CHIPPED: Part 4
water wars, 12 years from rock to e-waste
It has been a century since physicist Julius Edgar Lilienfeld, an immigrant to the United States, patented the idea of using a semiconductor material to make a transistor. A hundred years later, silicon microchips, some with tens of billions of transistors, are in everything from computers to cars to coffeemakers. They make our modern world possible. Now, with Artificial Intelligence, they are poised to run the world.
CHIPPED is divided into five parts. Each can be read independently, but there is an arc to the order.
Chips are a hot commodity. Meta has been stockpiling Nvidia Graphics Processing Units (GPUs)—the $30,000 chips needed to train and run AIs—by the hundreds of thousands. Nvidia, whose customers also include Microsoft, Google, xAI and Amazon, has sold out of its top of the line chips through 2025. Elon Musk just announced plans to turn Colossus, a supercomputer with 100,000 GPUs, into a super duper computer with a million of them. And Nvidia competitor Broadcom reports at least three of its customers will be using its XPU AI chips to build million-chip supercomputers of their own by 2027.
The boom in sales for high-end chips means a boom in sales for all kinds of chips, including the more run-of-the-mill and much cheaper Central Processing Units (CPUs), data storage chips and sensors.
Chips aren’t only a hot commodity, they also run hot. The faster the chip, the hotter it runs, but all chips need to be kept cool in order to work their computational magic. A data center is effectively a radiator: jam-packed with racks of servers stuffed with chips converting the energy in the electricity that powers them into heat. A lot of heat. And the best, cheapest way to cool things down is with water.
Data centers are as thirsty for water as they are hungry for energy. A small data center might use 18,000 gallons of water per day for cooling, but a hyperscale data center, the kind filled with GPUs, can go through 550,000 gallons per day.
Where water is abundant, it isn’t a problem. But water is a site-specific, limited resource. And the kind of water data centers need to keep water pipes from corroding is the exact same kind humans need to survive: fresh potable water.
Even prior to AI, there was growing concern that data centers were sourcing water from moderately to highly stressed watersheds, putting them in direct competition with people for an increasingly precious resource.
This has already been an issue in Uruguay, a small South American country on the Atlantic coast bordered by Argentina and Brazil. In the early 2020s, the country experienced its worst drought in almost 75 years, which nearly emptied some of its reservoirs. An unprecedented three years of back-to-back La Niña-influenced dry weather conditions, combined with record heat attributed to climate change, led to the declaration of a water emergency. While there has long been concern in Uruguay about the amount of water used for agriculture (rice, soy and cattle operations) and by UPM, a Finnish forestry company that uses a lot of water in its wood pulp plants, the specter of Google building an $850 million water-guzzling data center sent people into the streets. Despite vigorous protests, last fall Google broke ground.
Even in places where water isn’t an issue right now, it could become one in the future. The 2024 Fairfax County Environmental Quality Advisory Council’s annual report includes a scenario where the continued proliferation and increased capacity of data centers in Virginia (which currently has the highest density of data centers anywhere in the world) shows a steep increase in water usage to 70 million gallons per day. That would be nearly double the current consumptive water use in the Potomac River basin, which provides drinking water for five million people, including the entire District of Columbia. The scenario does not take into account the potential impacts of climate change in Virginia, which is expected to become much hotter over the next 30 years. In the data center-saturated counties of Loudoun, Fairfax and Prince, the number of days with temperatures of 92°F or higher will more than quadruple to at least 33. Both people and chips will be in need of a lot more cooling.
•••••••••••••••••••••••••
According to market analyst Synergy Research Group, there are now more than a thousand hyperscale data centers around the world, with another 440 in the pipeline. The combination of more data centers, bigger data centers and more powerful chips will triple capacity by 2030, predicts Synergy. More data centers, bigger data centers and more powerful chips will also dramatically increase the need for cooling.
The most common way to cool a data center is with an aptly named “chiller” system, which pumps water through pipes around the building. As long as the water in the pipes is cooler than the heat given off by servers and chips, it will absorb their heat and carry it off. Some systems even have little pipes that go directly into servers to circulate coolness in closer proximity to chips. The warmed water is then pumped to a cooling tower on the roof where some of it evaporates, taking the captured heat with it. This is called evaporative cooling, basically a building’s version of sweating that releases heat into the atmosphere. Water that condenses in the cooling tower is cycled back into the system. Since some of the water is lost due to evaporation, this is an open loop system.
Last year, Microsoft announced a closed loop design it claims “will avoid the need for more than 125 million liters of water (33 million gallons) per year per data center.” Plans are to pilot the new system in 2026; it could be in use in all new Microsoft data centers starting in 2027. Although the closed loop design requires more power to operate than a chiller, it is a move in the right direction.
Water for Power
Another way data centers use water is linked to how the electricity that runs them is generated. Coal, gas and nuclear plants use heat to turn water into steam to spin turbines that rotate generators that convert mechanical energy into electrical energy. Nuclear plants also use water to keep spent fuel rods from overheating.
E-TREASURE TO E-WASTE
Just five to ten years. That’s the average lifespan of a chip. Tack on the couple of years it takes to turn quartz into pure silicon and a silicon wafer into a chip and the journey from quarry to landfill might be a dozen years. Or less.
Why such a brief existence? Chips wear out. Heat erodes functionality. Chips also become obsolete, replaced by new, more powerful chips. Chip-filled gadgets and appliances become obsolete, with consumers constantly encouraged to upgrade to the latest models of phones, tablets, laptops, desktops, televisions, doorbells, smart watches and refrigerators.
Manufacturers go out of business. The 2024 bankruptcy of EV startup Fisker left car owners panicked about software upgrades. Would their cars, which have thousands of chips, become undrive-able e-junk?
Then there is the sad story of the untimely demise of Moxie, a companion robot designed to help children develop social skills. When Moxie’s maker failed to secure a funding round, the servers that brought the charming blue robot to life were shut down. Young children experienced real grief watching their once animated friends turn into a blank-faced, lifeless piles of plastic, computer parts and chips.
•••••••••••••••••••••••••••
In 2022, the global tally for e-waste, loosely defined as anything with a plug or a battery, came to 62 million tons. With a recycling rate of only 20%, most of it ends up in massive landfills, mostly in the Global South. By 2030, the amount of e-waste is expected to hit 82 million tons, a one third increase in less than a decade.
“Unfortunately, phones and electronics are designed in a way that you can’t really use components and parts,” explains Bas van Abel, the founder of Fairphone, a startup selling phones specifically designed for disassembly. Fairphone is proof of what’s possible.
Instead, electronics are burned to melt the valuable minerals inside, making them easier to harvest, explains van Able. A devil’s brew of toxins is released in the process. “It’s a very stupid process. It’s kind of stupid to put something into an incinerator that so much effort was put into making.”
CHIPPED: Part 5 | a century of progress, collision course and the future