Behold the demand revolution

Measuring the cloud’s clout

Arkansas Democrat-Gazette demand illustration.
Arkansas Democrat-Gazette demand illustration.

Everything begins with demand. There would be no need to solve the challenges associated with energy supply but for the fact that everything humans invent, build and do demands energy.

Silicon Valley claims to be revolutionizing energy demand by making homes, buildings, cars, washers, refrigerators, motors, tools, toothbrushes, pretty much everything, smarter. Is silicon intelligence combined with the Internet of Things (IoT) ushering in an era of peak energy demand? The IoT is a big deal, but it's not a demand revolution.

And Tesla? Its storied stock has vaulted the company's market value past Ford, although the latter sells 60 times as many cars and actually makes a profit ($10 billion last year) while Tesla burned a nearly $700 million hole in its balance sheet. The Tesla is an impressive car, but it's not creating a demand revolution. All things equal, physics dictates the same energy per mile whether burning a battery's electrons or barrels of oil.

Where's the revolution then? While predictable, it's not revolutionary when more people and wealth drive up energy demand. Real demand revolutions happen with the invention of something like the internal combustion engine or an airplane. Such inventions create new fuel demand that had never previously existed.

Virtually every class of product or machine in widespread use today was invented before World War II. The lone exception? Look no further than your pocket or purse. Just as Detroit democratized transportation a century ago, Silicon Valley has democratized information with a transformation that only began in earnest a decade ago. Both classes of technologies require vast infrastructures creating new energy demands. The cloud--a sprawling constellation of massive data centers and communications networks--is genuinely new and revolutionary.

And in energy terms, operating today's global digital infrastructure surpasses the fuel demand of global aviation.

It has taken the cloud two decades to reach a level of demand that it took aviation 70 years to reach. Only the dawn of the automobile age drove

as fast a rise in energy demand as has the cloud. Now that's a demand revolution.

The seeds of the new silicon demand revolution were planted in October 1957 when Fairchild Semiconductor opened its doors in Silicon Valley as the progenitor of the modern commercial transistor factory. And 30 years ago the second critical piece of the new demand revolution fell into place: when the Internet began its explosive growth from mere thousands to today's billions of users. Rounding out the third and final catalyst in the digital infrastructure triad, this year marks the 10th anniversary of the iPhone that democratized supercomputing in the cloud.

Almost daily we see media stories about digital-era products creating new services and upending old businesses. All these new companies and products--from Amazon and Uber to artificial intelligence-assisted diagnostics--are built on the vast new digital infrastructure. The more exciting or novel the services, the faster the growth in demand becomes for a bigger, faster infrastructure.

It may seem incredible to think that infinitesimally small transistors and ethereal bytes in wireless flight to smartphones could require more power than hauling humans at the speed of sound in leviathan aircraft. But that's where the law of large numbers comes in. As every student of zoology learns, bacteria collectively weigh thousands of times more than all the whales on earth.

The world fabricates more transistors each year than the number of grains--not bushels, but grains--of wheat grown globally. And those silicon devices, along with other complex components from lasers to optical systems, manifest not only in billions of consumer devices but in megatons and trillions of dollars of hardware in the hidden infrastructure of communications networks and warehouse-scale data centers. The data coursing through the world's wired and wireless networks is now countable north of two zettabytes, an incomprehensible number.

For those not steeped in the physics of information, this is what you need to know: All bytes are electrons or photons (electrons' quantum cousins). While mere atomic-scale energy quantities are used per single byte of information, a zetta is an astronomical number. A stack of two zettas' worth of dollar bills would go from the earth to the sun and back one million times.

The scale of the infrastructure that produces, massages, transports and stores all those bytes is well documented. It starts at the epicenter with thousands of data centers, many the size of a Wal-Mart, filled with hot humming computer memories and servers. A typical enterprise-class data center demands as much power as a utility-scale power plant can produce. Billion-dollar data centers are proliferating faster than billion-dollar oil platforms. Global capital spending on energy-using information-communication technology hardware is running at $3.5 trillion annually, rivaling global capital spending on energy-producing oil and gas equipment.

