Do You Really Understand Snowflake’s Revenue Model? (SNOW)

The Snowflake stock (NYSE:SNOW) has lost more than half of its value this year, yet it still trades at a significant premium to DevOps software peers with similar growth profiles. The company has been most commonly known for pioneering its consumption-based model in software, a pricing strategy that has historically been more prominently deployed by hyperscalers like AWS (AMZN), Azure (MSFT), and GCP (GOOG / GOOGL).

While the consumption-based pricing model has been viewed as a risk for software service providers during times of economic turmoil, as it does not provide any revenue security, many corporate customers view the strategy as a favourable option in times of mounting macro uncertainties. Although this may seem like a paradox at first glance, and potentially good news to consumption-model-based software providers like Snowflake, the resulting rush of demand does not necessarily provide immediate relief to the company’s near-term performance. Specifically, the consumption-based model will continue to be a major overhang on Snowflake’s near-term performance, subjecting it to the forefront of fundamental impacts ahead of a possible recession.

As the Federal Reserve remains fixed on raising interest rates on mixed recent economic data, alongside protracted market volatility in response to rocky investors’ sentiment this year, Snowflake’s consumption-based model, as well as other dark spots like the unprofitable nature of its business today, remain core downside risks to the stock. Given Snowflake’s lofty valuation premium to peers (SNOW forward EV/S 17.7x; high-growth DevOps average forward EV/S 9.3x), the stock is exposed to greater susceptibility to a protracted downtrend ahead of an unfavourable operating environment and market backdrop. The following analysis will provide a detailed overview of Snowflake’s consumption-based revenue model, and gauge the implications of said pricing strategy on the company’s near-term fundamental and valuation prospects ahead of a looming recession.

Snowflake’s Revenue Segments

Snowflake’s revenues are primarily generated from the sale of data storage, compute and transfer services through its “Data Cloud” platform. Specifically, Snowflake’s Data Cloud enables customers and technology partners to consolidate and access data within one single reliable source – the more customers / data participants, the biggest and more valuable the Data Cloud. Via its cloud-based storage, compute and transfer architectures, Snowflake’s Data Cloud facilitates a wide variety of key data workloads spanning “data warehousing, data lakes, data engineer, data science, data application development, and data sharing”. Related revenues generated from the sale of Data Cloud services are recorded through Snowflake’s Product segment, which represents about 94% of Snowflake’s consolidated sales.

The remainder accounts for sales generated from Snowflake’s provision of processional services, including “consulting, onsite technical solution services, and training related to platform use”, as well as other services such as the offering of access to a “dedicated instance of a virtual private deployment”.

Snowflake’s Revenue Model

Snowflake’s Product revenues are primarily generated from a consumption-based model, where customers only pay for what they use. Only a nominal portion of Snowflake’s revenues are generated from subscriptions. The consumption-based model is separated into two types – pre-committed capacity and on-demand:

1. Pre-committed capacity consumption-based billing model – This is akin to a pre-paid mobile phone plan. The majority of Snowflake’s customers typically sign a contract with pre-committed capacity to secure better unit pricing – the higher the pre-committed capacity, the cheaper the per-unit economics for customers. The average duration of related pre-committed capacity contracts are currently 2.4 years at Snowflake, with many providing payment upfront. Related revenue is recognized by Snowflake only when said pre-committed capacity is used by the customer. Any overages beyond the contracted capacity pre-commitment are billed at the standard “on-demand” rate, typically billed to customers in arrears.
2. On-demand consumption-based billing model – A nominal portion ($7.2 million, or less than 2%) of Snowflake’s Product revenues in the fiscal second quarter were generated based on an on-demand consumption-based billing model. Under said pricing model, customers are not hooked on a pre-committed capacity contract. Hence, said usages are not subject to any unit pricing discounts. Customers under the on-demand consumption-based billing model are typically billed in arrears for the amount of service consumed, akin to monthly utility bills.
3. Subscription billion model – The traditional software billing model can be found in Snowflake’s Professional Services and Other business. For instance, the deployment of virtual private instances is priced at an annual fee billed to clients upfront, and is recognized into Snowflake’s consolidated revenues “ratably over the contract term”. Related deployment fees charge currently represent 1% of Snowflake’s annual consolidated revenues.

