Software Eating The World, Revisited
By Victor Bennett, Ph.D. and Todd Hall, Duke University
Almost a decade ago, venture capitalist and Netscape-founder Marc Andreessen published an essay in the Wall Street Journal suggesting that “Software is eating the world.” The point? Software companies were making real changes to how we live our daily lives, and the market should value them as such.
That idea seems almost quaint now, in an era with major concerns about whether software companies like Facebook have an outsized influence in our daily lives; and whether software naturally leads to bigger companies, lowering the wages of — or completely replacing — human workers, particularly the most vulnerable.
Given the major change in how we interpret the idea of software eating the world, it seems like a good time to check in on the facts.
By our reading, Andreessen’s prediction encompassed four ways in which software was “eating the world.”
- The first is traditionally-non-software firms moving their business to software, which Andreessen exemplified by HP’s plan, at the time, to distance itself from its PC business.
- The second is software companies acquiring non-software firms, as illustrated by Google buying Motorola Mobility.
- The third was non-software companies making software central to their business, reflected in Wal-Mart’s logistics.
- Finally, there are non-software companies buying software companies, such as by Disney’s acquisition of Pixar.
We took a rough initial cut at some data to see how each of these has tracked over time.
1) Non-software companies increasing investment in software
The first example Andreessen gives is HP’s plan to explore the possibility of distancing itself from its PC business.
To measure how common this has been over the years, we start by looking at the increase in expenditure on software by non-software businesses. We do this by looking at data from the Information and Communication Technology (ICT) survey administered by the U.S. Census Bureau, and excluding SIC industry code 51, which indicates information companies. Figure 1 shows that overall expenditures in millions of dollars are increasing, and that this increase is driven largely by increases in software licensing costs instead of development. This seems to suggest that non-software companies are indeed allocating more dollars to software, and that much of that is driven by spending on software built outside the company.
Figure 2 breaks out these non-capitalized expenditures by sector. Interestingly, there are some sectors where we’re seeing declines in overall software expenditure, but big increases are happening in finance, a sector that Andreessen highlights as being “eaten.”
Moving away from the world of dollars, we wanted to see if non-software companies were developing more new ideas in software. To do this, we wrote some software to read through companies’ patent descriptions looking for indications that a patent was a software patent (for those interested, we followed the procedure in Arora, A., L. G. Branstetter and M. Drev (2013)). We then measured the average count of software patents by non-software companies by year. Figure 3 shows that this has been increasing dramatically over time. For the statistically-inclined, this graph’s shape holds up when we plot the regression-estimated count controlling for firm-fixed effects.
2) Software companies acquiring non-software firms
The second category Andreessen mentions is acquisitions of non-software companies by software companies. We used the SDC Platinum database of acquisitions to look at the percentage of acquisitions by software firms that were of non-software firms. We defined software firms as those with a NAICS industry code of 737, indicating Computer Programming, Data Processing, and other Computer Related Services. In Figure 4, we see that the raw number of acquisitions of software firms by software firms has actually been decreasing since the bubble, and that the percentage of all acquisitions by software firms that are non-software companies are declining steadily.
3) Non-software companies centering their business around software
Andreessen’s third category is non-software firms beginning to center themselves around software. To get at this, we look at companies’ descriptions of themselves in their 10-Ks over time. Figure 5 plots the percentage of firms using the term “software” in their description of their business.
The bump around 2000 should be taken with a grain of salt because many businesses used the term “software” in the “Risks” section of their 10-K describing Y2K compliance. Even omitting those years, though, we see a steady upward trend.
This pattern could be driven by the changing population of firms, i.e. more firms that always used software, or by firms that didn’t use software beginning to center themselves around software. Interestingly, here we are able to see firms whose original descriptions of their business didn’t include software begin to use the term “software” in their self-descriptions.
There are some interesting examples of firms that begin to talk about software after being well established. Arconic is a producer of lightweight metals such as nickel and titanium. Outside of year 2000 compliance, Arconic did not mention software in its 10-K filings until 2002 when it began noting the significant role software plays in its productivity improvement and cost reduction efforts. Arconic has mentioned software in every 10-K filing since 2002. Angelica Corp. provides outsourced linen management services. The company did not mention software in its first eleven 10-K filings in our sample outside of year 2000 compliance. However, in 2004 Angelica Corp. began mentioning software as an important part of route optimization in each 10-K filing. The pattern in figure 5 continues to hold up even after we remove firms that have always described themselves using the term “software.”
4) Non-software companies acquiring software firms
For the fourth category, we return to our data from SDC Platinum and our definitions of software and non-software firms based on industry code and look at the propensity of non-software companies to acquire software companies. Figure 6 is the analogue to Figure 4. Here we see a dramatically different pattern. Even omitting the bump around 2000, we see a steady increase through the beginning of the decade of acquisitions of software companies by non-software companies, in both raw numbers and percentage. Since then, both count and percentage are relatively stable. Both in leading up to Andreessen’s article, and since, there have been a number of high profile acquisitions of software companies by non-software companies aiming to build significant software capability within the firm including Wal-Mart’s acquisition of Jet.
In summary, the data supports software becoming more important across industries.
Firms that aren’t primarily software companies are more likely to bring up software in a short description of their business. They are also spending more on licensing software and are producing software patents at unprecedented rates. In the twenty years leading up to Andreessen’s essay there was an increase in acquisitions of software firms by non-software firms, though that has tapered off in recent years. While we have some anecdotal evidence of acquisitions by software firms of non-software firms, it doesn’t seem to be terribly common, nor is it becoming more likely.
All in all, while it may be hyperbole to say that software is “eating the world,” it certainly seems to be becoming more important across sectors, primarily in finance.
It is important to understand what shifts toward software will bring. There are certainly many positive effects of the increasing prevalence of software. Automation can drive down costs, which, in some settings, can translate to lower prices. Some recent work suggests that when software becomes available for automating an industry, entrepreneurship may go up.
Those benefits do not come without costs though.
The same study that shows that software increases entrepreneurship also suggests that the new firms may be displacing the existing, less efficient firms. That may be good for productivity, but less for workers at those firms.
Other studies have suggested that automation can reduce wages and drive inequality by replacing workers even at the firms that survive. Furthermore, others have argued that a shift toward software might not only affect wages and inequality directly through workers being replaced, but through a second channel as well. If software makes scaling up cheaper, it may lead to naturally fewer and larger firms. This can lead to workers having fewer options of where to work, possibly also pushing down wages and driving inequality. To make matters worse, fewer and larger firms could potentially raise prices as well. Furthermore, machine learning and artificial intelligence are increasing the ability to automate tasks that were not previously at risk of automation, potentially spreading the impacts to more industries.
In light of these effects, the possibility of software “eating the world” seems much more like something that the public and private sector should collaborate on preparing for — even as it unlocks new productive capacity and potential.
Victor Bennett, PhD is a friend of Initialized Capital. He’s faculty at Duke’s Fuqua School of Business, former White House Council of Economic Advisers Senior Economist for Technology and Competition Policy, and an Ex-Googler.
Todd Hall is a doctoral student in Strategy at Duke and an alum of Epic Systems.
