This story was originally published by Grist and is reproduced here as part of the Climate Desk collaboration.

Earlier this year, the e-commerce corporation Amazon secured approval to open two new data centers in Santiago, Chile. The $400 million venture is the company’s first foray into locating its data facilities, which guzzle massive amounts of electricity and water in order to power cloud computing services and online programs, in Latin America—and in one of the most water-stressed countries in the world, where residents have protested against the industry’s expansion.

This week, the tech giant made a separate but related announcement. It plans to invest in water conservation along the Maipo River, which is the primary source of water for the Santiago region. Amazon will partner with a water technology startup to help farmers along the river install drip irrigation systems on 165 acres of farmland. The plan is poised to conserve enough water to supply around 300 homes per year, and it’s part of Amazon’s campaign to make its cloud computing operations “water positive” by 2030, meaning the company’s web services division will conserve or replenish more water than it uses up.

The reasoning behind this water initiative is clear: Data centers require large amounts of water to cool their servers, and Amazon plans to spend $100 billion to build more of them over the next decade as part of a big bet on its Amazon Web Services cloud-computing platform. Other tech companies such as Microsoft and Meta, which are also investing in data centers to sustain the artificial-intelligence boom, have made similar water pledges amid a growing controversy about the sector’s thirst for water and power.

One recent estimate found that ChatGPT requires an average-sized bottle of water for every 10 to 50 chat responses it provides.

Amazon claims that its data centers are already among the most water-efficient in the industry, and it plans to roll out more conservation projects to mitigate its thirst. However, just like corporate pledges to reach “net-zero” emissions, these water pledges are more complex than they seem at first glance.

While the company has indeed taken steps to cut water usage at its facilities, its calculations don’t account for the massive water needs of the power plants that keep the lights on at those very same facilities. Without a larger commitment to mitigating Amazon’s underlying stress on electricity grids, conservation efforts by the company and its fellow tech giants will only tackle part of the problem, according to experts who spoke to Grist.

The powerful servers in large data centers run hot as they process unprecedented amounts of information, and keeping them from overheating requires both water and electricity. Rather than try to keep these rooms cool with traditional air-conditioning units, many companies use water as a coolant, running it past the servers to chill them out. The centers also need huge amounts of electricity to run all their servers: They already account for around 3 percent of US power demand, a number that could more than double by 2030. On top of that, the coal, gas, and nuclear power plants that produce that electricity themselves consume even larger quantities of water to stay cool.

Will Hewes, who leads water sustainability for Amazon Web Services, told Grist that the company uses water in its data centers in order to save on energy-intensive air conditioning units, thus reducing its reliance on fossil fuels. 

“Using water for cooling in most places really reduces the amount of energy that we use, and so it helps us meet other sustainability goals,” he said. “We could always decide to not use water for cooling, but we want to, a lot, because of those energy and efficiency benefits.”

In order to save on energy costs, the company’s data centers have to evaporate millions of gallons of water per year. It’s hard to say for sure how much water the data center industry consumes, but the ballpark estimates are substantial. One 2021 study found that US data centers consumed around 415,000 acre-feet of water in 2018, even before the artificial-intelligence boom. That’s enough to supply around a million average homes annually, or about as much as California’s Imperial Valley takes from the Colorado River each year to grow winter vegetables. Another study found that data centers operated by Microsoft, Google, and Meta withdrew twice as much water from rivers and aquifers as the entire country of Denmark. 

In Pennsylvania, one Amazon data center consumes about 20 percent of the electricity capacity of the nuclear power plant nearby.

It’s almost certain that this number has ballooned even higher in recent years as companies have built more centers to keep up with the artificial-intelligence boom, since AI programs such as ChatGPT require massive amounts of server real estate. Tech companies have built hundreds of new data centers in the last few years alone, and they are planning hundreds more. One recent estimate found that ChatGPT requires an average-sized bottle of water for every 10 to 50 chat responses it provides. The on-site water consumption at any one of these companies’ data centers could now rival that of a major beverage company such as PepsiCo. 