As discussed in the earlier section, Snowflake’s Data Cloud platform operates primarily via three architectures – storage, compute and transfer – and customers consuming said services are primarily billed under a pre-committed capacity consumption-based model:

1. Storage – Snowflake’s proprietary platform is capable of storing and managing all of “structured, semi-structured, and unstructured data” into one centralized source to ensure easy access and enhanced data analytics. Snowflake’s data storage service is billed by the volume of capacity consumed by customers – the greater the pre-committed storage volume capacity contracted, the lower the per-unit cost for contracted customers. The pricing is determined based on the pre-contracted price per terabyte and the “monthly average amount of data stored after compression”.
2. Compute – Compute services provided by Snowflake primarily assist customers to “simultaneously access common data sets for many use cases with minimal latency”. This would essentially permit Data Cloud access to as many authorized users across as many use cases as required in a seamless and secure way. Specifically, Snowflake has noted that its “storage layer is independent of compute”, which would allow users to access data from the centralized source and use/manipulate it without compromising the integrity of the original data in the storage architecture. Compute capacity consumption is billed by credits per hour. Customers typically sign onto a pre-contracted commitment a specific compute capacity based on estimated need. The consumption-based billing model would allow customers to scale up or scale down compute needs on demand. Instead of an “always-on” approach, Snowflake’s platform can be “switched off” when “engineers are not active on the platform” so billing and usage is more “intention-based”. This differentiates the company’s compute service from other consumption-based DevOps software providers that may operate under an always-on approach that requires users to keep the underlying software application running even when the engineer is not actively on it. Snowflake’s compute services are provided via three categories – virtual warehouse compute, serverless compute, and cloud services compute. Virtual warehouse compute is wholly “user-managed”, with compute credit consumption entirely dependent on the usage of compute resources such as executing queries, loading data, and performing other data manipulations operations. Serverless compute, on the contrary, provides compute resources that are “managed by Snowflake, rather than traditional virtual warehouses”, meaning that credit consumption will be based on Snowflake’s determined requirement for specific workloads. Snowflake also provides Cloud Services compute, which provides “services that tie together all the different components of Snowflake to process user requests, login, query display, and more”. This is similar to a web-hosted cloud-based portal like Google Drive where users can access any component of the productivity tool – or compute request, in the case of Snowflake’s Cloud Services compute – with said resources managed by the service provider.
3. Transfer – Snowflake’s platform also facilitates secure transfer and sharing of data within and beyond organizations worldwide, without compromising the integrity of the underlying source. This again plays to Snowflake’s vision of “eliminating data silos, empowering security and governed access to data, and [removing] data management and infrastructure complexities”. Data transfer services are billed to clients depending on the size of data transferred, the region of the data’s origination and destination, as well as the host public cloud on which the data is based / transferred to. Snowflake’s Data Cloud platform is currently hosted by three major public cloud service providers – AWS, Azure and GCP – which allows the company to support a wide variety of workloads across 31 regions worldwide.

Pros and Cons of the Consumption-Based Model for Snowflake

Unpredictable Revenues (Con). There are a few inherent business risks to the consumption-based model for Snowflake, with the unpredictable nature of revenue recognition being most commonly discussed. Specifically, a consumption-based model allows customers to scale up or scale down usage whenever they please, as opposed to the subscription-based model where customers can only opt-in / opt-out of subscribed services upon renewal. This essentially subjects consumption-based service providers like Snowflake to a highly choppy revenue recognition trajectory, even if the majority of its customer base have contracted pre-commitment capacity. The hope is that “total revenue generated over time [under the consumption-based model] will be meaningfully higher to traditional models” as customer demand increases, which is dependent on the effectiveness and viability of the product offered.

The choppy nature of the consumption-based model’s revenue recognition trajectory also introduces complexities to performance metrics. Traditional performance metrics typically applied to subscription-based revenue models such as subscriber growth / subscription count, or annual recurring revenue (“ARR”) are irrelevant in the case of consumption-based revenue models because even if the number of pre-contracted customers grow, it does not necessarily translate to immediate (or sometimes, future) revenue growth, nor is there any annual recurrence of a standard fee charged.

The most reliable performance metric for consumption-based revenue models for now is net revenue retention (“NRR”), which compares the growth of revenues generated by the same cohort of customers in the current year and the prior year to determine if the service is consistently being demanded, or if it is generally “shelfware” (i.e. installed/subscribed by not consumed). Given the nascent implementation of a consumption-based revenue model across DevOps SaaS, some industry participants have also suggested the future disclosure of a “consumption run rate” metric to better gauge service’s demand by unit/volume (as opposed to NRR which gauges demand by dollar value, and may not be reflective of real usage growth).

