“How Green Is The Internet?” was the title for a Google-sponsored Summit early this summer. Al Gore was the keynote, which, along with the title should tell you something.
No longer at issue is where the world’s energy comes from to drive and fly to conferences, or to light up homes and shopping malls. Now on the table is the energy needed to light up silicon and antennas. The Internet itself and especially its proliferating warehouse-scale central computers are so enormous that the associated energy use is as meaningful as that for other infrastructure sectors.
The global information-communications-technology (ICT) ecosystem consumes more energy – likely at least 50% more – than does global aviation. Both are growing, but ICT energy demand is growing far faster. And while aircraft use oil, computers use electricity.
U.S. Share of GDP
Since all bits are electrons, it should be unsurprising that creating, transporting, using and storing data at the zetabyte level consumes terawatt-hours of electricity. Perhaps what surprised, still surprises, many in Silicon Valley is the (inconvenient?) reality that during the past decade’s unprecedented growth in the Internet, nearly 70 percent of the world’s new electricity came from coal, and at least half will come from that hydrocarbon for the next decade (this according to the International Energy Agency, not the coal industry). Hence the title of our new analysis, The Cloud Begins With Coal: Big Data, Big Networks, Big Infrastructure, and Big Power.
But for investors the more interesting question is not how much green electricity the world might produce, regardless of what it is used for, but how much more electricity the ICT sector will use, and why.
Think of the kilowatt-hours hidden behind consumer and business activities as the digital exhaust of the global ICT ecosystem. Energy used for computing and communications is a direct measure of the scale and growth of ICT technologies. And the associated energy engineering challenges – what it takes to power everything from CPUs to handhelds and from cell towers to data centers – are where we find a deep well of opportunities for innovation, investment, profit and disruption.
While Apps, the Cloud, big data analytics and sensor-driven datafication of everything are entering a hot cycle, so too is everything behind the scenes associated with the infrastructure hardware that makes all that possible and affordable. IDC forecasts $3 trillion in global ICT infrastructure spending for the coming decade, all of which uses electricity. Mapping out the energy features of using that hardware — the digital exhaust – reveals specific technology challenges and investment opportunities.
Electricity in the Digital Ecosystem
For simplification purposes, we can summarize this macro analysis by grouping the above map into four main (energy-consuming) ICT domains: data centers; wired and wireless networks; end-user equipment; and the factories that manufacture all the ICT hardware.
In the following, looking through the lens of ICT energy use for each of the four domains, we identify one key driver or surprise that has investment implications.
Data centers sit at the core of the Internet and comprise the heart of the emerging Cloud architecture. While data centers collectively represent only about one-fifth of all ICT-centric energy use they are a magnet for attention and kilowatt-hours. Single enterprise-scale data centers are now commonly the size of shopping malls, but consume 100 times the electricity per square foot. Facebook, despite industry-leading efforts in data center efficiency, reported a 33% one-year rise in electric use.
A new global survey by DataCenterDynamics (DCD) found cost and availability of energy for data center operators ranked as the number one concern, as did a recent survey by European commercial real estate firm Jones Lang LaSalle. The DCD team points out that the collective electric use of all data centers would, if it were counted as a country, rank 12th in the world. Three years ago, that digital-power ‘country’ would have been 22nd on the list of nations.
And in just three more years, DCD forecasts global data centers will use 25% more energy. A typical data centers costs roughly $7 million per megawatt to build, and another $9 million per megawatt for the cost of electricity over the facility’s 10-year operating life (assuming low-cost power). While the cost of ICT hardware is falling, in most of the world electric rates are rising. Accordingly, there is a near obsession in the industry to chase efficiency to cut energy used per gigabyte, and chase cheap power to cut operational costs.
Data center operators increasingly face the same operational and thus capital spending challenges that electric utilities do – building to manage peak demand.
Data Center Traffic
Power plants and data centers are similarly capital-intensive, and they are similar in at least two other ways: the need to build sufficient capacity to meet peak, not average demand; and the need to supply the ‘product’ essentially contemporaneous with when it is demanded. Thus, in both cases, efforts to maximize, or levelize, utilization of a capital asset are stymied by the nature of consumer behavior which creates peaks and valleys. Yes, software makes the oscillations more predictable. But that’s different than making it less peaky.
