Reason 11 For Not Building Data Centers Now

A few stories recently caught my eye about moving data centers away from where we live, far away, in fact.

 China's government plans to test space-based data centers starting in 2026, with a rollout of ones with costs as low or lower than earth-based centers slated for the 2030s. Meanwhile in the States, Amazon's Jeff Bezo has hatched similar plans. Not to be left behind by his billionaire rival, Elon Musk wants to use his Starship mega-rockets to orbit data centers. Then there's Google Starcloud, with a test planned for 2027. That system would also depend upon something like Starship to make the venture cheap enough to construct.

At first blush, it seems like a crazy idea, but as I considered the benefits versus costs, it makes sense to move this industry skyward:

• With reusable rocket boosters, costs to orbit per kilogram have dropped radically in recent years. If Starship prospers and others copy its model, we'll see the door opened for very cheap rocket launches.
• We have abundant solar energy in orbit. There's no need for generators, nuclear reactors, or natural gas.
• We also have easy cooling, without using water. Rotate a satellite and you have a solar heating on one side, the utter cold of space on the other.
•  Space-based data centers need not be huge to do their job. They could be a constellation of large satellites that talk to each other, as Starlink does already. Right now, however, big centers seem to be the model for space-based construction.
• If we do go big, we know how to do this already thanks to the International Space Station. 
• Parts of orbital centers can be replaced with one launch, and the old centers can be upgraded by robots or small enough to burn up on a deorbit. 
• Beaming data to ground stations 200 miles away on Earth is not a problem. We already do this.

My hope is that this technology will mature fast, to avoid more environmental and social disruption on Earth. And closest to home, I hope my County's Board of Supervisors pays attention, before we end up with a huge and obsolete building placed on agricultural land next to residences.

One issue that does bother me, beyond the possible climate-change effects of launching so many rockets?

It's called The Kessler Effect (or Syndrome). Readers may have seen the film Gravity, which one explosion in space results in a cascading set of collisions and, consequently, a massive cloud of space debris hurtling around 10 miles per second, chasing an underwear-clad Sandra Bullock.  

How serious is a collision in space? I once saw a piece of Space Shuttle Challenger's front windows, removed from a mission before the craft's 1986 catastrophe. The section of thick glass was damaged badly by hitting a tiny fleck of paint tossed off some forgotten rocket-booster, perhaps decades earlier. Now imagine millions of these objects, large and small, forming a cloud around the Earth, making any rocket launches an exercise in futility, destroying telecom networks, and grounding human space travel for many decades. Or centuries.

A center as large as Starcloud makes for a huge target, were the Kessler Effect to begin. 

Belatedly, firms and governments are considering ways to mitigate space debris and harden orbital infrastructure. Let's hope they get is right, as they'll only have one opportunity. I'd like to get my data from the heavens, not from Earth with more carbon pollution and wasted groundwater.

image: Starcloud center from Google video 

A Strangely Easy Idea No Writing Teacher Has Discussed

I may be out of the loop here, but I've come up with a simple idea that may be grant-worthy. Yes, steal it. Beat me to the grant. The consequences for students' learning may be too great for me to worry about being the first to put my name on it.

What I propose is simple: a writing environment online with the ease of Google Docs, but with one enormous feature disabled: copy/paste. In that regard, it would resemble the Respondus Lockdown Browser. 

The difference? Faculty would organize their classes and assignments there, as with a learning-management system, but they would pick certain AI tools that could employed at each stage of the writing process. The tools would work only from within the writing environment, and the final stages of each step--from rough-draft to revision to later drafts--would be sent from within the browser to the faculty member, along with how each writer used any AI tools permitted.

In that regard, my writing environment looks something like an educational version of Grammarly that an executive once demoed for me. In that product, all work copied and pasted into to interface would be watermarked and AI tools could be disabled or enabled by the user. I'd put that power in the hands of the course instructor, where it belongs. As students tell me again and again, using AI is simply the norm now, and we faculty are wasting time focusing on detecting plagiarism with AI detectors I don't trust. Time and again, studies I've read indicate that they turn up too many false positives. 

My idea requires some coders adept at building a good client and we'd have to negotiate licensing for AI programs such as Research Rabbit, an LLM, a image / slide-deck generator, and multimedia tools such as ElevenLabs' podcast and text-to-voice generators.

By placing the tools within the writing environment and forbidding copy/paste from outside, drafting by cognitive offloading would no long be an issue. 