Increased Churn Exposure (Con) – Given the nascent implementation of a consumption-based revenue model across software providers, many remain inexperienced in recommending capacity commitment to customers looking to lock-in more favourable unit pricing. This has often caused customers to experienced frustration over unexpected overage charges, resulting in reduced usage in the short-term to prevent hefty bills, and churn in the long-run as they move to rival vendors. This has been a prominent headache for Snowflake, leading to its recent introduction of various platform efficiencies to “help reduce costs for customers”:

As a result of that consumption model, some early users are often surprised to get much bigger bills from Snowflake than they anticipated.

Source: Bloomberg

Efficiency cannibalization (Con). This brings us to the third inherent business risk introduced by consumption-based revenue models – efficiency cannibalization. Improvements made to platforms to enable improved usage and pricing efficiency (e.g. enhanced storage compression; improved cloud infrastructure processors) could essentially result in a lower consumption requirement by customers, thus lower revenues for consumption-based service providers.

In the case of Snowflake, its recent implementation of various “improvements to the company’s data storage and analysis products” are expected to decrease overall customer consumption requirements, resulting in expectations for a drastic deceleration in revenue growth over the near-term. This, alongside broader market deterioration this year, has cost Snowflake about $50 billion of its market cap earlier this year after the stock plunged on its first deceleration from previously eye-popping triple-digit y/y growth to now double-digit y/y growth, with management forecasting a $97 million revenue headwind in fiscal 2023 from the implementation of said platform enhancements alone.

Improved mass market penetration (Pro). Yet, the consumption-based revenue model, paired with Snowflake’s enhanced pricing and performance efficiencies makes a positive appeal to the mass market. Many software operators struggle from having their revenues saturated at large corporations often due to the complexity and hefty cost of services provided.

But under Snowflake’s “industry-vertical-focused” go-to-market strategy, which tailors its services to specific business solutions including “Financial Services Data Cloud, the Media Data Cloud, the Healthcare and Life Sciences Data Cloud, and the Retail Data Cloud”, paired with its attractive consumption-based revenue model, customers are offered a lower barrier to adoption on software solutions that were previously viewed as complex and expensive. This would essentially open the door for Snowflake to greater mass market penetration opportunities ahead of longer-term digital transformation tailwinds, which is consistent with management’s aspirations for the company to “attract more clients in the future” with its recent implementation of platform improvements.

The growing success of the consumption-based revenue model in attracting new customers is further corroborated by software peer Autodesk (ADSK), which recently introduced a “Flex” license priced at $21 per day as opposed to its standard $235 per month subscription plan in a move to provide greater usage flexibility to customers. Autodesk noted 40% of Flex users are new to its services, underscoring how the consumption-based revenue model could be a “powerful tool” in attracting less affluent corporate users (e.g. small- and medium-sized enterprises) that otherwise could not have been able to afford a full subscription to premium / complex software.

Pros and Cons of the Consumption-Based Model for Users

Cost opaqueness (Con). As mentioned in the earlier section, the nascent implementation of the consumption-based revenue model across software deployment has led to customer frustration over the lack of transparency in pricing. While the consumption-based model is intended to improve spending efficiency for customers by only charging for services consumed, many are often surprised at the month-end bill, ladened with higher-than-expected unit consumption and/or overages billed at less attractive standard rates.

A primary driver for this inefficiency is the lack of experience across software service providers in recommending the appropriate capacity commitment to customers looking to lock-in attractive unit pricing, resulting in a high occurrences of costly overages. The lack of transparency in what “consumption” comprises of is often a driver of cost frustration as well (as discussed earlier, different consumption-based models define consumption differently, with some service providers charging under an always-on approach, and others on an intentional usage approach). The majority of Snowflake’s software solutions provided are also hosted on non-user-managed servers (except serverless services compute), which potentially subjects customers to higher-than-expected “peak capacity” rates as well, which is likely a more prominent impact for users charged based on an on-demand rate.

Flexibility (Pro). While a potential risk for service providers, the flexibility offered under a consumption-based revenue model is welcome news for customers as it allows them to scale up or scale down usage according to need. Although the majority of customers are locked into a fixed capacity contract, the prepaid credit is typically not used until service is actually consumed. And any unused credit upon expiry of the contract can easily be rolled over by a new top-up, just like most pre-paid / pay-as-you-go mobility plans. This makes it significantly less costly to try out a new service, especially for SMEs as discussed in the earlier section. It also reduces shelfware, making a consumption-based model much more economical from customers’ perspective by eliminating onerous payment obligations (e.g. do not need to wait until year-end renewal period to cancel / upgrade service).

