I am sitting where I normally sit to write: at my desk, which faces the neighboring apartment and its lattice fence, over which tumble bunches of white jasmine flowers and their pink buds and spiky dark green leaves, beyond which is MacArthur Avenue and farther, Lake Merritt, and farther still, the East Bay Hills. I am, in other words, in Oakland. I am also on my computer, which means, on the internet, which feels, more and more these days—the internet mediated increasingly by all manner of invisible intelligent assistants—like nowhere and everywhere at once.
But the internet, and its artificial intelligence, is a place.
What counts as a place—or as nature—has never been static. There can be, of course, a concrete enough geographic demarcation. The Bay Area, for example, stretches from roughly San José to the south to roughly Santa Rosa to the north. But people move, animals migrate, currents swoop and air flows and soil erodes and fault lines shift. Humans have long lived in a globalized world of trade and industry, dependent upon natural resources moving across oceans and continents. Watch cargo ships unload at the Port of Oakland and it is easy to get a sense of how global the Bay Area’s local nature is. For example, pollution from tugboats, trains, and international ships has a direct impact on West Oakland’s air quality, leading to hospitalizations for asthma at a rate that is 88 percent higher than the county average.
The Human Animal
A quarterly column from Endria Richardson on how people are part of Bay Area nature. Find the rest of her columns here.
Still, mostly we have understood place to be measured and marked in distance, weather, geography, limits.
Chatbots like Anthropic’s Claude and OpenAI’s ChatGPT seem designed to defy a sense of physical place. Ask them a question and an answer appears almost instantaneously, contributing to the mirage that the internet is not geographically determined, not dependent on matter or constrained by local and global resources. Sam Altman, the CEO of OpenAI, has described artificial intelligence as technology that will make resources—both “intelligence” and energy —abundant enough to abolish, in theory, all limitations on human progress.
As of last spring, San Francisco houses 22 of Forbes’s “Top 50 AI Companies,” including Anthropic, OpenAI, and Crusoe, a company that specializes in building the massive data centers that make these AI tools possible. Anthropic’s offices are on Market Street in San Francisco, just a few blocks away from the hustle of the Ferry Building. OpenAI has three main buildings clustered off Third Street, just south of Oracle Park and Pier 52 and the launching ports for the Bay View Boat Club and the Mariposa Hunters Point Yacht Club. A stretch of the San Francisco Bay Trail, from Oracle Park to Heron’s Head Park, passes the buildings and offers views of the Bay, the Bridge, Emeryville, Oakland, and Alameda.
More important, the internet lives in its data centers—the physical structures that contain servers that process data; storage systems that store data; networking equipment to connect and direct data; cooling equipment to slough off the heat generated by these activities; batteries and generators to ensure uninterrupted power.
Increasingly, these centers are built not in the cities like San Francisco where people dream up the software, but in water-stressed areas in Texas or the Nevada desert, currently home to the largest data center in the country. Anthropic announced a $50 billion investment to build centers, starting in Texas and New York. OpenAI’s massive data center, the Stargate Project, began construction in Abilene, Texas, in 2024. Stargate plans anticipate at least 100 megawatts of energy per building for operation, and each building will require one million gallons of chilled water to circulate through the walls, to manage the significant heat produced by graphics processing units, or GPUs, the hardware used to train AI. In other words, enough energy to power a small city, or enough water to fill a swimming pool nearing the size of a football field.
Data centers built in the desert to fuel artificial-intelligence companies in San Francisco may put an even greater strain on water sources like Pyramid Lake in Nevada, stewarded and depended on by the Pyramid Lake Paiute Tribe. More than $15 billion worth of data centers are currently planned in California’s Imperial Valley—which already draws more water from its main water source, the Colorado River, than Arizona and Nevada combined. And in California, at least, there is currently no requirement that data centers report how much water they use for operations, nor that they use water-efficient technology. This is the sleight of hand behind the energy “abundance” of artificial intelligence.
Artificial intelligence is not magic, or placeless, or limitless. It is nature: wind, water, coal. These resources do not travel at the speed of light. Human labor drains water from specific reservoirs and rivers, digs copper, tellurium, lithium from specific mines. Artificial intelligence is a recent iteration, but this has long been a part of the human experience of the natural world: a commodity hides its physical cost.
I am on the internet, which means, I am in nature. A bay, a lake, a mine, a desert, at once.