Data traffic at peak is already twice as high as average traffic. (See the Cisco Video Networking Index.) And while average traffic is forecast to roughly double in the next five years alone, peak traffic is forecast to rise nearly four-fold. It is unlikely that data markets will find it feasible to employ pricing solutions akin to those employed by aviation to manage its peak traffic. The solutions will come mainly in hardware and software to accommodate – not suppress – both peak data demand, and the associated peak energy demand.
Telephony (a straight extension of telegraphy) constituted the first disruptive era of communications, from globe-spanning transatlantic cables to the local loop. The second era came with high-bandwidth optical fiber ‘wires’ enabling distributed (personal) computing – the wired broadband Internet. The third communications era came with the untethered and now ubiquitous cell phone. Now we enter the fourth disruptive era — the wireless, mobile broadband Internet.
The beginning of the mobile Internet was marked by the 2007 introduction of Apple’s iPhone, followed quickly by the iPad, which together igniting a new hypercycle of products, competitors, service, and software from Apps to situationally-aware computing. Total mobile traffic is up some 400-fold since 2007.
Growth in mobile data traffic has stunned everyone in the industry. Global mobile network traffic is expected to rise at least 10-fold in the next five years alone. In the physics of energy, delivering wireless bits is radically more energy-intensive than using wires and fibers. And expanding bandwidth on wireless networks is particularly costly in energy and engineering terms.
Compared to the now world-dominant voice-centric 2G cellular networks, the new broadband LTE networks use 60 times more energy to offer the same coverage. But with the latter, users get real-time streaming video where they formerly were restricted to voice and text. Notably, almost every analysis of ICT energy use is anchored in pre-iPhone era data or assumptions.
Energy use is rising more rapidly in mobile networks than anywhere else in the ICT ecosystem, and is emerging as a greater challenge than in data centers.
At least one major cellular network operator has recently published data on how mobile broadband impacts energy use. China Mobile Research Institute points to a 50% improvement in the energy efficiency of China Mobile’s network over the past half-dozen years. But contemporaneous with the advent of broadband wireless, that network system electricity use grew far more rapidly than the rise in the number of subscribers and base stations. The number of base stations doubled and energy use rose 600%.
Electricity used in homes in computing-related devices already exceeds the combined use for cooking, and washing clothes and dishes. Residential computing energy is on track to exceed that for lighting. And as lighting becomes more efficient, total energy use for lumens (at least in mature markets) may well fall. But as computing becomes more efficient, total energy used for ICT-related activities and services has risen and will rise in all markets.
If the accounting were done properly by including digital TV, residential digital energy demand already exceeds illumination. A single TV set-top box can use as much electricity annually as a modern refrigerator.
The migration of TV to the Internet will dominate ICT end-use energy for years, and creates daunting technical (and energy) challenges especially for wireless networks and data centers.
The share of TVs that are digital will reach 50% by 2020. Television’s migration into the Internet, and the expansion in general of video, represents a tectonic transformation. And shortly, yet another video growth cycle begins as the next era of (energy-hungry) displays emerges in the form of glasses-free 3D, and as well wall-scale displays. DisplayWalls are poised to move out of the rarified worlds of research and military domains and Hollywood imaginations, to consumer and commercial markets as costs plummet (and the non-trivial technology challenges succumb to innovators). The data traffic and energy implications are unprecedented in the Internet ecosystem.
The energy needed to manufacture one PC is about the same as that used to fabricate a refrigerator. But annualized, the energy to fabricate a PC is three to four times that of a refrigerator because the latter is used three to four times longer. Even enterprise ICT hardware is replaced frequently. An Uptime Institute 2012 Data Center global survey found operators rank rapid adoption of new tech as a higher priority than product lifespan by a 71% to 26% margin. The faster ICT products become obsolete, the greater the energy used in manufacturing. The ICT obsolescence rate is, if anything, accelerating.
Semiconductors dominate energy in the manufacturing of digital hardware. It takes 1 – 2 kWh to make a square centimeter of microprocessor in a world that produces hundreds of billions of square centimeters of silicon chips. And at the end of the useful life of the silicon device, there is essentially no inherent material or energy value left – it’s trash. The embodied energy can’t be recycled; it has been consumed.