Recession Implications for the Consumption-Based Model

Looming recession risks are making consumption-based models increasingly attractive in a boardroom where executives are “talking pennies, not millions”. Many software service providers – such as Autodesk, as discussed in the earlier section – have rushed to implement their own consumption-based offerings to accommodate the increasingly price sensitive customer base ahead of mounting macro uncertainties. With businesses “increasingly shifting to become more customer-friendly”, the eventual implementation of a consumption-based model either in replacement or alongside the more common subscription-based model is likely imminent.

However, the new pricing strategy does not necessarily safeguard software service providers looking to salvage their wobbly growth prospects under tightening financial conditions. While it is “unsurprising tech leaders are willing to try new pricing strategies to keep growth going”, the consumption-based model does not necessarily translate to any more revenue growth than the traditional subscription-based model within the immediate term. It is one thing to appease the shift in customers’ preference for a consumption-based model to better align with belt-tightening goals, and another thing for software service providers / operators to safeguard their respective top- and bottom-line growth trajectories.

As discussed in the earlier section of consumption-based model pros and cons for software service providers and customers, respectively, each benefits from different incentives under the pricing strategy. There is often higher demand from customers for consumption-based models during a recession given it helps them secure long-term pricing discounts, without disrupting immediate plans to tighten spending patterns. This is in contrary to traditional subscription-based models that would require customers to pay the annual fees, regardless of headcount cuts or volume reductions during the subscription period, with no / minimal changes permitted in the arrangement until the next renewal cycle.

What this flexible scale up / scale down benefit offered to customers for consumption-based software service providers like Snowflake is that its growth and margins would be the first to see a drastic adverse impact at the first sight of an economic downturn – even if they are seeing a higher number of customers. Recall that customer growth is not equivalent to immediate revenue growth for consumption-based software providers – customers can sign up today for an “x” amount of storage or compute capacity to secure long-term unit pricing discounts, but use less or not use it at all as they brave through the looming macro storm, subjecting consumption-based software providers to immense near-term revenue risks. This also means that NRR, the key performance metric for consumption-based software providers, will likely see drastic deceleration (or even declines) during early stages of a looming or confirmed recession.

In Snowflake’s case, which is already expected to experience a decelerated pace of growth compared to the prior year due to recent implementation of platform improvements, is now likely facing compounded vulnerability to revenue risks due to looming recession risks as well. With Europe – one of Snowflake’s core operating regions – now anticipating a recession soon due to “elevated uncertainty, high energy price pressures, erosion of households’ purchasing power, a weaker external environment and tighter financial conditions”, the company will likely become one of the firsts to experience a more prominent impact on its revenues.

And similar risks are faced in the U.S., where Snowflake generates the majority of its revenues. While the recent release of weaker-than-expected CPI and PPI figures for October indicate that inflation might have already peaked, new data on unemployment claims and retail sales indicate that the economy remains strong, meaning the Fed will remain on its hawkish tightening trajectory to ensure decades-high price increases are structurally brought back in line with the 2% long-term target. This means demand will inevitably slow across all verticals, including IT spending which was previously considered relatively resilient due to digital transformation trends, making a near-term headwind for Snowflake.

Final Thoughts

Anticipated growth deceleration, paired with projected market fragility is likely to bode unfavourably for Snowflake’s still lofty valuation premium compared to peers with similar growth profiles. The unprofitable nature of the business is also likely to drag on sentiment in the near-term as investors’ preference shift from growth to profitability under the dire market climate.

However, given the favourable long-term secular demand environment for data management software like Snowflake’s Data Cloud platform, as well as the proven viability of its product, the potentially inevitable decrease to the stock’s price could make a compelling entry opportunity to capitalize on longer-term upside potential. A consumption-based model company is also likely to be the first to benefit from an economic recovery, given there is no time-lag in re-ramping up revenue (e.g. no wait time for subscription-based customers to rehire previously laid-off talent). And considering the anticipated increase in new customer volume acquired over the looming economic downturn due to the shift in demand preferences from subscription-based to consumption-based, Snowflake will likely be in an advantageous position once macro headwinds clear.

But it would likely be prudent to stay on the side-lines for now as there is still significant headroom for downside risks to play out for Snowflake, particularly as market continues to adjust to an unravelling macroeconomic environment.