The recent spectacle of Senator David Pocock grilling Anthropic’s leadership over water consumption is a masterclass in political theater and a failure of technical literacy. It’s the kind of performance that makes for great headlines but ignores the physical reality of how global infrastructure actually functions. We are watching a repeat of the 1990s panic over "vampire power," where politicians fixated on VCR clocks while ignoring the massive inefficiencies of the industrial power grid.
The consensus is lazy. The media wants a narrative where "Big Tech" steals water from the mouths of thirsty citizens to cool humming black boxes. It’s a clean, easy-to-digest story of corporate greed versus environmental necessity. It is also fundamentally wrong about how cooling cycles, thermodynamics, and resource displacement work. You might also find this connected article interesting: Newark Students Are Learning to Drive the AI Revolution Before They Can Even Drive a Car.
If we want to have a serious conversation about Australia’s resource security in the age of generative AI, we need to stop pretending that a data center is a bottomless drain. It’s a heat exchanger. And compared to the alternatives, it might be the most efficient one we’ve got.
The Evaporation Fallacy
The "water use" numbers cited in parliamentary inquiries are almost always Water Withdrawal figures, not Water Consumption figures. This is a distinction that lawmakers routinely fail to grasp, either out of ignorance or because the truth doesn't poll as well. As discussed in detailed reports by MIT Technology Review, the effects are widespread.
When a data center pulls water from a utility, it isn't vanishing into a digital void. In closed-loop systems, that water is circulated to absorb heat and then returned to the source. Even in evaporative cooling systems—the ones that actually "consume" water—the fluid is being transitioned from a liquid state to a vapor state to manage thermal loads.
Compare this to the mining sector or large-scale agriculture, where water is often contaminated with heavy metals or lost to runoff that carries pesticides into the water table. Data center water is clean. It’s warm, sure, but it’s chemically inert.
The Efficiency Paradox
Senator Pocock and his contemporaries are obsessed with the input without ever measuring the output of efficiency. This is the Jevons Paradox in action: as we make a resource more efficient to use, we end up using more of it because the value it provides skyrockets.
Let’s look at the math that nobody in Canberra wants to touch. A single large-scale AI model can optimize the logistics of a national trucking fleet or the power distribution of a smart grid, saving millions of liters of fuel and gigawatts of energy.
If an Anthropic cluster consumes $X$ amount of water but enables an algorithm that reduces the water waste in Australian cotton farming by $15%$, the net result is a massive win for the environment. By fixating on the "thirst" of the server, we are ignoring the "hydration" of the broader economy.
Thermodynamics Doesn't Care About Your Optics
Every computation generates heat. That is an immutable law of physics. We have exactly three ways to deal with that heat:
- Air Cooling: Massive fans that require enormous amounts of electricity.
- Water Cooling: Using the high heat capacity of water to move energy away from the chips.
- Phase-Change Cooling: Complex, expensive, and often reliant on refrigerants with high global warming potential.
When politicians demand that AI companies "reduce water use," they are effectively demanding that they "increase electricity use." In a country like Australia, where the grid is still heavily reliant on coal and gas, trading water for electricity is an environmental disaster.
Water is a local, manageable resource. Carbon emissions are a global, existential one. By forcing data centers to move away from water-efficient cooling, we are inadvertently forcing them to draw more power from a dirty grid. It’s a classic case of fixing a localized "problem" by exacerbating a global crisis.
The Sovereignty Myth
The subtext of the Anthropic inquiry is a fear that American companies are coming to Australia to "steal" resources. This ignores the reality of data sovereignty.
If Australia doesn't host these data centers, Australian businesses will simply use API calls to servers in Virginia or Singapore. The computation still happens. The water still evaporates. The only difference is that Australia loses the jobs, the tax revenue, and the physical control over its own data.
I’ve seen dozens of companies try to "virtue signal" their way out of infrastructure costs. They move their workloads to "water-neutral" regions, only to find that the latency kills their product and the carbon footprint of the long-distance data transmission is higher than if they had just built a well-cooled shed in Western Sydney.
Beyond the Cooling Tower
If we actually cared about water, we’d be talking about the source of the water, not the quantity.
Smart data center operators are already moving toward recycled "grey water" or seawater desalination. They don't want to use your drinking water—it’s full of minerals that gunk up their expensive heat exchangers. They want ultrapure water or industrial-grade runoff that they can treat themselves.
The real "disruption" here isn't asking how much water Anthropic uses. It’s asking why the Australian government hasn't mandated that all new data centers must be integrated into district heating schemes. In Northern Europe, the "waste" heat from data centers is piped into homes to provide heating.
In Australia, we could be using that heat for industrial desalination or large-scale greenhouse agriculture. Instead, we have senators asking 20th-century questions about 21st-century hardware.
The Brutal Reality of Scale
Silicon is the new oil. You can either be a producer or a vassal state.
Every liter of water spent in a Sydney data center is an investment in the country's technical autonomy. If you want to run a modern economy, you have to pay the thermal tax. There is no such thing as "dry" intelligence.
The next time a politician points at a cooling tower and screams about a drought, ask them to show you the energy-to-water trade-off calculations. Ask them how many liters of water were used to grow the grain for the steak they ate at the parliamentary lunch.
The data center is the most efficient engine ever built by man. It turns electricity and cooling into insight. If you can't handle the heat, get out of the digital age.
Stop asking how much water the models drink. Start asking what we are getting in return for every drop.
Build the centers. Turn on the taps. Optimize the world.