Amazon doesn’t provide statistics on its absolute water consumption; Hewes told Grist the company is “focused on efficiency.” However, the tech giant’s water usage is likely lower than some of its competitors—in part because the company has built most of its data centers with so-called evaporative cooling systems, which require far less water than other cooling technologies and only turn on when temperatures get too high. The company pegs its water usage at around 10 percent of the industry average, and in temperate locations such as Sweden, it doesn’t use any water to cool down data centers except during peak summer temperatures. 

Companies can reduce the environmental impact of their AI business by building them in temperate regions that have plenty of water, but they must balance those efficiency concerns with concerns about land and electricity costs, as well as the need to be close to major customers. Recent studies have found that data center water consumption in the US is “skewed toward water stressed subbasins” in places like the Southwest, but Amazon has clustered much of its business farther east, especially in Virginia, which boasts cheap power and financial incentives for tech firms.

“A lot of the locations are driven by customer needs, but also by [prices for] real estate and power,” said Hewes. “Some big portions of our data center footprint are in places that aren’t super hot, that aren’t in super water stressed regions. Virginia, Ohio—they get hot in the summer, but then there are big chunks of the year where we don’t need to use water for cooling.” Even so, the company’s expansion in Virginia is already causing concerns over water availability.

To mitigate its impacts in such basins, the company also funds dozens of conservation and recharge projects like the one in Chile. It donates recycled water from its data centers to farmers, who use it to irrigate their crops, and it has also helped restore the rivers that supply water-stressed cities such as Cape Town, South Africa; in northern Virginia, it has worked to install cover crop farmland that can reduce runoff pollution in local waterways.

The company treats these projects the way other companies treat carbon offsets, counting each gallon recharged against a gallon it consumes at its data centers. Amazon said in its most recent sustainability report that it is 41 percent of the way to meeting its goal of being “water positive.” In other words, it has funded projects that recharge or conserve a little over 4 gallons of water for every 10 gallons of water it uses. 

But despite all this, the company’s water stewardship goal doesn’t include the water consumed by the power plants that supply its data centers. This consumption can be as much as three to 10 times as large as the on-site water consumption at a data center, according to Shaolei Ren, a professor of engineering at the University of California, Riverside, who studies data center water usage. As an example, Ren pointed to an Amazon data center in Pennsylvania that relies on a nuclear power plant less than a mile away. That data center uses around 20 percent of the power plant’s capacity.

“If they are able to capture some of the growing water and clean it and return to the community, that’s better than nothing.”

“They say they’re using very little water, but there’s a big water evaporation happening just nearby, and that’s for powering their data center,” he said.

Companies like Amazon can reduce this secondary water usage by relying on renewable energy sources, which don’t require anywhere near as much water as traditional power plants. Hewes says the company has been trying to “manage down” both water and energy needs through a separate goal of operating on 100 percent renewable energy, but Ren points out that the company’s data centers need round-the-clock power, which means intermittently available renewables like solar and wind farms can only go so far.

Amazon isn’t the only company dealing with this problem. CyrusOne, another major data center firm, revealed in its sustainability report earlier this year that it used more than eight times as much water to source power as it did on-site at its data centers. “As long as we are reliant on grid electricity that includes thermoelectric sources to power our facilities, we are indirectly responsible for the consumption of large amounts of water in the production of that electricity,” the report said.

As for replenishment projects like the one in Chile, they too will only go part of the way toward reducing the impact of the data center explosion. Even if Amazon’s cloud operations are “water positive” on a global scale, with projects in many of the same basins where it owns data centers, that doesn’t mean it won’t still compromise water access in specific watersheds. The company’s data centers and their power plants may still withdraw more water than the company replenishes in a given area, and replenishment projects in other aquifers around the world won’t address the physical consequences of that specific overdraft.

“If they are able to capture some of the growing water and clean it and return to the community, that’s better than nothing, but I think it’s not really reducing the actual consumption,” Ren said. “It masks out a lot of real problems, because water is a really regional issue.”

This story was originally published by Grist and is reproduced here as part of the Climate Desk collaboration.

