Amazon.com (AMZN -0.30%) may be leading the growing market for cloud infrastructure, but Alphabet (GOOG -0.23%) (GOOGL -0.13%) is still putting up a fight.
Cloud infrastructure vendors have been crucial in enabling smaller start-ups, particularly those that are building mobile apps, to get up and running without having to worry as much about the back end. It's an appealing model since it allows smaller companies to simply pay for what they need without having to invest in building the infrastructure themselves. In other words, it's an OpEx model versus a CapEx model, and Google is betting big that the market is going to shift to the OpEx model in a major way in the years ahead.
Image source: Getty Images.
But as companies grow, they often choose to invest in their own cloud infrastructure. This gives them much more control over their own destinies, reduces risk, and over time the cost efficiencies improve after the large upfront investment.
Social media peers Facebook (META -1.12%) and Twitter both operate their own data centers; Facebook announced its seventh data center a few months back, while Twitter's infrastructure operations are expanding more slowly, as it operates in co-located facilities. Data centers are so strategically important to Facebook that it helped create the Open Compute Project back in 2011, publicly sharing Facebook's infrastructure innovations for the greater good.
But not Snap (SNAP -3.15%).
Hey, big spender
Snap primarily relies on Google Cloud for its infrastructure needs. While this is entirely par for the course for mobile app start-ups, Snap has grown to a size (nearly 160 million daily active users) that would typically warrant investing in infrastructure. The sheer scale of Snap's use of Google Cloud is what's unprecedented here. The Information reports that Google landed Snap with numerous discounts and other benefits, and Snap very much relies on Google Cloud at this point, although it also uses Amazon's AWS to a lesser degree.
Back in 2014, Snap CTO Bobby Murphy said the company picked Google Cloud over AWS because the company could scale better on Google Cloud. Snap poached AWS' head of data centers in October of last year, which is a peculiar move as it may imply that Snap has longer-term ambitions to build out its own infrastructure.
But for now, Snap is firmly in bed with Google Cloud, and spends quite a bit of money. Infrastructure costs comprise the bulk of Snap's cost of revenue. From the S-1: "Cost of revenue consists primarily of payments to third-party infrastructure partners for hosting our products. Hosting costs primarily include expenses related to bandwidth, computing, and storage costs." Cost of revenue in 2016 was $452 million, and Snap has committed to spending $2 billion with Google Cloud over the next five years.
But at what cost?
In discussing its operating leverage, Snap believes that using a capital-light business model will translate into "lower costs for us in both the short and long term." That's definitely true in the short term, but much more debatable in the long term.
Having a larger proportion of fixed costs actually helps with operating leverage as the business grows, since those fixed costs get spread out over an increasing revenue base over time, reducing total average costs. In contrast, if a company's cost structure has a larger proportion of variable costs, those costs will scale commensurately with revenue, so you don't get as much margin expansion. It's already telling that Snap's cost of revenue exceeded revenue in 2016 ($405 million), resulting in negative gross margin.
We saw this all play out years ago in the semiconductor industry, with the rise of the "fab-less" and "fab-lite" models that are dominant today. Building a chip fab costs billions of dollars, which was a significant barrier to entry. Then contract chip manufacturers came around and allowed companies to focus more on designing the chips while the manufacturing partners would produce them. By outsourcing production to a few large chip manufacturers, the industry collectively saved billions and innovation flourished. Semiconductors mostly shifted to the OpEx model long ago, and there are only a handful of companies that currently manufacture their own chips these days.
Snap is wading into uncharted territory by relying on third-party infrastructure partners at this scale. Whether or not this proves right or wrong will also depend on how Snap grows from here. One thing is for sure, though: Both aspiring start-ups and infrastructure vendors will be watching closely to see how this experiment unfolds.
