Demand for the computing power that undergirds AI models — "compute," in the lingo of the industry — has skyrocketed.

Tech giants are committing hundreds of billions of dollars to construct massively resource-intensive data centers across the country, but the aging power grid and an increasingly resistant public are quickly turning these expansion efforts into nightmare.

As an alternative, many AI companies — including Elon Musk's SpaceX — are vowing to move the entire affair into Earth's orbit, saying their orbital data centers will harvest the Sun's energy around the clock and convert it into raw computing power.

But the emerging consensus is that the plan isn't just overly ambitious; it would require several technological revolutions to overcome glaring engineering challenges.

A scathing takedown video — a collaboration between Irish aeronautical engineer Brian McManus from the YouTube channel Real Engineering and tech publication IEEE Spectrum — discusses the many pain points of Starcloud, a Y combinator company that raised $170 million earlier this year to develop data center spacecraft with the help of SpaceX, and how the white paper outlining the project feels far more like it was dreamed up by an AI than a team of human experts.

"It really seems like anyone with some renders and a white paper written by someone being gassed up by an overly agreeable AI can get VC funding these days," McManus argued. "Billionaires will attempt to pull the rug over your eyes and convince you that this technology makes total sense, but reality is, this technology is dumb."

The scathing rebuke comes not long after SpaceX's IPO, which turned the rocket company into one of the most valuable companies in the world over night. A huge chunk of its multitrillion-dollar valuation is tied up in Musk's orbital data center vision, setting enormous stakes — and once again illustrating how much Wall Street is leaning on the richest man in the world's beliefs, despite his abysmal track record when it comes to making good on his promises.

The engineering challenges of building out a constellation of enormous data centers in space are numerous.

For one, keeping piping hot AI hardware cool as it's pushed to its limits is extremely difficult and requires sophisticated equipment even here on Earth. In the near-vacuum of space, it's even harder, as all of that heat can't simply escape, requiring an extensive network of pipes running coolants, presumably along the rows of solar arrays, according to McManus.

In the case of coolant fluids like glycol, each data center would have to pump over 150,000 pounds of the stuff per second, which is like "emptying an Olympic swimming pool in 40 seconds" — rates only common among gravity-fed hydroelectric dams.

To achieve an advertised capacity of five gigawatts of compute, Starcloud is looking to go big. Each data center, including its enormous array of solar panels, would cover 1.6 square miles, nearly 5,000 times the surface area of the International Space Station's panels.

The mass could therefore balloon quickly thanks to the required cooling hardware.

"Even ignoring the pumps, coolant, radiation shielding, fuel, inertia wheels, structures and other stuff, Starcloud's station exceeds a 113 million kilograms," McManus argued in the video.

That's "more than an aircraft carrier sitting in orbit," he said. "More than six times the total mass launched into space in history."

All of that surface area could open up these satellites to damage from the millions of items of space debris already cluttering our planet's busy orbit. Even the smallest pieces could punch a hole in the panels, requiring costly repairs greatly complicated by the necessary journey into space.

It's a very real problem SpaceX is already all too familiar with. The company revealed that its Starlink network of internet satellites had to perform 300,000 collision avoidance maneuvers in 2025 alone.

And on top of all that, sending even crunched data through the emptiness of space could require extensive oversight.

"Ionizing particles passing through satellites will burn out a transistor or flip a bit of information stored inside," McManus explained. "This would result in the mother of all AI hallucinations without a software constantly checking results." Existing computers on board the ISS have to run redundant calculations and compare results to "weed out corrupted data."

The price tag for the envisioned Starcloud project also appear to be pulled out of thin air and are "overoptimistic on launch weights and launch costs," McManus argued.

Maintaining the enormous network alone would also be a massive and costly undertaking. The lifespan of current AI chips is only two to four years — and that's on Earth, where degradation is far less of an issue when compared to the extreme environment of space.

In short, Starcloud appears to be serving as an AI-pilled billionaire's pipe dream inherently designed to appease on-edge investors.

"This is just one early rushed concept to fundraise and move on," McManus concluded. "In the ever evolving world of tech, first movers are being heavily rewarded."

More on orbital data centers: Elon Musk's Orbital Data Centers Are Staggeringly Huge

Nvidia says it may have a potential answer to one of the AI boom's biggest environmental concerns: water consumption.

As companies build ever-larger data centers, the chipmaker says its newest server design could, in some climates, reduce on-site cooling water use from millions of gallons per day to nearly zero.

What happened?

According to Fast Company, a large data center may use up to 5 million gallons of water each day for cooling — about as much as a town with tens of thousands of residents. 

That level of demand has made AI infrastructure more controversial in communities already concerned about drought, overburdened utilities, and rising costs.

However, Fast Company reported that Nvidia's Vera Rubin platform, the company's newest AI servers, is intended to run at significantly higher temperatures than earlier versions. That shift enables a closed-loop liquid-cooling setup instead of one that depends on evaporating water to remove heat.

Fast Company explained that in the new system, a water-and-propylene-glycol mixture enters at around 113 degrees Fahrenheit, absorbs heat from the chips, and can then rise to 131 degrees Fahrenheit before exiting the system to cool back down.

Since that fluid can be cooled again without evaporation, many sites may no longer need large amounts of fresh water on-site for cooling. Nvidia says that in many climates, water use for that purpose could fall to "close to zero."

Why does it matter?

AI's rapid growth is closely tied to the energy grid and water tables. Training and running advanced models require enormous amounts of electricity, which can strain local utilities and, in some cases, raise concerns that households could eventually shoulder infrastructure costs through higher bills.