Earlier this year, the e-commerce corporation Amazon secured approval to open two new data centers in Santiago, Chile. The $400 million venture is the company’s first foray into locating its data facilities, which guzzle massive amounts of electricity and water in order to power cloud computing services and online programs, in Latin America—and in one of the most water-stressed countries in the world, where residents have protested against the industry’s expansion.

This week, the tech giant made a separate but related announcement. It plans to invest in water conservation along the Maipo River, which is the primary source of water for the Santiago region. Amazon will partner with a water technology startup to help farmers along the river install drip irrigation systems on 165 acres of farmland. The plan is poised to conserve enough water to supply around 300 homes per year, and it’s part of Amazon’s campaign to make its cloud computing operations “water positive” by 2030, meaning the company’s web services division will conserve or replenish more water than it uses up.

The reasoning behind this water initiative is clear: Data centers require large amounts of water to cool their servers, and Amazon plans to spend $100 billion to build more of them over the next decade as part of a big bet on its Amazon Web Services cloud-computing platform. Other tech companies such as Microsoft and Meta, which are also investing in data centers to sustain the artificial-intelligence boom, have made similar water pledges amid a growing controversy about the sector’s thirst for water and power.

One recent estimate found that ChatGPT requires an average-sized bottle of water for every 10 to 50 chat responses it provides.

Amazon claims that its data centers are already among the most water-efficient in the industry, and it plans to roll out more conservation projects to mitigate its thirst. However, just like corporate pledges to reach “net-zero” emissions, these water pledges are more complex than they seem at first glance.

While the company has indeed taken steps to cut water usage at its facilities, its calculations don’t account for the massive water needs of the power plants that keep the lights on at those very same facilities. Without a larger commitment to mitigating Amazon’s underlying stress on electricity grids, conservation efforts by the company and its fellow tech giants will only tackle part of the problem, according to experts who spoke to Grist.

The powerful servers in large data centers run hot as they process unprecedented amounts of information, and keeping them from overheating requires both water and electricity. Rather than try to keep these rooms cool with traditional air-conditioning units, many companies use water as a coolant, running it past the servers to chill them out. The centers also need huge amounts of electricity to run all their servers: They already account for around 3 percent of US power demand, a number that could more than double by 2030. On top of that, the coal, gas, and nuclear power plants that produce that electricity themselves consume even larger quantities of water to stay cool.

Will Hewes, who leads water sustainability for Amazon Web Services, told Grist that the company uses water in its data centers in order to save on energy-intensive air conditioning units, thus reducing its reliance on fossil fuels. 

“Using water for cooling in most places really reduces the amount of energy that we use, and so it helps us meet other sustainability goals,” he said. “We could always decide to not use water for cooling, but we want to, a lot, because of those energy and efficiency benefits.”

In order to save on energy costs, the company’s data centers have to evaporate millions of gallons of water per year. It’s hard to say for sure how much water the data center industry consumes, but the ballpark estimates are substantial. One 2021 study found that US data centers consumed around 415,000 acre-feet of water in 2018, even before the artificial-intelligence boom. That’s enough to supply around a million average homes annually, or about as much as California’s Imperial Valley takes from the Colorado River each year to grow winter vegetables. Another study found that data centers operated by Microsoft, Google, and Meta withdrew twice as much water from rivers and aquifers as the entire country of Denmark. 

In Pennsylvania, one Amazon data center consumes about 20 percent of the electricity capacity of the nuclear power plant nearby.

It’s almost certain that this number has ballooned even higher in recent years as companies have built more centers to keep up with the artificial-intelligence boom, since AI programs such as ChatGPT require massive amounts of server real estate. Tech companies have built hundreds of new data centers in the last few years alone, and they are planning hundreds more. One recent estimate found that ChatGPT requires an average-sized bottle of water for every 10 to 50 chat responses it provides. The on-site water consumption at any one of these companies’ data centers could now rival that of a major beverage company such as PepsiCo. 

Amazon doesn’t provide statistics on its absolute water consumption; Hewes told Grist the company is “focused on efficiency.” However, the tech giant’s water usage is likely lower than some of its competitors—in part because the company has built most of its data centers with so-called evaporative cooling systems, which require far less water than other cooling technologies and only turn on when temperatures get too high. The company pegs its water usage at around 10 percent of the industry average, and in temperate locations such as Sweden, it doesn’t use any water to cool down data centers except during peak summer temperatures. 

