Power Use by Computers and Data Centers
Where possible, this discussion will use MKS power units - watts, gigawatts, terawatts and MKS energy units, joules and megajoules. Units such as kilowatt-hours per year and quads per year are confusing, and inherently "resource-depletion" oriented. I won't touch such units with an 87.66 kilofoot-hour per year pole.
Global Power Use
World power use averaged 15 Terawatts in 2008 according to Wikipedia . The US Energy Information Administration estimates world usage in 2010 at 509.7 Quads per year. A Quad per year equals 33.4 gigawatts, so that is 17.0 Terawatts. The world usage is extrapolated to reach 721.6 quads or 24.1 Terawatts by 2030, a growth rate of 1.75% per year.
88% of electricity is generated by fossil fuel, leaving 12% generation by nuclear and hydro, and a tiny fraction by wind, solar, and other alternatives. Generation is mostly coal and some oil, which averages about 33% efficient. We can estimate that the multiplication factor from electricity to primary power is approximately 3.0. The fraction of that which generates CO2 is 88%, or 2.68. Every gigawatt consumed by computers requires 2.7 gigawatts of CO2 producing primary power.
The 2010 CIA World Fact Book estimates global electricity usage for 2010 at 18.8 TKWH per year, or 2.15 Terawatts. This increased from 1.63 Terawatts in 2005, a 5.7% per year increase.
China is increasing electric power and computer usage much more rapidly. Electricity production has more than doubled in 5 years, from 186GW in 2005 to 392GW in 2010, an increase of 16% per year. China, India, and other rapidly emerging nations will soon dominate world usage, and perhaps increase the usage growth rate as well. We compete on the world market for oil, which is 40% of US power consumption. Increasing Chinese consumption, and the US debt to China, may reduce US oil imports and severely curtail US power generation.
Global Computer Power Use
David Sarokin estimated 2007 computer energy use at 868 billion kilowatt hours per year, or 99 gigawatts. 5.3% of global electricity consumption. Forrester estimates a doubling of PCs in 7 years ( 10% per year). The DOE estimates US data center power usage as a fraction of total US usage (increasing 1.4% per year) will double every 5 years (15% per year), for a total power increase of approximately 16% per year. The average power usage increase might be 13 percent.
In 2010, we can estimate the computer power use to increase by 44% (1.13^3), so the power in 2010 is approximately 140 GW of electrical power, corresponding to 380 GW of carbon primary power in 2010.
Comparison to Aviation
For comparison, Schafer estimates that aviation used 2.5% of the world's primary energy in 2005, and that this will increase to 9% of the larger total by 2050. If this is accurate, the fraction increases by a factor of 1.0289 a year, and extrapolates to 2.88% of world energy usage in 2010. Times 17.0 TW, that is 490 GW of primary power for aviation in 2010. The increase in power, rather than power fraction, is multiplied by 1.0175, so aviation power increases by 4.69% per year.
If aviation power increases 1.0469 a year, and computer power increases 1.13 a year, computers are overtaking aviation by 8% per year. In 2014, computers will use 620GW of primary power, and aviation will use 590GW. By 2020, computers may be using 1290 GW of primary power, compared to aviation's 770 GW.
Limits on Power Production
Although the earth intercepts 170,000 TW of solar power, only 89,000 TW makes it to the ground. Of that, perhaps 20,000 TW is intercepted by plant life, and converted into plant-usable power by 3% efficient photosynthesis. Of the resulting 600TW, 2/3 is used by the plant for metabolism, resisting disease, etc. The remaining 200 TW is "fixed" as consumable structure - food for animals, bacteria, and other dependent lifeforms, on and under the land, and in the sea. Humans consume about 1 TW of that directly as food, a lot more of that as wood and fiber, and a great deal more primary power to cultivate and process the crops that produce that food with current energy-intensive methods.
Supporting a world population of 10 billion with western European levels of consumption (an optimistic goal, since western Europe draws on energy-intensive primary production elsewhere) requires 50 TW of primary power. It is unlikely that we can extract that out of the total 200 GW produced by plants without completely destroying the remnants of wilderness. Much of the "waste" energy in nature is not actually waste at all. It drives the scavengers and soil bacteria that close the cycles of biomass that permit plant growth in the first place. We may already be extracting more energy from nature than is available in the long term, and causing irreversible damage to the biological carrying capacity of the Earth.
Perhaps the only likely source of additional "natural" energy is biotechnology, creating extremophile plants adapted to polar and desert conditions. Most of the ocean is devoid of nutrients or laden with pathogens that attack plankton, limiting plant growth. Adding nutrients for genetically-engineered disease-immune plankton might boost the oceanic production of biomass. Or it might destroy existing plankton communities and kill the oceans. We don't know enough to safely do this kind of global-scale bio-engineering.
Photovoltaic, wind and ocean power are touted as saviors, but these are intermittent and unconcentrated sources. A parallel infrastructure, driven by natural gas, is needed to match the peaks and valleys in the supply to the demand. Nuclear fission with fuel recycle and long-term waste storage is a long term option favored by some, but the political climate (at least in the US) is hostile.
The Problem with Extrapolation
All these exponential extrapolations assume usage can keep growing forever. In fact, we are approaching real limits, and geopolitical economic conflict, which will invalidate any extrapolation of "business as usual". All that can be said for sure is that assumptions of continued growth in computers and internet infrastructure, which are the unacknowledged basis for the projected growth of many internet-based businesses, may be overly optimistic. Unless we find new sources of energy, we may be approaching a brick wall.
