AI is making enormous demands on electricity grids around the world. Exactly how enormous isn’t easy to say, since statistical agencies aren’t reporting the AI sector separately and the companies themselves aren’t exactly forthcoming about the topic. So researchers have to do a bit of guesswork to come up with some numbers.

A recent article in MIT Technology Review by James O’Donnell and Casey Crownhart is one of the latest efforts to do so. Unusually, the authors start from the query level and work upwards to the macro. A simple text query doesn’t make many demands—”about what it takes to ride six feet on an e-bike, or run a microwave for one-tenth of a second,” in O’Donnell and Crownhart’s words. Generating a simple image takes about two to four times that. A reasonably high-definition video five seconds in length requires lots more: the equivalent of “riding 38 miles on an e-bike, or running a microwave for over an hour.”

These numbers add up. The authors offer the example of using AI to launch a charity run, involving querying about how best to fundraise, generating a flyer, and producing a five-second video for posting to Instagram. That would “use about 2.9 kilowatt-hours of electricity—enough to ride over 100 miles on an e-bike (or around 10 miles in the average electric vehicle) or run the microwave for over three and a half hours.” And that’s just one person with one modest task.
At the aggregate level, the sums get enormous. Lawrence Berkeley Laboratory estimated that in 2024, US data centers of all kinds used enough electricity to power Thailand for a year. AI alone accounted for about a quarter to a third of that total—enough to power more than 7.2 million US homes for a year. Before ChatGPT’s launch in November 2022, AI usage was next to nothing.

This AI-driven onset of rapid electricity demand growth follows years of relative flatness. Data centers proliferated, but got more efficient, resulting in little increase in energy demand. That’s changing radically. On current trends, by 2028, AI alone will use enough electricity to power nearly a quarter of US households.

And since the major AI companies are negotiating favorable electricity deals, residential customers may wind up footing a large share of increased construction and generation costs. An investigation by the Virginia legislature reports that residential customers could see their bills rise by as much as $450 a year because of this cost-shifting. Electricity demand in the state, one of the world’s leading data center locations, is likely to double over the next decade because of demand from AI servers. That’s a major change from the previous 15 years, when there was little growth in demand for juice. And while these data centers are hulking and expensive, once constructed, they generate little employment. Building one can generate 1,500 jobs over 12 to 18 months, but after that, a typical facility employs just 50 workers.

Reflecting on these numbers prompted us to look over some stats on electricity. Nationally, as the graph below shows, electricity production, as measured by the Federal Reserve’s industrial production series, was, like Virginia’s, virtually unchanged between 2006 and 2020—and 2024, for that matter. (We’re using five-year intervals in the graphs because yearly numbers are quite volatile.) It’s begun to rise—up around 2% a year since 2023, compare to an average annual rate of 0.1% between 2000 and 2023, and if the projections are to be believed it’s only the beginning.

And costs, graph below, are also starting to rise. Measured by the CPI, electricity prices have risen more rapidly over the last five years than at any time since the early 1980s. Electricity prices also surged between 2006 and 2010, driven by sharp increases in natural gas prices, but not by fresh demand. They eased, along with natural gas prices, as the fracking boom kicked in. They’ve been rising again, up over 10% at an annualized rate in late spring and early summer.

Mention of natural gas brings up the topic of energy sources for electricity generation. These have changed enormously over the last few decades. (Graph below) For much of the 20th century, half or more of our electricity was generated from coal; in 1988, 57% was. From there, coal’s share began a long decline; so far this year, it’s averaged 16%, up a point from last year all-time low. (Coal may be filling some of that AI-increased demand.) Over the same period, natural gas went from 10% to 38%—and renewables (geothermal, solar, wind) went from under 1% to 19% (3 percentage points more than coal).

That growth in renewables is driven by cost, not wokeness. Graphed on p. 7 are the costs of various energy sources as estimated by Lazard. (They present ranges; the graph shows the means of those ranges.) Onshore wind installations are 22% cheaper than gas and 50% cheaper than coal. Individual solar facilities on houses and industrial sites aren’t so cheap, but large-scale installations managed by utilities are 26% cheaper than gas-fired generating facilities and 52% cheaper than coal.

