The following remarks were delivered by Mark P. Mills at a recent Manhattan Institute forum on his forthcoming book, "Digital Cathedrals' (Encounter Books, January 2020).
My plan today is to paint a picture of the technological landscape in the near future.
Let me start with the past. In 1913—106 years ago—the world witnessed an architectural marvel here in New York City, at 233 Broadway: the completion of the Woolworth building, a 792-foot-tall skyscraper, then the world’s tallest. It would hold that distinction for two decades.
That building epitomized the convergence of technologies that propelled the 20th century. Skyscrapers became possible because of the maturation of the then-new technology domains of steel, steam, and electricity—and the car. Electricity made possible both the subway and the elevator, and the automobile expanded the reach and density of traffic, which enabled the economic concentration of skyscrapers.
On the day the Woolworth building was christened, the New York Times called it a “cathedral of commerce.” People then were in awe of such a building because it had taken 600 years to surpass the previous record height for a habitable structure. Until 1913, no building had topped England’s Lincoln Cathedral, with its 524-foot tower—completed in 1311.
It bears noting that the Middle Ages marked civilization’s first epoch of industrialization. The gear, pulley, water wheel, and windmill—and, later in that age, the clock—propelled unprecedented economic growth and human wellbeing. The technological prowess and wealth of that era enabled the construction and proliferation of the world’s great cathedrals. Quite aside from their theological roots, those cathedrals were powerful economic focal points, not least because they served as hubs for non-religious commerce.
Today, we’re witnessing another rare pivot in the nature of economic infrastructure. And it is again associated with an entirely new class of building, this one called a data center—though I prefer the label “digital cathedral.” Like previous pivots, it is enabled by a confluence of technologies, this time of silicon, software, and photonic highways.
But, in a remarkable break from the iconic infrastructure of earlier eras, digital cathedrals are almost entirely out of sight, essentially invisible. This feature has, I believe, helped fuel a popular belief that the digital revolution somehow promises a kind of dematerialization of our economy—that the magic of cyberspace and virtual reality has de-linked economic growth from hardware and energy use.
In order to appreciate just how wrong that idea is, it’s helpful to use metrics that aren’t themselves invisible. Instead of counting the bytes of digital traffic, we can view digital cathedrals using physical metrics relevant to skyscrapers: specifically, their total number and their size in square feet.
The world’s biggest data center, near Reno, Nevada, has twice the square footage of the world’s biggest skyscraper, the 3,000-foot tall Burj Khalifa in Dubai.
The ten biggest data centers in the world today contain more square footage collectively than do the ten biggest skyscrapers. And more than 5,000 enterprise-class data centers are operating in the world, compared to 1,500 enterprise-class “office centers”—i.e., the Woolworth-class of skyscrapers. Smaller data centers number some 8 million.
Globally, there’s at least ten times more square footage under construction for data centers than for skyscrapers.
Building both skyscrapers and digital cathedrals costs roughly the same per square foot. For those steeped in the key measure of merit for buildings in cities: a square foot of a data center rents for five times as much as a square foot of skyscraper.
And, coming back to energy, each square foot of data center inhales 100 times more electricity than does a square foot of skyscraper.
Given these facts, one might not be surprised to learn that, collectively, digital cathedrals consume about 30 times more electricity than do the 5 million electric vehicles in the world today.
The infrastructure analogy with skyscrapers also extends to the transportation systems integral to the economic power of digital cathedrals. Of course I’m referring to the invisible traffic of bits moving through underground fiber-optic cables and over ethereal wireless networks.
In another example of the astonishing scale of this new infrastructure, consider that the world’s physical highways collectively span some 20 million miles—a distance that would get you halfway to Mars. Meanwhile, the so-called “information superhighway” has an effective aggregate length of 100 billion miles. That’d take you to Mars and back 500 times.
Coming back to the energy metric—because, in the universe we live in, all things and all traffic consume energy—the energy appetite of the digital transportation system roughly matches the amount of electricity used by all the world’s data centers.
Thus, given the recent fad of flight-shaming, it’s useful to note that digital cathedrals use more energy than global aviation—and they’re growing faster, too.
There’s really no mystery to what’s causing all this digital construction and traffic. We see it over these last few decades in the myriad businesses that have been enabled by, or disrupted by, the digital infrastructure. We see it in the wealth of the tech titans, and in the angst over the social and political disruptions from e-commerce and social media.
What comes next? What will drive yet further expansion of this new infrastructure?
Thus far, digital disruptions have nearly all been associated with the news, advertising, finance, entertainment, and communications industries—all information-centric enterprises and thus the ones most readily invaded by new, information-centric tools.
But all those businesses collectively comprise less than 20 percent of our economy. We should expect far more disruption ahead. In fact, we’re not at the beginning of the end of the expansion of digital infrastructure, but rather at the end of the beginning.
Despite my analogies to physical-world metrics, it’s the quantity of data in transit that remains the key measure of digital cathedrals—that’s what drives the build-out of the infrastructure.
To glimpse what’s coming, consider the velocity of what’s already happening to data traffic. Nine years ago, the consensus forecast was that mobile traffic would increase 90-fold by 2020. What’s actually happened is that traffic will have grown nearly 200-fold by next year.
Six years ago, Google held a conference entitled, “How Green Is The Internet?” Google now reports its use of electricity has at least tripled since then. The company is far from alone; this is typical for all players in digital infrastructure.
Today, the consensus forecast for the coming decade is that data traffic will grow at least another 100-fold, and that’s obviously on top of a far bigger starting point.
There is no other commodity in commerce that has or can experience this kind of growth. That’s because data is totally unlike its natural-resource analogues—engineers literally create more of it at an ever-accelerating pace.
