The High Cost of Energy Illiteracy and Sustainability Hype

What engineers and journalists have in common is their utter inability to communicate with one another reliably and compellingly.

A raising of standards would be helpful in improving idea exchanges about energy transition and industrial sustainability challenges.  Lifecycle cost approaches can improve the framing and discourse around challenges faced by manufacturers and asset-intensive industries, and the utilities and other owner/operators of the critical infrastructure that serve them.

Part of the high cost is due to narrow framing of issues. This prevents the right questions from being asked. It also places everyone at a disadvantage.  In some ways, it is more difficult to address narrow framing than outright falsehoods.

Energy illiteracy and sustainability hype stem from a paradoxical fact:  What engineers and journalists have in common is their utter inability to communicate with one another reliably and compellingly. An engineer can lose their audience if they focus too much on reporting details accurately, while a lot of important nuances about complex situations can fall between the narrative lines of a journalist’s “good story.”

This has made life very interesting for those of us who work as analysts, particular with regard to recent reporting around the following:

• The electric energy usage of hyperscalers’ infrastructure in support of AI- and cloud computing

• The GHG impacts of the Bloom AI solution’s training run

• Wind turbine cement foundations

• Challenges associated with carbon capture and sequestration (CCS)

To analyze these things well, and ask the right questions, provides great value to decision-makers and stakeholders across the industrial, manufacturing, utility, transportation, and other major sectors of the global economy. In contrast, narrow focus reporting, and false claims that get repeated and magnified, place industry leaders, government policy-makers, investors, and even R&D resource decision-makers, in much riskier strategic landscapes, creating more difficult decision-making scenarios which stifle creativity and healthy competition.

This has been made clear in numerous reports in recent months about the carbon footprint of one or another renewable energy source, or the electric energy usage of the massive data centers that support cloud-based computing platforms and solutions, or of the carbon footprint associated with training Chat GPT and other generative AI models.

The examples that follow do have a positive side.  After this ARC analyst reached out to the journalist, two of the four changed their article based on the suggested correction—both were major magazines with circulation above 500,000.

• The electric energy usage of hyperscalers’ infrastructure in support of AI- and cloud computing: An article stated that the costs of the electricity alone associated with powering the computing infrastructure for training the Chat GPT-3 model exceeded $100,000,000.  (The actual electric energy usage was 1,287 MWh, which, at last year’s industrial average rate of $0.08/kHhr, equaled $102,960, not $100,000,000). The same article associated this and other AI and cloud computing costs with what it repeatedly described as high costs associated with the hyperscalers (AWS, Google, Facebook) massive data centers. The size of a hyperscaler’s data center electric energy load, in and of itself, no matter how large, is not a “bad” thing, absent a comparison to the energy footprint of the alternate scenario—one where there are no hyperscaler data centers and where there is no cloud. The much less energy efficient on-premise customer server infrastructure that those end users would have had to install would involve highly overbuilt, underutilized, and less energy efficient operations. Note that related coverage in other publications repeatedly cited the ChatGPT-3 training’s electric usage of 1,287MWh as if it were a stunningly large amount of electric energy, even though it is a fixed cost for an asset that had been utilized by more than 100,000,000 users at an average rate of a billion uses per month. It would have been easy to make the 1000x electric bill error after all the misplaced outrage over the 1,287MWh. Ironically, hyperscalers not only directly help their customers lower the energy footprint of their compute requirements due to the higher energy efficiency and lower per-unit carbon footprint of their electricity supply sources generally, they also offer an increasingly deep and broad array of solutions that enable their customers to lower their energy footprint in other ways. (See the August 30, 2023 ARC blog, "Hyperscalers Continue Push to Address Meeting Industrial Sustainability Challenges.")

• The GHG impacts of the Bloom AI solution’s training run: A frequently re-quoted study states that there are serious environmental impacts associated with AI and as its proof, stated that BLOOM’s training run emitted 25 times more carbon than a single air traveler on a one-way trip from New York to San Francisco.” Why state the footprint of one person, when the same one-way trip Boeing 757, carrying 200 passengers, could yield a clear fact: the same BLOOM training run’s emissions could have been described as emitting 1/8th the carbon of a full airplane going from NY to San Francisco. Since the training run was a “fixed” cost spread over numerous users, even this way of stating it is not providing a clear picture.

• Wind turbine cement foundations: An article states the number of tons of cement required in a wind turbine foundation and complained about the high carbon footprint associated with production and transportation of the cement, without comparing it to the much higher tonnage per MW of generating capacity associated with the cement foundations of nuclear and other power plants.

• Challenges associated with carbon capture and sequestration (CCS): A journalist who wrote about his visit to one of the world’s largest direct air carbon capture facilities described the CO2 reduction goal as the equivalent of trying to pull a single droplet of water out of an Olympic swimming pool. Actually, it is the equivalent of trying to pull 374 gallons out of such a pool.

The wider remedy is to call for a simple standard to be raised, one which requires us to be more energy and sustainability literate.  This is in progress, and will take decades, in much the same way that nutritional labelling of food products did not instantaneously make everyone more conversant in the key issues and areas for useful comparisons.

And the ability to make good lifecycle cost and environmental impact comparisons, as a standard is what ultimately matters—it is more helpful than a blanket critique one solution’s carbon footprint or energy usage in the abstract, without comparing it to the viable actual alternatives involved.

Raising a standard may sound like engineering-speak.  It is not—it is refreshing to look back, in a more journalistic fashion, at the use of that term by George Washington at his Farewell Address. He said, “Let us raise a standard to which the wise and the honest can repair.” Think of “repair” as pairs of people, perhaps journalists and engineers, marching behind the same flag-bearer at the head of the group—that is the type of standard to strive for here.