These are not transient developments. In the 2020 edition of its World Energy Outlook the International Energy Agency (IEA) said that solar power “is now the cheapest source of electricity in history.” Wind’s advantage isn’t as dramatic, but it’s real—and utilities are acting on these cost comparisons. Again, according to the IEA, this its Global Energy Review 2025 (which has replaced the World Energy Outlook), renewables “made up almost three-quarters of the overall increase in power generation” in 2024, with solar in the lead. Add nuclear and hydro and you approach four-fifths. Fossil fuels accounted for almost all of the remaining fifth, well below their existing share.

Fossil fuels are on the way out. Over the longer term, the IEA projects that global oil demand growth will slow into a peak in 2030, and begin declining thereafter, and what increase in demand there is likely to be will be for petrochemicals, not energy. (Projections like these have led the Trump administration to threaten to withdraw from the IEA.) Natural gas is another story—demand for it, according to McKinsey, is projected to grow over the next decade, the only fossil fuel to do so after 2030, to a peak around 2037 and then flatline thereafter. While policies could accelerate or retard the post-fossil transition, the relative costs of the energy sources are doing much of the transformative work.

Sadly, US policy is paddling against this current. As we were writing this, news came in that the Trump administration was intensifying its previously declared war on windmills, assembling a joint task force staffed by six cabinet agencies to put an end to what the president sees as an ugly (tastes differ) and expensive (it’s not) energy source. (He also hates solar, but he’s carrying on that war with less intensity for now.) Instead, his administration is encouraging utilities to turn to more expensive options like gas and coal. Even if you’re a skeptic on climate change, this emphasis makes little economic sense.

We have worked closely with the Bureau of Labor Statistics for decades and, in the belief that people are more likely to value what they understand, are adding some historical context focusing on the early days at the Bureau, to the ongoing discussion of the many probable repercussions of President Trump’s firing of Commissioner Erika McEntarfer on August 1st, AKA “Jobs Friday.”

The Bureau of Labor Statistics was established in 1884 to study the many issues affecting working men and women. In the words of the first commissioner, Carroll D. Wright, the mission was to conduct “Judicious investigations and the fearless publication of the results.” That was a tall order for a team of three on a $25,000 budget!

Since their founding, the BLS has published more than a thousand monthly employment reports, and this is the first time a commissioner has been fired directly after the release of one of those reports.

There have been accusation of data manipulation, some apparently substantiated, and there is one suspect dismissal that was cloaked as a retirement-age requirement.

Leading into the 1932 election, President Herbert Hoover announced that the Great Depression was almost over, based on the fact that the employment rate had grown by 4% in the previous month. Francis Perkins, who was soon to become President Franklin D. Roosevelt’s Secretary of Labor, pointed out that his statement was inaccurate, the jump was caused by seasonal Christmas hiring, not permanent jobs, something she believed Hoover would have understood on his own. And Ethelbert Stewart, then BLS chief, had told Hoover this. Hoover continued to make the claim, as did his Secretary of Labor, William N. Doak, and the press approached Stewart for his opinion. He was direct in his criticism, which Doak tried to refute, and then gave Stewart a public ” tongue-lashing for daring to contradict his chief.”

Although details of Stewart’s departure from the BLS are somewhat unclear, Time reported under the lede, “Last week the Government’s foremost expert on joblessness found himself jobless,” that although Stewart was supposedly forced into retirement by his age, many disputed that. Stewart himself rejected the term “retired,” adding, “Don’t put it that way. I’ve had a tin can tied to the end of my coat tail.”

In 1971, President Richard M. Nixon ordered the BLS statistician who pointed out that a fall in the unemployment rate from 6.2% to 5.6% was likely caused by a “statistical quirk” be identified, and fired. And in the coming months when the assistant commissioner Harold Goldstein pointed out that one decline in the unemployment rate was “marginally significant,” and another “sort of mixed,” Nixon cancelled the briefings that traditionally followed the release of the monthly jobs report. Rumor has it that Senator William Proxmire, known as a critic of wasteful government spending, requested that the Bureau report directly to the Joint Economic Committee on which he served so they could have discussions free of political spin.

