Location Still Matters in the Era of Bits and Bytes: Interview with Shane Greenstein

Does distance still matter in the digital economy?

The question has been debated for decades, ever since journalist Frances Cairncross argued in the book The Death of Distance that electronic communication would erode the importance of location in business decision-making.

Two research papers coauthored by Harvard Business School Professor Shane Greenstein indirectly revisit the question. This time, he offers fresh evidence based on separate analyses of the construction boom in data centers and the forces shaping investments in new technologies.

In “Where the Cloud Rests: The Economic Geography of Data Centers,” published by the journal Strategy Science in December, Greenstein and Rice University’s Tommy Pan Fang analyze how companies choose the locations for the data centers that form the backbone of the modern digital economy. Their conclusion: Geography still matters in the era of bits and bytes—but in sharply different ways, depending on the business model. Specifically, they found that:

• Third-party data centers tend to be located in densely populated areas to remain close to customers—particularly in finance and other information-intensive industries where latency and reliability are the highest priorities.

• Cloud data centers are usually found in more remote and less densely populated areas, where they can take advantage of an abundance of energy and water.

Meanwhile, in the August working paper “New Economic Forces Behind the Value Distribution of Innovation,” Greenstein teamed with Stanford University’s Timothy Bresnahan and Amazon’s Pai-Ling Yin to examine how technological change travels and creates value. The team analyzed 2,400 products that represented both incremental and novel inventions, comparing innovations for the digital markets with those in the local radio and newspaper industries.

They found that while web and mobile infrastructure spread widely across the country, the most valuable and novel inventions did not. Instead, those breakthroughs concentrated in a handful of locations, with 96% of their value accruing to Seattle, San Francisco, Los Angeles, and New York, and the media, entertainment, and retailing fields.

Those regions offered the right mix of attributes to foster novel innovation, including access to specialized skills and collaborative, learn-from-each-other environments. In contrast, incremental innovations were far more geographically dispersed, with roughly a third of their value generated outside those major hubs.

We talked with Greenstein, the Martin Marshall Professor of Business Administration, about his research in an interview edited for length and clarity. Here’s what we learned about the role of location and innovation.

Location has always been a digital consideration

“It's a revelation to many people to think about how the labor markets for technical talent or domain knowledge work in other places, or how organizations relocate their capital to take advantage of different situations. Many people just never have to think about this.

I remember reading about Google and Facebook putting up their first data centers in Oregon, I recall thinking: What’s going on there? Both companies were facing a problem that no one had had to face before, where they were connecting everybody.

If every person (on Facebook) has 90 or 120 or 500 friends, when you scale that up to everybody, the draw on computing resources turns out to exceed what anyone had ever tried to engineer before. It means millions of pieces of information have to be kept in virtual memory. Nobody had ever put that much information instantly into a computer for instantaneous draw (of power). They were at their technical frontier.

The only other firm that had done anything similar was Google, which had also faced a similar problem, but on a smaller scale.

For the architecture, you had to make this very large data center and you could only draw pieces of it at a time. You would have to fool the user and sometimes delay a bit before they could draw too much. I remember looking at that and thinking, ‘Gosh, what an interesting design for the problem they face.’”

AI changes how companies think about proximity

“AI algorithms take anywhere from a couple of days to a week (when employed to train new AI models). So there's no compelling reason to build a data center geographically co-located with customers.

Once you're free of that constraint, then other service dimensions become more salient to the buyer. And that generates demand for hyperscale data centers because they can handle larger data sets faster.

In the US, sufficient energy to run a hyperscale data center turns out to be the primary constraint and the primary determinant of where a lot of these centers are locating. It's often (HVAC) talent, the people who keep the electrical generators running and those kinds of professionals. But hyperscale centers don’t need a huge number of these people.

It's the same set of things. The checklist is the same. It's just the rank is different.”

Companies still need to be close to talent

“If AI follows the patterns we saw in the previous generation of technology, Web 2.0, you would expect those with novel aspirations to look for the rare resource. And the rare resource in AI is technical talent for sure right now—definitely favoring San Francisco and Seattle, and New York to some extent.

You would also expect that domain knowledge and AI services are going to be constraints in determining where people develop some of the novel services for AI.

If the growth area is in media again, then LA and New York have an advantage. But there are applications in other areas, too—logistics, science, engineering, architecture, design, and chemistry. And that could take AI development into different locations. It’s about the domain that can take advantage of the opportunity.”

Choose your company’s location wisely

“A real fundamental question to ask right away is if geography is something that can be changed. For many businesses, if you're in a service business, you're not changing that.

So if that’s not going to change, then that requires the manager to have a sober analysis about what is, or is not, possible. Because you can reach the rest of the world, but you still have to produce it where you are. And when you're reaching the rest of the world, you're in competition with other people who produce it where they are. And they may have more resources at their disposal.

And the point is, you may have an advantage based on where you are. And so you have to be very sober about that. That's something I really came to appreciate.”

Weigh the tradeoffs of incremental and novel innovation

“First, who's the customer you're aiming at? If it's for existing customers, almost always, incremental is a better way to go. If it's for a new customer, then that’s an open question.

Another consideration is risk tolerance for the manager and risk tolerance net for the firm.

Incremental projects tend to be shorter and much less expensive, whereas novel projects almost always take a long time. They have lots of unexpected twists and turns to them.

Another one is how big a team can you draw on internally or in your labor market to get things done? And when it comes to domain knowledge and technical knowledge, what kind of team do you need to get a project going?

For example, consider a chatbot in a service environment for a small e-commerce site. That takes someone with technical talent and it takes a while to put that team together.”

Novel innovations often raise existential questions

“Another thing you have to worry about is what you might call the ‘existentialism’ of the project. Are you going to be facing somebody who's trying to put you out of business? Or, if you do this, could you put yourself out of business? You have to think quite hard about the time and expense, and whether it affects your probability of being alive.

Tim and I looked at terrestrial radio. It was at the top of its game in the mid-90s when the internet showed up. And it was still at the top of its game through the dot-com boom. But Web 2.0 finally started to grab a lot of user attention and share.

Most traditional radio stations went online. We looked particularly at music, and many of those firms lost share to Spotify and Pandora. It's not like they didn't see it coming. There just wasn't any way for them to get from where they needed to be. They didn't have the technical talent, or they didn't have the money, or they didn't have the appetite for risk.”

AI innovation might be incremental now, but novel soon

“I expect the same thing for AI at an early stage anyway. Those easiest data-intensive activities are going to be the easiest places to apply technology because the firms already have processes for data in place.

The problem with AI is that large language models are not sufficiently reliable to use everywhere without humans being in the loop. For instance, AI speeds up coding, but humans still have to be there to do the architecture and check for errors.

An interesting example I heard was when Russia invaded Ukraine, it was obvious to every human on the planet that the price of grain was going to go up, because Ukraine is the source of a lot of wheat in world markets.

But no computer could make that prediction because there was no historical data, so you can’t automate that. But once you can set that level, it’s going to spill through the entire market. It’s a very subtle combination of human and computer expertise.”

Photo credit: Russ Campbell