As the cloud enters the next and bigger expansion phase--where data and software are increasingly becoming utility services--we see the emergence of so-called hyperscale data centers. There are already 300 of them globally, with 400 expected by next year. One leading data center expert recently concluded: "Data center power is out of control. ... Google's total power usage seems to have gone up 12-fold in the last four years. ... Most of the rampant growth is caused by growth in consumer services like Facebook, YouTube, Netflix and messaging."

Next, consider the essential communication networks connecting the data centers with billions of mobile devices and soon hundreds of billions of things. Here it was the emergence of high-speed wireless technology that accelerated the early Internet tethered to desktops to today's planet-spanning wireless network.

And physics comes into play again as we migrate from wired to wireless. Wireless convenience is inherently far more costly in energy terms per byte. The new broadband networks use 60 times more energy to provide the same coverage as the earlier voice-centric 2G cellular networks still dominant around the world. (No surprise then that the electric bill for wireless operators can exceed 40 percent of operational costs.)

In aggregate, the networks demand far more power than do the data centers. And every forecast sees wireless bandwidth demand ballooning exponentially. Researchers at a Bell Labs project in the University of Melbourne, Australia, recently concluded: "Wireless networks are the biggest threat to the sustainability of cloud services, not data centers."

Finally, we add hardware manufacturing to the pantheon of hidden energy demands. Digital things are the most complex products ever produced at scale and require on average 1,000 times more energy pound-for-pound compared to typical products manufactured last century.

For all of history, the energy costs of manufacturing conventional products largely tracked the weight of the thing produced. A refrigerator weighs about 200 times more than a hair dryer and takes about 200 times more energy to manufacture. But it takes nearly as much energy to make one smartphone as one refrigerator, even though the latter weighs 1,000 times more. And the world produces 30 times more smartphones annually than refrigerators.

To use another example, despite a car weighing 10,000 times more than a smartphone, global fabrication of smartphones now uses 15 percent of the energy used by all car factories. And unlike the steel in fridges and cars, at the end of a short useful life a silicon device has no inherent or recyclable value. It's trash.

Thus, in energy terms smartphones are the SUVs of the information highways. Greenpeace has eagerly pointed to these energy realities to put PR heat on the tech community. Greenpeace's 2017 Clicking Green report found that the global cloud uses more than twice the electricity of the entire nation of Japan.

Apparently embarrassed by the scale of cloud energy demand, the tech community serially promises more efficiency will solve the problem. But it's not a problem; it's the sign of a wonderful new economy-boosting infrastructure.

And it is precisely because of astounding gains in energy efficiency that the Internet exists at all. A single Amazon data center, if it had to operate at the level of 1987 computing efficiency, would demand fivefold more power than does all of Manhattan. Without far better efficiency there wouldn't be a single data center, never mind thousands. The reality is every company in the digital ecosystem eagerly pursues big efficiency gains in order to expand markets for silicon logic.

While we're dispensing with shibboleths, we noted that electric cars actually don't change demand, nor do they cause a meaningful shift on the supply side either. Over the decade of Tesla's rise, about 70 percent of America's new electricity supply came from burning shale gas, effectively displacing the use of, well, shale oil.

As for the future, all trends point to explosive growth in global cloud traffic. Odds are that overall cloud power demand will double or more in a decade or so.

Meanwhile, even if optimistic forecasts for global EV sales are realized, two decades from now the cloud will demand far more power than do battery-powered cars.

But there is one way that future cars will help revolutionize demand. Making cars silicon smart and self-driving, bragging rights for Tesla and every other automaker, means automobiles become a new generator of data. Executives at Audi and chip-maker Nvidia talk in terms of cars as supercomputers or data centers "on wheels." The amount of data such cars will generate is countable in the zettabtytes per year.

Editorial on 05/14/2017

Upcoming Events