Nixon also believed the BLS included a “Jewish Cabal” out to get him, which led to the “Nixon Jew count,” where H.R. Haldeman supervised an investigation into BLS employees with “Jewish-sounding names.” In what Timothy Noah once described as the “last known act of official anti-Semitism conducted by the United States government,” thirteen employees identified as Jewish were moved into positions that did not involve compiling the politically sensitive employment reports.

It’s easy to think BLS staff, and Dr. McEntarfer herself, were exhibiting the fearlessness Wright was committed to, but releasing and correcting data is just what the BLS does. Years ago, an economist there told us it is indeed painful to release unusually large benchmarks, but “we do it anyway.”

Looking back to the early days of the BLS, there’s a good bit of colorful language. Amid the ongoing debate concerning founding a bureau that concentrated only on the world of work, one senator argued, “A great deal of public attention in and out of Congress has been given to the American hog and the American steer. I submit, Mr. Chairman, that it is time to give more attention to the American man.”

Perhaps that should have been man and woman, and within its first three years the BLS published a study of woman working in “manufactories” in big cities.

Commissioner Wright’s first annual report provided information on the character and potential causes of industrial depressions in a global context.

Here’s a timeline of the development of the many data sets, and we’ve put together some highlights.

Publications in the 1880s included the first consumer expenditure survey, and in the next decade the BLS looked into the effect of tariffs on wages and prices, and investigated the effects of machinery on employment, production and productivity in Hand and Machine Labor.

Later BLS added producer and consumer price indexes, industrial accidents, and in 1915 began their monthly surveys of employment and payrolls, now the Current Employment Statistics program. The first Monthly Labor Review was published that year, and the BLS signed cooperative information agreements with the states the following year.

BLS’s expertise is long-standing. By the 1940s they were training international economists and statisticians, and in 1959 they set up the Household Survey, now officially known as the Current Population Survey. Soon after they conducted comparative studies of international unemployment rates, and of federal and private-sector compensation.

In the 1960s, BLS began recording occupational injuries and illness, and the first job openings reports were produced. The full JOLTS report would be later.

Data on workers with disabilities appeared in 2009, on green jobs in 2012, and that same year BLS Tweeted for the first time!

During the covid pandemic BLS added questions about job-site respiratory illnesses, and began tracking much-needed data on indigenous populations in 2022. That was long in the making. BLS is now using QCEW data to identify populations facing violent weather events.

True to their mission of accurate impartial investigations of labor conditions, BLS established the Business Research Advisory Council and the Labor Research Advisory Board in 1947. In 2000 the BLS, together with the Bureau of Economic Analysis, founded the Federal Economic Statistics Advisory Committee, in 2007 the long-standing labor advisory board was renamed the Data Users Advisory Council, and in 2010 the BLS put together the Technical Advisory Committee.

All of these committees, made up of experienced and unpaid professionals in economics and related fields, were disbanded by President Trump in late March.

The boards were established to monitor data quality, evaluate BLS methodologies, and suggest improvements. This work was intended to be ongoing—technology changes, and pressure from administrations varies—which made the statement that members had “fulfilled their mission,” as advisors to the BEA were told, downright humorous.

At the time, former BLS Commissioner Erica Groshen commented that given the Committees’ mandate, “I would say that if an administration wanted to try to manipulate data, then they would not want these advisory committees to be around.”

The monthly payroll reports the BLS has published since their founding covered the Great Depression, all manner of lesser recessions, wars, and the effects of inflation spikes. Many of those reports were far worse than the limpid July report that caused Erika McEntarfer to lose her job. President Barack Obama had to accept huge monthly declines, and a -0.9% benchmark revision, multiples of average at the time. President Joe Biden was accused of having cooked the books when the unusually large benchmark revision to 2024 payrolls came out late last August, just two months before the election. (Of course, he was no longer the candidate, but if the BLS were tampering with the records to favor the Democrats, they were doing a lousy job.)

In her history of the BLS, Janet Norwood notes that many of the problems faced in the early days of the BLS are “unresolved to this day.” A major problem then and now is low and late response rates. As early as 1885 a state commissioner argued with appealing candor, “If questions are asked of five hundred men indiscriminately, and two hundred …give answer, those two hundred will not be average representatives of the whole five hundred. They will, on average, have more brains than the other three hundred. The very fact that they answer, while others do not, shows this.”