The Sun produces 1366 watts per square meter at the distance of Earth's orbit (150 million km on average). The total solar output is 4 \pi \times 1.5e11^2 \times 1366 or 3.86e26 or 386 trillion Terawatts. The "optical radius" of the earth is around the top of the clouds, an average radius of about 6380 km, so the total power intercepted is \pi \times 6.380e6^2 \times 1366 or 174,700 Terawatts.
A disk 1.28 light seconds in diameter around the earth ( 384,000 kilometers, the distance from the Earth to the Moon ) has an area of 4.63E17 square meters, and 6.33E20 watts of sunlight passes through it - 633,000,000 Terawatts. If we could capture 1,000 TW of that power, and send it to the power grid of the Earth with 5% efficiency, the resulting 50TW could feed the entire world's energy demand without depleting Earth resources. If an extra 100TW of heat was added to the Earth (50TW as terrestrial inefficiency, 50TW as end use), the additional black body radiation would heat the Earth by 0.04 Celsius, which would be more than compensated by a massive reduction in greenhouse gasses.
Server sky can move both the poewr generation and consumption for data centers and computation off the Earth, with no direct increase in Earth heating. Most big computation tasks can tolerate the high latencies of distant computation arrays (beyond the moon), so vastly more of the Sun's power is available in the long term. The resources available for space computation may be trillions of times the limits for Earth computation, and quadrillions of times current usage.
Replacing Carbon Fuels in the US
The vast majority of power in the US is produced by coal, petroleum, and natural gas. The US uses 3.4TW of primary power on average, and more than 2.2TW is carbon fuel. "Heating fuels" cannot be converted to useful work with 100% efficiency, so transportation and electricity generation are about 30% efficient; electricity can be closer to 100% efficient. For heating things, heat pumps can move more heat with less watts. So if we use 2.2TW of carbon fuel, we can replace it with 30% of the electricity, 660GW on average. However, because our usage of carbon fuel changes over the day, we must be able to provide "peak power" in excess of the average. System capacity might need a 50% additional reserve (WAG, assuming a lot of energy storage capacity), resulting in a needed capacity of 1.5*0.66 or about 1TW.
Assume (again, optimistically) that the cost of replacing gasoline cars with electric cars is part of the normal wear-out-and-replacement process, and the same holds for other infrastructure.
Nuclear power plants cost about \$5/W to build and connect to the grid (including lifetime fueling, later decommissioning, but not finance charges), and are available about 80% of the time. So, it would cost \$6 trillion to replace all carbon fuel usage with nuclear electric. Assuming this is done for an average wage of \$20/hr, that is 300 billion hours. With 150 million working Americans, we will on average have to work an additional 2000 hours per worker to accomplish this, over whatever time frame we chose to do it in.
A 1GWe nuclear plant uses about 250 acres for the plant itself, though it is often surrounded by a larger security buffer zone, often used for wildlife habitat. So, the actual plant area for 1000 nuclear plants would be 250,000 acres, about 390 square miles, about half the area of Washington County, Oregon.
Hydro is built out, and wind is poorly matched to power demand. Most wind farms peak at night, when demand is lowest, so the power is sometimes dumped into resistors. Worse than useless. Solar photovoltaic is better matched to the load, but has many other problems. The Nellis AFB photovoltaic array can (if it is kept clean) produce a peak power level of 14MW on sunny days, but the availability is 20%, perhaps half that in winter. It uses 140 acres and cost \$100M.
Solar thermal can be more efficient, and the heat can be stored more cheaply than electricity can. However, a solar thermal plant cannot efficiently use diffuse sunlight, and the mirrors must be kept clean and polished. They will be scoured into uselessness by blowing sand, and there will be a lot more sand as sunlight is blocked to desert soil bacteria that hold the sand down. Those will die (releasing carbon) and release the sand to the wind. Solar thermal will require cooling water (a scarce commodity in deserts). Both solar thermal and solar photovoltaic require water for cleaning the collecting surfaces and tracking mechanisms, huge quantities to keep thousands of square kilometers clean. Solar power works well in space, not on the ground.
Solar plants must be tied into nation-sized distribution grids (through lossy wiring) so that clouds in one place are compensated by sunshine in others. Perhaps we can approach 10% availability for the whole ensemble. So, the cost for the Nellis array is \$70/W to build and connect to the grid. The land usage is 100 acres per megawatt. At that rate, it would cost \$70 trillion to replace all US carbon fuel usage with solar photovoltaic, and the land area consumed would be 156,000 square miles, about big as California. An additional 23,000 hours per US worker would be needed.
It is reasonable to assume that in both the nuclear and solar cases, that manufacturing and deployment will be highly automated, and that economies of scale will lower the price. But that applies to all alternatives, including status quo carbon alternatives like natural gas, and even more strongly to space energy alternatives like Server Sky. On the other hand, massive deployments of anything will run into unexpected problems such as scarce resource depletion, endangered species, shortages of special skills, and the waste and corruption that can hide in large projects. So the economies of scale may be overwhelmed by the unexpected.
However you approach the problem, replacing cheap and dirty carbon fuels will not be easy, and will require vast amounts of resources and labor, diverted from other pressing needs. Most people have no idea how much this will cost - those who do usually expect to shift the burden to others.