do other modern data tools, like dbt, that companies have widely adopted to make the data-crunching process much easier. MongoDB, another large publicly traded provider, also relies on consumption-based pricing.
Now, as companies look to cut costs, some may cut down on those data operations. That could manifest as increasing the efficiency of existing processes, like daily updates through tools like dbt, or cutting those processes altogether. Each incremental improvement in efficiency or dropped query is one fewer dollar a company looking to tighten its budget has to spend.
“We would contend databases are less recession prone than other consumption companies, especially in the data analytics space, given databases are mission-critical to software development (i.e. cannot build an application without a database), whereas data warehouses, data science platforms, and even observability solutions all support more discretionary analytical projects,” RBC Capital Markets wrote in a note this week.
Given that consumption-based pricing is a recent development, there isn’t much data on how it will perform. Emil Eifrem, the CEO of Neo4j, another database provider that operates a consumption-based model, said there would be a sharp learning curve for all players.
“We don’t have answers in the downturn,” he said. “We’re all going to learn a lot over the next several quarters, and I think we will all come out of it smarter. Amazon Web Services was around in 2008, so they might have some data on it, but that’s it — no one else has anything.”
Snowflake offers an early glimpse of the liability of consumption-based pricing
While Snowflake showed that its pricing model was exposed to the forces of an economic downturn, it also offered a glimpse of the pricing model across the industry. On the earnings call, Wall Street analysts pressed executives for more information about the impact of a reduction in spending to gauge the model’s sensitivity.
Snowflake defended its data-warehousing model — though it did take a jab at its chief rival, Databricks, which focuses heavily on machine learning and data science, saying that kind of spending could be seen as discretionary.
CEO Frank Slootman said on the earnings call the company’s types of core data warehouse workloads “are not going anywhere.”
“They’re not optional,” he said on the earnings call. “They’re not like, what do I feel like doing today? That, by the way, there are workloads like that, that’s far more on the data lake side, where essentially you have a massive repository of files. You may have data scientists that are just sort of fishing files out of the lake and trying to decide to do some interesting stuff for that. That sort of thing is highly discretionary, but that’s not the focus of our business.”
Snowflake, however, is still heavily investing in machine learning and data science by launching new features to support the Python language and spending $800 million on a small machine-learning startup called Streamlit. It’s an area where Databricks has traditionally been stronger, while Snowflake has blossomed thanks to its data warehousing and analytics.
“Our data lakehouse architecture covers the full journey, from the initial data to predictive modeling, prescriptive analytics, and automated decision making,” Databricks CEO Ali Ghodsi told Insider. “We see some of the most valuable data use cases on the latter. Those projects are not discretionary, but highly strategic to organizations. We’re therefore seeing a continuous incremental growth in our business thanks to that.”
Still, there may be room for optimism for Snowflake, as it might offset decreased spend among some major customers by growing the usage of other customers, Wood said.
“I wouldn’t convict the model,” said Brad Zelnick, a managing director of software equity research for Deutsche Bank. “I would say that it’s the more forward-leaning, innovative companies out there that are employing consumption models, whether it’s AWS, MongoDB, or Snowflake.
“It’s just the nature of the model is such that if the customer at a point in time is going to decide to moderate their consumption in some way, there’s a real-time impact on the income statement.”