The torrid rate of adoption of new wireless broadband services, big data analytics and the Cloud architecture are driving a new cycle of capital investment, and it is transferring much of the energy cost of the digital ecosystem from end-use operations into manufacturing.
Cellular Base Station
Even in torpid Europe, commercial real estate firm Jones Lang LaSalle found (in the same survey noted earlier) that the data market is a bright light of economic activity. The pursuit of new, more complex, feature-laden and in particular, more efficient ICT equipment drives demand back to manufacturers across the digital domains.
In effect, an energy trade-off is made: the energy embodied in manufacturing is swapped for energy used in operations. The operator pays for the swap in capital, motivated by the newer device that generates higher revenues (especially new broadband services & capabilities) and that offers lower operating costs from the more efficient new equipment. But through the lens of total ICT system use, overall societal electricity demand rises. But instead of rising at the point-of-use, it rises in the factories (measured then as “embodied” energy in the devices).
In theory, over time the new more efficient equipment’s operational energy savings of, for example, a cellular base station can repay the energy debt embodied in manufacturing that hardware. But odds are high, however, that the new equipment will end up being obsoleted and replaced long before the operational efficiency pays for the manufacturing energy debt. It is a cycle that is bullish for manufacturers, consumers, and investors.
The Next Investment Hypercycle
The implications of the trends briefly highlighted herein have not gone unnoticed in scientific circles, even if they’ve been largely ignored and are even unfashionable in Silicon Valley. While business and media markets were pre-occupied with watching green-tech investments, both the (few) successful ones and the prominent failures, the National Academy of Sciences published in 2011 a seminal report — The Future of Computing Performance: Game Over or Next Level? — examining the energy implications of computing trends. That study’s core conclusion was certainly a clarion call for tech opportunity:
- “As the use of the Internet continues to grow and massive computing facilities are demanding that performance keep doubling, devoting corresponding increases in the nation’s electrical energy capacity to computing may become too expensive.”
Epitomizing widespread investor cognitive dissonance on ICT realities we find storied tech investor John Doerr and his venture fund, Kleiner Perkins– Al Gore sits on Doerr’s Board. With a stellar reputation anchored in early and successful bets on ICT start-ups, over the past decade Doerr’s fund famously lost a lot of money in green energy tech.
Doerr’s team chased the same chimera as did many others in Silicon Valley; the hoped-for equivalent of a Moore’s Law for energy. There isn’t one, and while a subject for another day, we note for the record that in the physics of energy there is, in effect, an inverse to Moore’s Law. (For more on energy realities, see my co-authored book The Bottomless Well.) Of the tens of billions spent by Silicon Valley investors, and billions more spent by the government over the past decade on green tech, the vast majority failed to yield anything useful at the scales promised, or make anyone any money.
The point isn’t to single out Doerr, as he was far from alone in the dream of a radical restructuring of the energy-supply ecosystem. Whether the world will one day, many decades hence, become far less dependent on hydrocarbons, is a moot point in terms of the time frames of interest to nearly any investor. For the record, hydrocarbons supply 85% of the world’s energy, a ratio that will change little for at least two decades.
But roughly contemporaneous with the above-cited National Academy report (perhaps no coincidence), Doerr brought onto his team the brilliant Mary Meeker to oversee a return to what Silicon Valley is so good at delivering: new ICT companies. Meeker’s annual state of the Internet report is a tour-de-force in mapping the key demand side-metrics for the rapidly evolving ICT ecosystem.
Much like the dawn of the Internet’s first hypercycle circa 1993, the investment opportunities now range across the entire spectrum from big companies like IBM, Qualcomm and Cisco, to newer hot players like Splunk (which I wrote about earlier) and Teradata. And then there are of course the hundreds of start-ups blossoming in Silicon Valley, Silicon Alley and beyond. Many of them will end up at Qualcomm or even IBM scale in due course.
You can look at the massive transformation of the Internet now underway through the lens of energy demand, or of investment opportunities. The energy used will continue to be the digital exhaust, not the driving force. But when you put the pedal-to-the-metal, the roar of the exhaust is quite something.
Original Source: http://www.forbes.com/sites/markpmills/2013/08/11/how-green-is-the-internet-wrong-question-how-big-is-the-internet-and-where-do-i-invest/