The revisions that sent President Trump to a place no has ever been before, firing an experienced commissioner because a report was not flattering to the current economy, are caused in large part by such late responses. Response rates have been declining for some time, and former Commissioner Groshen believes some of that decline has been driven by a loss of confidence in public data, and a growing disregard for public research. An inadequate budget also gets a nod. With the BLS’s current, nominal, budget slashed by eight-percent, BLS staff will probably be unable to follow up on particularly pesky sectors, like public education, as they once did.

Bureau of Labor Statistics employees have long argued that with proper funding they could update their data series in order to make them more timely and accurate. For example, the Department of Labor designed the weekly unemployment claims series as a administrative tool, which is why it can sometimes be misleading in gauging unemployment. The series includes a wealth of regional and demographic data that could be restructured as an economic indicator, a kind of early warning system, that would give policy makers notice on developing weakness they could then address, and those who want the 411 on national employment for market reasons would have more timely data as well.

The other is funding an update of the Quarterly Census of Employment and Wages. As we often mention, QCEW covers 97% of the employment universe as defined in the establishment survey, but is released with a lag. We won’t have data for the first quarter of 2025 until early September. And recent budget cuts just pushed that forward from late August.

The decision to cut the BLS budget, and staff, was made by the current administration. When the administration disrupted the carefully crafted structure, including remote work, the BLS had put together to deal with the fact they were moving into a space 40% smaller than the one they had occupied, and budget cuts were announced, many analysts raised red flags suggesting these disruptions could lead to larger revisions in upcoming releases.

Correlation, of course, does not imply causation—we could be seeing big downward revisions because the labor market, driven by all the unknowns and supply disruptions, is weakening more than we knew. That would make the birth/death model too additive, as often happens in downturns, which the BLS took into account during the pandemic, hence the very small benchmarks in 2020 and 2021.

But that possibility doesn’t erase the question. How quickly will the administration’s actions cut into the quality of the BLS’s data, respected around the world as the gold standard, and crucial to our creditors?

Karina A. Sanchez, Amanda J. Bevan Zientek, and Emily A. Holt just released their study of college students’ awareness of institutional, structural, and cultural racism and sexism in the field of ecology. The team, all part of the University of Northern Colorado’s department of biological sciences when the study was conducted, first had students fill out a five-question survey on their own perceptions of racism and sexism in the ecological sciences. The students then collected and analyzed data to evaluate patterns in different publications and data, identifying those that might be problematic, and considered how racism/sexism affected the field. Following this intervention, the students were resurveyed.

All fields of science are now addressing legacies of injustice that can shape who participates in a given field, research projects, and policy outcomes, but many focus on research labs, or are top-down administrative changes. The authors identified college classrooms as centers of social change around the world. Starting there, they intend their work to help fill the research gap on how inequities are being addressed in colleges and universities.

The lack of diversity in ecological science has long been a matter of concern. Some years ago, npj Biodiversity, a journal within Science, posted an editorial identifying biodiversity research as a science of crisis. Over the millennia that crisis has been driven by unsustainable exploitation, and changes made to our landscapes, both directly and via human-assisted migration of nonnative species that may become invasive. The editors detailed how overemphasis on the temperate regions of the northern hemisphere and terrestrial ecosystems to the detriment of marine and freshwater ecosystems, drylands and other extreme habitats, all limit our understanding of interrelated forces, which in turn hampers our ability to improve outcomes. The editors also highlight the lasting impact of Darwin’s emphasis on competition among species. To them that has had a “tremendous, and to a considerable extent spurious,” effect on our understanding of evolutionary processes, perhaps especially our understanding of species co-exiting in diverse habitats. The “painfully consistent biases,” built into the ecological sciences also devalue the work of women, underrepresented minorities and indigenous peoples, as well as researchers from the Global South, all of whom have “unique knowledge” and views on how to proceed in our current crisis.

In their opening paragraphs, Sanchez, Bevan Zienek, and Holt note that efforts to reform medical sciences have intensified in part because social inequities in the field are so visible, and are now the focus extensive research. Ecological science has not had the benefits of this kind of intense scrutiny, and although the share of women has grown in higher education overall, in the ecological sciences, their focus here, there are persistent publication biases, as well as citation and funding imbalances, between men and women.