I'm currently talking to a few colleagues who run our AI cluster on campus. I hope to have something sketched out, and if we can find the coders and money, tested in 2026.

Creative-Commons Image courtesy of oercommons.org 

Ten Arguments Against Hyper Data Centers

 

We lost the vote in Goochland County; 4 of 5 Supervisors voted to approve a Technology Overlay District (TOD) and Technology Zone that could include data centers. We learned that a likely center would be two million square feet, or 3000' x 200' in size. That's three times the length of USS Nimitz supercarrier or twice the square footprint of the Short Pump Town Center, a huge outdoor mall about 20 minutes from where I live. I've always hated Short Pump and that mall in particular; it was once prime farmland where a friend stabled her horse. It's now a suburban asteroid belt.

The proposed data center is worse. And soon it may be utterly obsolete. I commend to you articles by Bryan Alexander and Noah Smith about why this entire industry could soon find itself in deep trouble and why that matters to the larger economy. 

Other localities will be fighting data-center behemoths that use enormous amounts of energy and water. Their parking lots create heat-islands and they lower property values. It's likely we citizens will be taking the county to court, as plans were modified at the final moment to include 900 additional acres without public discussion. The entire process seemed rushed through, perhaps spurred on by money from industry, before a new Governor in January. She has promised to develop a statewide strategy. I hope so, and perhaps it can preempt the sort of hasty decisions made here. My state has more data centers than any other.

So what are arguments that money-hungry county officials will heed? Quality-of-life issues may help, but I think environmental concerns, sadly as usual, fall on deaf ears. Our nation's brand of capitalism, one I despise, values short-term thinking and profiteering. I'm a Distributist, not a Socialist; I want capital held and decisions made by the largest number of citizens possible. I want data centers small and as rare as possible. 

In any case, here are 10 technical and economic arguments to make with your local officials as you organize against this bonanza. They are based on my remarks at our recent county meeting. You can share these in 3 minutes at a meeting, if you practice! 

1.  AI drives this rapid growth in data centers. 
2. The AI industry is not turning a profit and has shown only limited ROI for hundreds of billions in venture capital and now, risky loans. Subscriptions make up a tiny portion of revenue.
3. The current model of AI relies on “brute force” computing to simulate human reasoning. This method requires huge data centers that drive up electric bills for ratepayers, and it uses lots of ground water. The Microprocessors in data centers need to be replaced every few years, meaning short life-spans for their hardware. 
4. Real “Artificial General Intelligence” (aka AGI or “Superintelligence”) is unlikely with silicon-based semiconductor technology, yet AGI is the stated goal of OpenAI, Anthropic, Meta, and Google.
5. Without more powerful new models, investment may well leave this sector, as 80% of firms that have already deployed current AI models have shown no gains in productivity. 
6. While Meta and Google have varied sources of income, other big AI firms are one-trick ponies. If the gold-rush turns into a bubble that bursts, it will harm the entire economy. What would happen in your locality when a data center goes dark, its owner bankrupt?
7. NVIDIA just announced a new Spark workstation that brings hardware-based AI to developers’ desktops. Coders no longer need to send work to data centers for processing. 
8. Soon consumer devices will have this capacity.  When we have AI-on-a-laptop or phone, which is Apple’s goal for Apple Intelligence, the data-center boom will likely go bust. 
9. Does your locality want to build power-hungry, thirsty data centers that may be obsolete in a few years? 
10. Quality-of-life issues are a form of return on investment. Start with that, not data centers.

Image source: first try with ChatGPT 5 for prompt "Generate an image of a two million square foot hyper data center in ruins." 

AI Bubbles and My County: Saying "Not Yet" to Big Data Centers

Note to readers: A version of this post will be submitted to our Board of Supervisors as a formal letter, as well as to a local news outlet. We'll see if it even makes a ripple. 

Even those wary of AI agree that it can do clever work. I use it regularly with students to shape their writing ethically and to design assignments. But can the industry make enough money to continue its progress? A great deal hinges on that answer, including development on the western border of the Richmond Metro Area.

A proposal before Goochland’s Board of Supervisors would allow data centers, perhaps powered by small nuclear reactors, to be built along the Route 288 corridor. Even without considering the environmental hazards, as one who researches the industry, I’ve found economic risks of rushing off quickly to join this AI gold-rush.