Companies can reduce the environmental impact of their AI business by building them in temperate regions that have plenty of water, but they must balance those efficiency concerns with concerns about land and electricity costs, as well as the need to be close to major customers. Recent studies have found that data center water consumption in the US is “skewed toward water stressed subbasins” in places like the Southwest, but Amazon has clustered much of its business farther east, especially in Virginia, which boasts cheap power and financial incentives for tech firms.

“A lot of the locations are driven by customer needs, but also by [prices for] real estate and power,” said Hewes. “Some big portions of our data center footprint are in places that aren’t super hot, that aren’t in super water stressed regions. Virginia, Ohio—they get hot in the summer, but then there are big chunks of the year where we don’t need to use water for cooling.” Even so, the company’s expansion in Virginia is already causing concerns over water availability.

To mitigate its impacts in such basins, the company also funds dozens of conservation and recharge projects like the one in Chile. It donates recycled water from its data centers to farmers, who use it to irrigate their crops, and it has also helped restore the rivers that supply water-stressed cities such as Cape Town, South Africa; in northern Virginia, it has worked to install cover crop farmland that can reduce runoff pollution in local waterways.

The company treats these projects the way other companies treat carbon offsets, counting each gallon recharged against a gallon it consumes at its data centers. Amazon said in its most recent sustainability report that it is 41 percent of the way to meeting its goal of being “water positive.” In other words, it has funded projects that recharge or conserve a little over 4 gallons of water for every 10 gallons of water it uses. 

But despite all this, the company’s water stewardship goal doesn’t include the water consumed by the power plants that supply its data centers. This consumption can be as much as three to 10 times as large as the on-site water consumption at a data center, according to Shaolei Ren, a professor of engineering at the University of California, Riverside, who studies data center water usage. As an example, Ren pointed to an Amazon data center in Pennsylvania that relies on a nuclear power plant less than a mile away. That data center uses around 20 percent of the power plant’s capacity.

“If they are able to capture some of the growing water and clean it and return to the community, that’s better than nothing.”

“They say they’re using very little water, but there’s a big water evaporation happening just nearby, and that’s for powering their data center,” he said.

Companies like Amazon can reduce this secondary water usage by relying on renewable energy sources, which don’t require anywhere near as much water as traditional power plants. Hewes says the company has been trying to “manage down” both water and energy needs through a separate goal of operating on 100 percent renewable energy, but Ren points out that the company’s data centers need round-the-clock power, which means intermittently available renewables like solar and wind farms can only go so far.

Amazon isn’t the only company dealing with this problem. CyrusOne, another major data center firm, revealed in its sustainability report earlier this year that it used more than eight times as much water to source power as it did on-site at its data centers. “As long as we are reliant on grid electricity that includes thermoelectric sources to power our facilities, we are indirectly responsible for the consumption of large amounts of water in the production of that electricity,” the report said.

As for replenishment projects like the one in Chile, they too will only go part of the way toward reducing the impact of the data center explosion. Even if Amazon’s cloud operations are “water positive” on a global scale, with projects in many of the same basins where it owns data centers, that doesn’t mean it won’t still compromise water access in specific watersheds. The company’s data centers and their power plants may still withdraw more water than the company replenishes in a given area, and replenishment projects in other aquifers around the world won’t address the physical consequences of that specific overdraft.

“If they are able to capture some of the growing water and clean it and return to the community, that’s better than nothing, but I think it’s not really reducing the actual consumption,” Ren said. “It masks out a lot of real problems, because water is a really regional issue.”

This story was originally published by Grist and is reproduced here as part of the Climate Desk collaboration.

When the Federal Emergency Management Agency spends millions of dollars to help rebuild schools and hospitals after a hurricane, it tries to make the community more resilient than it was before the storm. If the agency pays to rebuild a school or a town hall, for example, it might elevate the building above the floodplain, lowering the odds that it will get submerged again.