The authors note that “systems of oppression,” can affect study outcomes. For example, natural history museums tend to collect species with a slight overall bias to males among birds and mammals that can lead to significant disparities at the species level. In some cases just 27% of specimens are female, obviously a research issue. Further, habitats in underserved communities, as well as war-torn areas, are chronically under-sampled. The authors believe that systemic racism, gender bias, and colonialism have not only transformed ecosystems throughout the world, they continue to exacerbate unequal access to natural areas among marginalized communities.

We touch on much of this as reports come out. Two examples: biases toward the temperate north led to the short-sighted belief that only male birds sing, recently corrected by a team of female ornithologists who studied common ancestors of our migrating birds in Australasia where males and females were signing away.

A recent study by the venerable American Forests showed one can determine redlining practices used in the 1930s by forest canopies in our cities. The average canopy is 39%, running from 23% in formerly redlined communities, to 43% in primarily US-born white communities.

Sanchez and her team believe that although hiring of diversity officers and other administrative changes hold promise, they may get stuck in rhetoric with “minimal” long-term effects, and administrations are of course slow to change. Colleges have a long history of advancing civil rights, with women’s study programs advancing gender equality, and efforts to decolonize curricula and pedagogy picking up steam in the humanities and social sciences. But, again, science has lagged. The authors suggest that may be the fault of the narrative that tells us science is objective and value-neutral, and hence outside the realm of social justice.

The study included thirty-eight upper division undergraduates, primarily white, 68%, and female, 79%, with 16% Latino, 8% Asian, 5% multiracial, and 3% African-American participants. About 60% were continuing generation students, meaning at least one parent attended college. In the twelfth week of the sixteen-week semester, the students filled out a five-question open response survey, anonymously and without pay: who does ecology, are they aware of one prominent ecologist to whom they can personally relate, is the field racist, is it sexist, did European colonialism affect the field, and does it still? Two weeks later, the team introduced their intervention, designed to outline social justice issues in ecology, and identify demographic representation of field ecologists. Students worked in groups to identify under-representation and inequality in ecological data, including the nationalities of “parachute” scientists, those who drop into a developing country and complete data collection and analysis with no input from local scientists or the community. For example, in Indonesia forty percent of published studies of coral reefs included no local scientists.

Students were given plenty of time to research their topics, to discuss them fully, and to look for patterns, and decide if the patterns they discovered represented problems. When the intervention was completed, students were resurveyed using the same five questions.

And the responses had indeed changed. The share who believed anyone with interest could participate in the field dropped by 16%, and shares agreeing that ecology was influenced by sexist or racist rose. In this day and age we need to point out that the idea was not to force beliefs on the students, but to discover how exposure to different data changed their thinking. In fact, the authors were unsure if some students had wondered about racism and sexism in ecology, but didn’t feel safe sharing those feelings until the study project was complete. They also stressed that some students expressed an interest in promoting diversity in their own careers after the intervention, and explored possible routes in study groups. And to us their statements included greater nuance in the second survey.

Fun fact: For generations, Draw-a-Scientist studies produced a stereotypical white man in a lab coat from K-12 students, despite ongoing changes in science. Other interventions cited in Sanchez, Bevan Zientek, and Holt’s study helped students move beyond this stereotype. Instead of scientists being defined as “people who do experiments,” descriptions focused instead on traits like curiosity and specific interests. Sanchez et al. suggest this may help potential ecologists both relate to scientists and to see themselves as scientists. And incorporating topics like discrimination and oppression helps “disrupt the façade of neutrality in science,” helping students understand that ecology is “imbedded in humanism with all its biases.”

And, as you might guess, the authors sign off with an acknowledgement that this is one step, which they classify as a transformational approach, and many more are needed to build curricula focused on solutions, and to increase opportunities that will support a diverse scientific community.

The New York Botanical Garden has long included diverse scientists in their programming, including indigenous scientists who are guiding restoration efforts with their own practices.

If of interest, you can watch some webinars here.

Photographs: Field science projects are a great way to involved diverse students. Eel populations are in severe decline, and Bard College students, along students from the region, help catch, count and release the glass eels caught in the fyke net to monitor their levels. The glass eels you see in the second photograph just spent the last year drifting up on the currents from the Sargasso Sea.
Credit: Philippa Dunne