In 2011, Marc Andreessen famously declared that software would eat the world. Last month, two members of his investment firm Andreessen Horowitz made the case that cloud has been eating the enterprise’s money.

In a post titled “The Cost of Cloud, a Trillion Dollar Paradox,” Martin Casado and Sarah Wang detailed an analysis of enterprise cloud spending, based on a review of financial reports from 50 public software companies. Their conclusion was that in operating at scale, the cost of cloud could double a firm’s infrastructure bill.

Moreover, they say, an estimated $100 billion in market value is being lost as a result, based on the impact of cloud costs on margins. One company cited in the post found that public cloud list prices were 10 to 12 times the cost of running one’s own datacenter.

When a prominent investment firm such as Andreessen Horowitz, which backed Facebook Inc., Instagram, Slack Inc. and Lyft Inc. in the early days, suddenly raises an alarm about a foundational piece of the enterprise world, the tech industry sits up and takes notice.

“The reaction from a lot of people in the tech industry has been: ‘Yes, we’ve known about this for a while,'” Rob Lee, vice president and chief architect at Pure Storage Inc., said in an interview with SiliconANGLE. “It really depends on the size and scale of the customer. People recognize that the cloud is costly.”

The analysis has generated a great deal of debate in Silicon Valley, including one pointed rebuttal. In his own post published in VentureBeat this month titled, “Andreessen Horowitz is Dead Wrong About Cloud,” Rich Hoyer, director of consumer FinOps at SADA Inc., made the point that it was more sensible to analyze the return on investment for one set of costs against the alternative. The benefits of cloud relate to revenue, not cost, according to Hoyer.

“The problem with the Andreessen piece is it has blinders on in terms of the cost itself,” Hoyer said in an interview with SiliconANGLE. “The way an enterprise should be looking at this is they should be doing an ROI analysis on their cloud costs. Don’t make the decision based on being cheaper. Ask how workloads will improve the business’ operations to create new opportunities.”

And even before Casado’s and Wang’s piece, Dave Vellante, co-founder of SiliconANGLE Media sister market research firm, had called the notion of “cloud repatriation” a “myth,” putting the lie to the notion that enterprises themselves are finding the cost of cloud computing enough of an issue to switch back to their own data centers.

The Dropbox example

The Andreessen analysis has drawn fire for citing Dropbox Inc. as a prominent example. Dropbox made the decision about eight years ago to migrate billions of files from 500 million users off Amazon Web Services Inc. and onto custom-built infrastructure in colocation facilities. The move saved nearly $75 million over a two-year period prior to its IPO in 2018, according to the researchers.

Corey Quinn, chief cloud economist at The Duckbill Group, built his business by analyzing AWS costs and advising clients on maximizing cloud value. He took issue with the Dropbox analogy, noting that the company actually migrated 34 petabytes of data to AWS in 2020.

“Ever notice there isn’t a second story like this?” Quinn said in a recent interview. “It was a single, very well-understood, very niche workload: storing user files with access patterns that didn’t (at the time) align with S3’s pricing dimensions.”

In a set of Twitter posts following the publication of his research, Casado noted that Dropbox was not “core” to the analysis. “We used Dropbox as an illustrative example because the data is public,” Casado wrote on June 2. “We didn’t use it to drive the savings assumptions on the analysis. So even if you ignore any mention of it, the results are the same.”

Repatriation re-examined

The Dropbox example has also opened a fresh look within the tech world at the thorny question of repatriation, a shift of data and workloads from the cloud to on-premises infrastructure. Is this really a trend?

One survey of 350 information technology decision makers, commissioned by Virtana Inc. and conducted by Arlington Research in November, found that 72% of respondents had moved one or more applications from a public cloud back on-premises.

However, using data provided by Enterprise Technology Research and other sources, Vellante found that evidence for repatriation becoming the next big thing was sorely lacking. ETR’s most recent data showed that 55% of enterprises surveyed, for example, indicated plans to increase spending with AWS in 2021.

“I definitely don’t believe you can simply move from cloud back to on-prem or hybrid without unintended consequences,” said Vellante. “Complexities around talent, resource allocation, IT alignment, the ability to attract developers may all be impacted by a decision to repatriate. Dropbox was the poster child of that piece, and I don’t see them as an inspiring example of repatriation translating to valuation growth.”