Alongside heavy power demand, AI data centers can also consume significant amounts of water.

Nvidia's cooling change appears to address one part of the problem, but not the broader reality that AI expansion also means more servers, more power use, and more pressure on local resources.

Despite concerns about energy and land use, Nvidia's breakthrough opens the door to less water-intensive data centers. 

Josh Parker, Nvidia's head of sustainability, told Fast Company: "The 45-degree intake temperature — that is really the newest innovation that's really transformative."

Better cooling may ease one major problem, yet it does not fully solve AI's larger resource footprint as data center construction continues to accelerate.

Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.

A quick AI prompt may seem inconsequential, but experts say the systems behind it have tangible environmental effects. 

Each chatbot answer or generated image depends on data centers that use huge amounts of electricity and water, and that demand is climbing rapidly.

What's happening?

With AI tools spreading quickly, concern about the resources behind them is rising too. Last year, global data centers consumed 448 trillion watt-hours of electricity. That's more than every country in the world except 10, and that total could more than double within four years, according to Newser.

To put it into perspective, producing an AI video can be comparable to 42 hours of an efficient light bulb staying on and about a gallon of water. An AI text reply uses the same electricity as about two and a half minutes of a light bulb, and ChatGPT alone handles roughly 2.5 billion text prompts per day.

Researchers say the public still cannot fully measure AI's footprint because companies reveal little about what their systems consume. Cognitive computer scientist Sasha Luccioni put it bluntly: "AI is going in the opposite direction to decarbonization efforts."

Why does it matter?

Power use is only part of the issue. Kaveh Madani, a water scientist and co-author, said that by 2030, meeting data centers' electricity needs alone could require nearly 2.5 trillion gallons of water, and that's not even including the water needed to cool the hardware. He also stated that AI is using enough drinking water to supply the entire world for 1.7 years.

AI is also closely tied to the energy grid. Every query passes through servers powered by local and regional electricity systems, so when demand spikes, utilities need to generate more power quickly. That cost can be passed on to everyday customers as well.

AI may help optimize systems, increase output, and accelerate research, but the challenge is that those potential benefits come with real risks. 

What can I do?

Being more selective about AI use is one of the biggest practical steps you can take. If a task does not truly require a chatbot or image generator, skipping it can help reduce unnecessary energy and water consumption.

As Luccioni says, "You can generate a chocolate chip cookie recipe with Claude, or you can open a damn book." Low-value, repetitive prompts add up to massive ecological strain.

Use a calculator for math, a map app for directions, regular search for simple questions, and maybe ask your friends what good movies are out instead of AI.

People who want to avoid the automatic AI-generated search summaries can try appending "-ai" to a Google search or turning off the feature in settings. When they do use AI, keeping prompts concise can help, as more information generally requires more computing power.

Luccioni's advice is a breath of fresh air in current times: "You are not obliged to use AI for everything."

Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.

Traditionally, data center operators have recommended an ambient temperature of 18 to 27 degrees Celsius (approximately 64 to 80 degrees Fahrenheit) for data center equipment. But in an unexpected twist, Nvidia's new 100% liquid-cooled AI data centers are running at a feverish temperature of 45 degrees Celsius (113 degrees Fahrenheit). This may seem counterintuitive, but Nvidia asserts that this cooling method is one of the biggest efficiency leaps in data center history.

In Nvidia's blog post by Josh Parker, head of corporate sustainability at Nvidia, it was announced that Nvidia DSX AI factories are hosting Nvidia's Rubin AI infrastructure with no fans or cold aisles. The idea that a data center must feel like a walk-in freezer is explained away as a misconception. All that was left was to find a way to transition from the hybrid liquid-and-air cooling to a pure liquid-cooling solution.

So, how does the Nvidia DSX design work? Liquid coolant enters the chip at 45 degrees Celsius, absorbs heat from across the chip's surface, and exits at 55 degrees. It was observed that processors can continue to operate at full performance at this temperature, and the process doesn't cause them to degrade. This is a huge boon for data centers, considering that every one-degree-Fahrenheit increase in operating temperature translates to up to 5% savings in energy costs. And with the U.S. pushing for data centers to pay for grid access, the timing couldn't be better for Nvidia.

Read more: 7 Companies That Are Owned By Elon Musk

What NVIDIA's high-temperature cooling means for consumers

The mention of an all-liquid-cooling data center may raise alarms for everyday people who are concerned about AI data centers doing more harm than just raising prices. These data centers are already consuming a reported 5 million gallons of water per day, to the potential detriment of nearby communities. However, Nvidia claims that its DSX AI factories have "zero water consumption" due to a dry-cooler-based design featuring a closed-loop system with no evaporative water cooling.

With that being said, this energy-efficient high-temperature cooling method might find its way into your home computer someday. For the CPU and GPU in your home computer, manufacturers explain that you should target a temperature as low as 30 degrees Celsius (86 degrees Fahrenheit) when the system is idle or under light use. That number might hit the new 45-degree standard if Nvidia's partner companies, such as cooling system manufacturer Motivair, start to implement similar engineering into consumer products.

If Nvidia is saving big on energy costs at the data center, does that mean the price of Nvidia's AI-powered products will go down? Not necessarily. As stated by Parker in the original blog post, "AI workloads are not getting lighter. The compute demand driving data center construction is growing faster than almost any other category of infrastructure investment." It's still very much a time when you should not wait to buy a new graphics card for your PC.

Enjoyed this article? Sign up to BGR's free newsletter and add us as a preferred search source or follow us on Google for the latest in tech and entertainment, plus tips and advice you'll actually use.

Read the original article on BGR.