CEOs constantly make utopian claims about AI's future, but numbers cloud those predictions. Subscriptions such as mine meant under $2.5 billion in revenue for ChatGPT maker OpenAI in the second half of 2024, according to sources that include The Wall Street Journal and The New York Times. The rest came from venture capital or circular investing, such as Nvidia's and Microsoft's massive stakes in OpenAI.

The firm estimates that building out its data centers would cost $400 billion. For such ambitious plans, companies need a lot more revenue to balance their books. An "AI Bubble" may be inflating, as several economic journalists have warned. Writing for The Atlantic, Rogé Karma notes that “In the first half of this year, business spending on AI added more to GDP growth than all consumer spending combined.” A crash in the AI sector would deal devastating blows to the economy, with local effects when superfluous data centers go dark.

For his part, OpenAI’s CEO Sam Altman focuses on big ideas that seem straight from science fiction. He writes that humans and computers will become one form of life soon, which he calls “The Merge.” I'll quote from his blog, "unless we destroy ourselves first, superhuman AI is going to happen, genetic enhancement is going to happen, and brain-machine interfaces are going to happen." Not all of us would call such things progress, but history shows us that technological breakthroughs take time and a lot of money. Quiet supersonic airliners and nuclear fusion plants have inched closer to reality, over decades.

It well may take that long for AI data centers to not increase ratepayers’ electric bills. We’ve seen that already from existing data centers in the Commonwealth. Before we consign so much open land to data centers locally, let’s consider what experts in the field who are not CEOs or marketers have to say.

A robotics researcher, hired away from my university by Google, insisted to me that human-level thinking cannot be engineered; at best we might imperfectly simulate it, at great expense via a "brute force" computing method run in huge data centers. I recently noted in class how we might improve that, given faster computing using fewer chips, following what’s called Moore's Law. A student challenged me, claiming that silicon-based chips have not enjoyed the leaps in power and speed we expected annually up through the early 2000s.

Several academic papers I looked up support that idea. Without new types of microprocessors or exotic quantum computing, we are stuck with today’s data centers. Concurrently, our AI models remain far from perfect. Alex Reisner of The Atlantic, as well as academic researchers, found developers fudging benchmark tests that measure AI's smarts. They give the software test questions in advance, a tactic beloved by college students for many years. When AI is faces novel tasks, however, it scores far lower. Rogé Karma’s investigative work bears this out; productivity of coders dropped by 20% when working with AI, as they spent more time correcting mistakes.

Other Bubbles have popped with local consequences. A few years ago, I addressed The Board of Supervisors, whose agenda that night included motions to increase suburban-style growth in the east end of the county. I opposed this for many reasons, but one stood out: we had made zoning changes before based on empty promises, when we developed an office park called West Creek for an anchor client, Motorola. Their semiconductor plant would, in theory, have provided many hundreds of jobs.

My late father-in-law, Edward Nuckols, had addressed the Supervisors in the 90s. He was a respected master mechanic and businessman with deep ties to the community; his auto shop, unlike Motorola, had been in operation here for decades. Ed warned that big companies change their minds.

So it came to pass: Motorola backed out of their promises and Goochland was left with over $100 million in debt. One Supervisor reminded me that the county had little choice but to search for other revenue streams to repay that sum. While a fair counterargument, it does not seem we have learned much. Even if Altman and other CEOs are right, such men have no ties to any locality save the posthuman utopias they dream of inhabiting. $100 million of debt means nothing to them.

I’d love to see cheaper, more sustainable AI as a partner in our work. Today's models such as ChatGPT 5, Anthropic Claude, as well as lesser known but powerful tools such as NotebookLM or Research Rabbit, greatly help my students. We engineer prompts to find credible sources that are not hallucinated, we check human-created work for accuracy and voice, we write scripts for podcasts that AI converts to broadcast-quality audio. Plagiarism, the bugaboo of academic AI, concerns me less than does helping my students gain skills for workplaces slowly adopting AI.

My classes emphasize critical thinking when working with AI. So I'll ask Goochland's Board of Supervisors to do some critical thinking and research not influenced by the hype of Silicon Valley's billionaires. Altman claims intelligence will soon be "too cheap to meter." Older folks will recall that fable about electric bills, one made in the 1950s and 60s.

Otherwise, we could be left with more debt and large, empty buildings crumbling in the rain along 288. 

Creative Commons Image: Bioethics.com