Several executives contacted by SiliconANGLE, whose firms would normally be closely involved in repatriation decisions, also indicated they did not see repatriation as a growing trend. One of these firms, CloudCheckr Inc., provides visibility and insight to enterprises seeking lower cloud costs.

“We’re not seeing very much of that,” Travis Rehl, vice president of product at CloudCheckr, said in an interview with SiliconANGLE. “The Dropbox example is a little more unique, the primary driver was storage cost. I get why some want to repatriate, but I don’t think that’s a scenario for most people.”

Higher cloud costs

Although there may not yet be a mass exodus of workloads from the public cloud tent, the cost of admission is still worth examining. Follow-up comments from Casado on his Twitter feed noted that the benchmarking conducted by his research with Wang found that cloud spend accounted for approximately 50% of cost of goods or COG, and topped out at 80% in the case of one large enterprise.

Vellante’s own analysis of public filings for Snowflake Inc., a data warehouse company that runs completely on cloud infrastructure, found a disclosure of $247 million in cloud purchase commitments over a five-plus-year period. When Snowflake’s latest 10-K report was reviewed, that same line item had ballooned to $1.8 billion.

Findings such as these provide evidence that cloud costs have become a significant percentage of total IT spend. Where the Andreessen analysis and viewpoints of others in the tech community diverge is over the true value received by the enterprise for that investment.

Casado and Wang note that by spending such a high COG level on cloud, enterprises could be leaving as much as $500 billion in market value on the table. Yet, as Hoyer and others point out, the focus should be on cloud optimization, not cost.

Hoyer described how SADA had worked with one client concerned about the amount it was spending on cloud. SADA’s experts got the company focused on frameworks for cloud usage and analyzed precise data over where the costs were going. A new FinOps team was created to drive this effort.

“The question is not: Is it too expensive?” Hoyer said. “It should be: How do we run it efficiently? Any one of the major cloud vendors can give you 20 cases of phenomenal success stories.”

Margin versus workload

Amid the debate over cloud cost and optimizing value is a key point made by Casado and Wang at the end of their post. If they are right, and enterprises decide to do something about it, the major cloud providers will have a critical decision to make. Do they cut prices and sacrifice profit margins or stand firm and accept the loss of workloads to on-premises infrastructure?

There is a significant amount of money involved in this decision. Gartner projects that public cloud spending will reach $304 billion in 2021 alone, up from $257 billion the previous year. Perhaps more significantly from a corporate perspective, the two largest public cloud providers – AWS and Microsoft Corp. – are the second- and fourth-largest companies in the world by market capitalization. Google LLC, which trails the other two in cloud share, ranks fifth.

None of the executives consulted for this story were prepared to accept that any of the major cloud providers would reach this fork in the road very soon. These companies are too well-capitalized and the level of innovation will be hard to match.

“I think that’s a question seven years from now,” Rehl said. “It’s not in the short term, but it’s not that far away. As long as those public cloud providers continue to provide new services that are innovative, I think they can continue to command higher costs.”

For his part, Andreessen Horowitz’s Casado is building a reputation for raising the level of dialogue around tech economics. In February of last year, Casado and Matt Bornstein published an analysis of the AI industry that challenged the notion that firms using AI were built on the same economic models as software businesses.

Casado and Bornstein analyzed the financial filings of software services companies and interviewed executives from AI firms. They determined that AI companies had lower gross margins and scaling challenges from the complicated work of edge computing. In a foreshadowing of Casado’s most recent research project, the two determined that cloud infrastructure was a substantial and sometimes hidden cost for AI businesses.

In 1999, Ben Horowitz and Andreessen founded Loudcloud, one of the first companies to provide computing, hosting and software services for consumer-facing enterprises. Unfortunately for them, the “dot-com bust” began a year later and the partners abandoned their cloud-based model and pivoted the company into a software solution, renaming it Opsware. They ultimately sold it to Hewlett-Packard Co. for $1.6 billion.

“The good news is we had this idea of cloud computing,” Andreessen said later. “The bad news is we were 10 years too early.”

On the basis of the Andreessen partners’ latest research about the cost of cloud, the hyperscalers will now have to decide if the prediction about margin and workload is imminent, 10 years down the road — or simply nothing to worry about at all.

Image: geralt/Pixabay