There are a few reasons to believe that forecasters today are again underestimating the future expansion of digital domains. The first is the fact that the information superhighway itself is about to enter a new phase. It’s a transformation, to use a transportation analogy again, akin to what happened in the early 1950s when we went from narrow, meandering roadways built on horse paths to superhighways.
For once, the advertising hyperbole is spot-on: the far greater traffic capacity and speed of emerging 5G wireless networks really is a big deal. The potential of this radically new network is amplified by the emerging capabilities to add wireless connections to things—not just people—and even the components within things. This so-called Internet of Things will also include biological things in nature, as well as inside our own bodies.
AI
And we’re also witnessing a contemporaneous revolution in the logic engines that consume and create data at the core of the digital cathedrals.
Computers were invented to, well, compute or calculate. A profound difference exists between software that calculates and software that estimates an outcome based on continuous awareness of and responses to physical events in the real world, in real time. Of course I’m referring to artificial intelligence, or AI.
AI is so different from conventional computing that it has spawned—and actually requires—a new class of logic engine. Instead of the central processing units that have dominated computing to date, we now have classes of silicon chips variously called graphic-processing units, or tensor-processing units, or neural-network chips.
This new AI–class of silicon engine is optimized for “machine learning,” not for calculating. To extend the transportation analogies, the capabilities and implications of AI engines are analogous to what happened to aviation once the propeller was superseded by the jet engine. AI-class logic engines are forecast to account for 80 percent of all semiconductor sales in the next half-dozen years.
Stock-market traders are already eagerly pursuing the power of AI in the Sisyphean hope of beating the “invisible hand” of the market. But for most citizens, AI’s benefits will mean such things as better forecasts of natural events like weather and earthquakes and far better medical diagnostics.
AI will, eventually, profoundly alter most services. Most of you know that Uber and Amazon, for example, already use early-stage AI to optimize their platform services. It also means that manufacturing and supply chains will not only see meaningful gains in productivity, but also, more consequentially, the ability to design and create entirely new classes of materials or machines.
It’s likely, though, that it will be in health care where we’ll see the biggest impact of AI-infused digital cathedrals. Health care, overall, hasn’t seen meaningful productivity gains in decades. Progress has been difficult to achieve because biological and human systems are so complex. Even though computing has become far more powerful, we haven’t had sufficient power to meet biological challenges—until the advent of AI.
AI will also enable researchers to plumb the depths of nature and develop therapeutics by simulating molecular behavior in silico—instead of in humans volunteering as guinea pigs.
The explosion in the scale of health-related data will be epic.
One can measure AI–driven discovery and economic growth in energy terms.
For example, the learning phase to develop a single, simple AI application can consume more energy than 50 cars a year. And, the number of applications for AI is unlimited.
For an example at the heavy-lifting end: a single research team running in silico simulations for drug discovery entails machine-use intensity that consumes the energy equivalent of flying a jumbo jet to Asia. Of course, in the future, AI will become ten- to 100-fold more energy-frugal. But that efficiency will merely propel tens of thousands of similar AI-centric applications and activities.
Facebook has already flagged AI as, in its words, the “major culprit” in the annual doubling of its overall data center power use. And that’s just to deploy today’s nascent AI to perform economically useful but, truth be told, relatively trivial applications of AI in social media and advertising.
At a fundamental level, the emerging Cloud of digital cathedrals is as different from the Internet infrastructure that preceded it as air travel is different from automobiles. Forecasts that see, in just a few years, over $4 trillion spent annually in global AI–derived businesses could easily be underestimates.
Consider the analogous forecasting circa 1984, at the last pivot in the computer-communications infrastructure: back then, no one forecast companies and services like Amazon, Uber, AirBnB, Facebook or Twitter, much less the associated infrastructure scale we have today.
Energy
Given the nature of our energy debates, many Cloud companies are offering opinions about how society itself should be fueled, never mind its data centers.
Early in 2019, Google touted that it had achieved two years of powering its operations 100 percent with renewable energy. This is not to single out Google—many other tech firms make similar claims or have such aspirations, not least Apple and Facebook, and now Amazon—but such a claim is simply not true. It’s based on paper credits—on purchased indulgences—not on actual energy use.
Data centers and the Cloud can no more run on wind and solar than an aircraft can fly by burning wood. Energy must be delivered in the correct form. And when it comes to digital cathedrals, the correct form is “always-on.” The digital infrastructure is in fact fueled, on average, the same way that the rest of society is: with roughly three-fourths of the energy coming from hydrocarbons.
The entire ecosystem of digital cathedrals already consumes about 10 percent of global electricity. Many forecasts see that share doubling in the next two decades. It is indisputably the case that the Cloud has emerged as a major, fast-growing, hardware-centric, energy-consuming infrastructure—something no one imagined a few decades ago.
Closing Thoughts
All this makes a lie of the contention that society is de-materializing, or even de-industrializing. It makes a lie of the idea that economies will soon glide effortlessly toward a more digital, bit-dominated nirvana in a future with fewer hardhats, fewer mines, and less energy.
The cathedrals of the Middle Ages, last century’s cathedrals of commerce, and the digital cathedrals of our era are all anchored in the tenacious grip of the physics of big, atom-centric, energy-consuming hard stuff.
Now, as the Cloud, like all infrastructures before it, evolves into an increasingly “critical infrastructure,” policymakers and regulators will be ever more tempted—or enjoined—to engage issues of competition, fairness, and social consequences, as well as claims of abuse of market power, whether valid or trumped up.
And that’s where the virtuality of digital cathedrals intersects with even more challenging realities—those in the world of politics.
Mark P. Mills is a Senior Fellow at the Manhattan Institute and Faculty Fellow at the McCormick School of Engineering, Northwestern University.