That sounds simple enough, but the policy hinges on a deceptively simple question: How do you define “floodplain”? FEMA and the rest of the federal government long defined it as an area that has a 1 percent chance of flooding in any given year. That so-called 100-year floodplain standard, though more or less arbitrary, has been followed for decades—even though thousands of buildings outside the floodplain go underwater every year. 

Now FEMA is expanding its definition of the floodplain, following an executive order from President Joe Biden that forced government agencies to tighten rules about how they respond to the increasing risk of floods. In a significant shift, the new standard will require the agency to factor in the impact of climate change on future flood risk when it decides where and how it’s safe to build.

The new rule will result in higher-elevated and better-fortified buildings, and could help break a cycle of destruction and reconstruction that has cost the government billions of dollars over the past few decades. In a press conference announcing the rule, FEMA administrator Deanne Criswell hailed it as a significant change in how the government responds to disasters. 

The rule “will allow us to enhance resilience in flood-prone communities by taking future flood risk into consideration when we rebuild structures post-disaster,” she said. “This is a huge win that will also allow us to end the repeat loss cycles that stem from flooding and increase the safety of families and save taxpayer dollars.” 

Under the new rule, the agency will “integrate current and future changes in flooding based on climate science” when it estimates flood risk, factoring in sea level rise and intensified erosion that will get worse over the course of the century. This will be easiest in coastal areas, where the science about sea level rise and flooding is well established. In riverine areas, where science is less robust, the agency will rebuild at least as high as the 500-year floodplain, or the land that has less than a 0.2 percent chance of flooding in a given year—and sometimes even higher for essential infrastructure such as bridges and hospitals.

“This is a huge win that will also allow us to end the repeat loss cycles that stem from flooding and increase the safety of families and save taxpayer dollars.” 

This is a dramatic shift from previous measurements, which relied on historical data to estimate future flooding. Because climate change has intensified since the collection of that initial data, previously the agency was systematically underestimating climate-related risk. Therefore, the new system assumes that flood risk is much higher than in the past, and that it will keep rising as time goes on. To mitigate that risk, FEMA will build farther from the water wherever possible and will raise structures on stilts and pilings when it can’t pull back from the coast.

“The federal government really has a duty to account for a future flood risk when it’s providing funding to build or rebuild homes or infrastructure, because it’s using taxpayer dollars,” said Joel Scata, a senior attorney at the nonprofit Natural Resources Defense Council and an expert on flood policy. Under the new rule, he said, FEMA is “going to be building in a way that’s not setting people and infrastructure up for future failure.”

FEMA has estimated that elevating and flood-proofing structures at this stricter standard could cost the agency as much as an additional $150 million over the next ten years—a proportionally small sum given the agency’s $3 billion annual disaster spending. The agency says that elevating structures by 2 additional feet adds around 2 percent to the cost of the average project, but that this spending will pay for itself over the next 60 years by preventing future damages.

There could still be trickle-down costs for local governments, which often have to pay around 25 percent of the cost when FEMA repairs a damaged school or installs a flood barrier in a community. Many small towns and low-income communities have struggled to provide these matching funds, and they have been excluded from federal resilience grants as a result.

The Biden administration is not the first to consider the 100-year floodplain standard inadequate. Then-President Barack Obama tried to expand the definition after Superstorm Sandy in 2012, but the Trump administration scrapped this revised standard just after taking office. President Biden’s rule has now advanced farther along in the regulatory process than the Obama administration’s rule was able to, which will make it much harder for a potential second Trump administration to repeal it.

Local updates to floodplain standards have already shown results: Houston, Texas, saw three massive floods in consecutive years between 2015 and 2017. After Hurricane Harvey struck in 2017, the city updated its building regulations to prohibit construction in the 500-year floodplain, forcing builders to elevate homes much higher or build farther back from rivers and streams. These standards likely prevented thousands of homes from flooding earlier this week during Hurricane Beryl, which caused several rivers and bayous to overflow and spill onto surrounding land.

This story was originally published by Grist and is reproduced here as part of the Climate Desk collaboration.

