Lux Research took a close look at the claims being made about energy usage by owners of big datacenters and found them to be suspect.
The problem is that these metrics produce results that allow companies such as Google and Amazon to make claims that do not reflect reality, according to Lux Vice President of Analytics Ory Zik. “Companies are making bold claims about being carbon neutral or that they have cut their carbon emissions by X percentage,” Zik said. “The question we were asking is, ‘How do they calculate their carbon emissions?’ If they don’t calculate their current emissions properly, definitely they cannot claim that they are reducing it by 50 percent (or some other percentage).”
In the United States, the estimate of how much energy a facility uses is determined by the Environmental Protection Agency’s Emissions & Generation Resource Integrated Database (eGRID) system. The system, Zik said, was developed in the 1990s. The country is sliced into 24 regions and the data collected based on the nature of the power generation in a particular region. Estimates lag by three years. Thus, an eGRID score arrived at today and the ensuring claim by the datacenter on its efficiency are based on data collected in 2013.
To Lux, an equation that was so broad geographically and depended upon three-year-old data was insufficient. The firm moved forward with a formula that based assessments on more timely and granular metrics. It characterized emissions from about 20,000 power plants nationwide using monthly data from the Environmental Information Agency (EIA) instead of the three-year-old eGRID statistics. It also increased the regionalization from 30 to 184 public control areas (PCAs). This painted a picture, Zik said, that statistically is 80 percent more accurate than that drawn using the old metrics.
The new approach – which was published in the American Chemical Society and shared with the energy-focused segment of the analytical community – was first used to assess datacenters. The reason, Zik said, was that these are very sophisticated energy users who can’t plausibly say that they don’t have the tools to do the proper analysis. In other words, they are big energy users and big targets.
Both the eGRID and Lux approaches are estimates. They both are based on taking a section of the country (sliced into 24 regions in one case and 184 in the other) and assessing its energy usage. It therefore is an estimate, since Lux doesn’t know precisely which energy suppliers within that region that a particular datacenter uses. So, for instance, it is entirely possible that a particular datacenter would use more wind and solar than the collect data suggests is representative of that region.
The unsurprising result of using more granular and timely inputs is that what companies blithely claim almost certainly is inaccurate, according to Zik. “They missed by 25 percent, 30 percent, up to 40 percent,” he said. “And sometimes they performed better than they reported because they just didn’t do the analytics.”
Lux didn’t shy away from naming names. The press release on the research results said that Google appears to rely more on coal than it claims in four of its seven U.S. data centers. Likewise, the release says that Amazon use 7 percent more energy from coal – 43 percent versus 35 percent – than the eGRID-derived computations suggest.
Zik believes that Lux’s approach provides more tailored information. “Distributed generation is one of the most expensive and valuable decisions people make,” he said. “[Lux’s approach] allows you to be more surgical and more accurate in distributed generation decisions. This is critically important as we move toward solar and other renewable and distributed sources.”
Energy Manager Today contacted Alphabet – Google’s parent company – and Amazon for comment. They did not immediately respond. The piece will be updated if they do.