When the Federal Emergency Management Agency spends millions of dollars to help rebuild schools and hospitals after a hurricane, it tries to make the community more resilient than it was before the storm. If the agency pays to rebuild a school or a town hall, for example, it might elevate the building above the floodplain, lowering the odds that it will get submerged again.

That sounds simple enough, but the policy hinges on a deceptively simple question: How do you define “floodplain”? FEMA and the rest of the federal government long defined it as an area that has a 1 percent chance of flooding in any given year. That so-called 100-year floodplain standard, though more or less arbitrary, has been followed for decades—even though thousands of buildings outside the floodplain go underwater every year. 

Now FEMA is expanding its definition of the floodplain, following an executive order from President Joe Biden that forced government agencies to tighten rules about how they respond to the increasing risk of floods. In a significant shift, the new standard will require the agency to factor in the impact of climate change on future flood risk when it decides where and how it’s safe to build.

The new rule will result in higher-elevated and better-fortified buildings, and could help break a cycle of destruction and reconstruction that has cost the government billions of dollars over the past few decades. In a press conference announcing the rule, FEMA administrator Deanne Criswell hailed it as a significant change in how the government responds to disasters. 

The rule “will allow us to enhance resilience in flood-prone communities by taking future flood risk into consideration when we rebuild structures post-disaster,” she said. “This is a huge win that will also allow us to end the repeat loss cycles that stem from flooding and increase the safety of families and save taxpayer dollars.” 

Under the new rule, the agency will “integrate current and future changes in flooding based on climate science” when it estimates flood risk, factoring in sea level rise and intensified erosion that will get worse over the course of the century. This will be easiest in coastal areas, where the science about sea level rise and flooding is well established. In riverine areas, where science is less robust, the agency will rebuild at least as high as the 500-year floodplain, or the land that has less than a 0.2 percent chance of flooding in a given year—and sometimes even higher for essential infrastructure such as bridges and hospitals.

“This is a huge win that will also allow us to end the repeat loss cycles that stem from flooding and increase the safety of families and save taxpayer dollars.” 

This is a dramatic shift from previous measurements, which relied on historical data to estimate future flooding. Because climate change has intensified since the collection of that initial data, previously the agency was systematically underestimating climate-related risk. Therefore, the new system assumes that flood risk is much higher than in the past, and that it will keep rising as time goes on. To mitigate that risk, FEMA will build farther from the water wherever possible and will raise structures on stilts and pilings when it can’t pull back from the coast.

“The federal government really has a duty to account for a future flood risk when it’s providing funding to build or rebuild homes or infrastructure, because it’s using taxpayer dollars,” said Joel Scata, a senior attorney at the nonprofit Natural Resources Defense Council and an expert on flood policy. Under the new rule, he said, FEMA is “going to be building in a way that’s not setting people and infrastructure up for future failure.”

FEMA has estimated that elevating and flood-proofing structures at this stricter standard could cost the agency as much as an additional $150 million over the next ten years—a proportionally small sum given the agency’s $3 billion annual disaster spending. The agency says that elevating structures by 2 additional feet adds around 2 percent to the cost of the average project, but that this spending will pay for itself over the next 60 years by preventing future damages.

There could still be trickle-down costs for local governments, which often have to pay around 25 percent of the cost when FEMA repairs a damaged school or installs a flood barrier in a community. Many small towns and low-income communities have struggled to provide these matching funds, and they have been excluded from federal resilience grants as a result.

The Biden administration is not the first to consider the 100-year floodplain standard inadequate. Then-President Barack Obama tried to expand the definition after Superstorm Sandy in 2012, but the Trump administration scrapped this revised standard just after taking office. President Biden’s rule has now advanced farther along in the regulatory process than the Obama administration’s rule was able to, which will make it much harder for a potential second Trump administration to repeal it.

Local updates to floodplain standards have already shown results: Houston, Texas, saw three massive floods in consecutive years between 2015 and 2017. After Hurricane Harvey struck in 2017, the city updated its building regulations to prohibit construction in the 500-year floodplain, forcing builders to elevate homes much higher or build farther back from rivers and streams. These standards likely prevented thousands of homes from flooding earlier this week during Hurricane Beryl, which caused several rivers and bayous to overflow and spill onto surrounding land.