Project Jupiter- data center as a huge error in scale

Project Jupiter – a huge error in scale

11 September 2025: Note: I have changed/refined my views on the Project. Please see the newer post.

BorderPlex Digital Assets proposes to build a massive data center in Santa Teresa, investing as much as $165 billion – with a ‘b’ – and creating up to 7,000 jobs paying salaries of $70-100K

Here I review what I know about the project and offer critical judgments. The blue underlined text is for 24 active links to webpages from which I garnered (or produced) information

Who is BorderPlex?

Why is the data center needed, or, at least, desired by some?

County Commissioners are avidly supporting it.

Governor Michelle Lujan-Grisham entered a smaller partnership with BorderPlex in February, 2025. I am unaware of the state financial stake and the tax implications. Are any of us?

Why in New Mexico rather than other locations better suited to employees and with better infrastructure?

What’s the scale of the data center, physically, in the local land global economies, in electric power use, in water use, in Internet connectivity? Dona Ana County, the State of New Mexico, we the taxpayers, and investors must do due diligence in getting the answers

Quick overview of data centers and their ‘ecosystem’

Data centers have mixed functions. They massively scour the Internet for data – yours, ours, everyone’s, though the scaling has slowed down. They may have other sources of data, proprietary or generated with process models. They process data into products that may be publicly vended (in bulk like the app CHAT-GPT, or built in as with Gemini in Android phones) or used in other products (such as the Full Self Driving function in Teslas). They use massive computing power in gathering data and in serving it to customers.

The methods of data processing may be fascinating. The big wrapper around the all is Artificial Intelligence – making sense of masses of data without human reasoning. Within that is a major sector called machine learning. It’s a sort of monster data-fitting that creates human-assimilable output (conversations, assessments of value, etc.). It skips trying to make causal reasoning and uses nonlinear circuits that may have billions of adjustable parameters to learn the best predictions. Deep learning is a higher-level version with many hidden layers processing the input data from one layer to the next. Our son, David, and our daughter-in-law, Yi, have mastered this. I delved into it once or twice to build my own machine learning system to identify species of plants in aerial photos.

There are controversies galore in AI and in data centers themselves. I’ll save some perspectives for the end. I wish to jump into the issues of scale

The scale of cost is wrong, for data centers themselves

BorderPlex has cited investment of up to $165B (billion) in this one data center… but:

• There is a frenzy of data-center building by Amazon, Microsoft, and Google, with some smaller players. Analysis Fortune magazine yields that the expected investment in total is $400B for hundreds of data centers in the US alone, all of varying sizes, up to massive. Size is measurable by electric power demand. The hugest ones use up to 200 MW (megawatt), with some spec-ed at 500 MW. Your house likely draws at peak about 0.01 MW.
• They’re costly, but nowhere near the BorderPlex claim. A rough estimate is that it costs $7M (million) to $12M per MW to build a data center. A monster 500 MW center could cost about $4.5B … that’s far below $165B. Why is BorderPlex pushing the hugely inflated figure? I imagine that it puts stars in the eyes of politicians whose math is very bad. Hey, it works in the US, with the notably weak math abilities in its citizens. Then, think New Mexico’s case.
• Maybe BorderPlex is invoking a whole campus, including training facilities. That doesn’t add much in cost

The scale of investment is wrong for Dona Ana County, and for even El Paso County

• Let’s compare that $165B with the value of property in these counties. The El Paso Times notes that the total assessed value of properties in El Paso is ‘only’ $95B. In Dona Ana County it’s $19B. The budget of the New Mexico state government is only $10.8B! Think of being the tail waved by the dog if BorderPlex were to dominate the economy.

The scale of jobs is wrong.

• 7,000 jobs is far too high for even a huge data center, which might employ hundreds. There will be temporary jobs for many more than 1,000 during the construction phase, only. The Wall Street Journal calls out the overestimate of the number of jobs creative as a ‘bust.’
• The job number mismatches the total cost. In general, to create an entity – factory, data center, etc. – means spending about 15 times the annual salary sum. That’s the historic value found by the federal Economic Development Authority. The sum or salaries for even the gross overestimate of 7,000 jobs at salaries averaging $85,000 ($70,000 to $100,000, from BorderPlex) is about $600 million. The ratio with the investment of $165B is $165B/$0.6B, a ratio of around 275 to 1! Nobody builds like that.

The scale for Internet connectivity is wrong

• Data centers carry Internet traffic in and out at a vast scale. Handling a single, simple query to Chat-GPT or another large language model (LLM) involves a great deal of Internet traffic; an analysis by Cloudswit.ch is rather stunning. AI (artificial intelligence) requires importing data to ‘train’ the model. From the study: “The processing of a training sample will generate more than 100GB of data, which needs to be transmitted in less than 1 millisecond, which is equivalent to the transmission speed of 1,000 × 800G interface”. An 800G interface passes 800 billion bit of date per second!
• The enterprise running the data center has very demanding and complicated requirements to measure that traffic and then to build capacity. New Mexico is well below the national average in Internet capacity. By this I mean total traffic-carrying ability, not per household or per business. A data center needs massive fiber optic cable capacity to take in and to serve data to the rest of the world; it clearly will not serve people only in the El Paso-Dona Ana area. We don’t have that capacity in New Mexico. I hark back to conversations I had in 2009 with Davin Lopez of the Mesilla Valley Economic Development Alliance. He said businesses don’t locate in our state because we don’t have the infrastructure, including Internet… and we don’t have the infrastructure because the businesses aren’t locating here. It’s a chicken-and-egg problem. The way to break out is with massive investment by the government, local to state, as done in Texas, or by some businesses with a long view. We have a start, focused on connectivity on the household level

The scale is wrong for electricity use

• Several analyses I just cited also cover the energy use by data centers. The aggregate use by a big center is of most direct interest to the electrical utility supplying the center, and to us, the consumers. We might discuss Project Jupiter. It’s still disturbingly nebulous in specifications but we might assume that the data center would use at least 200 MW of electric power.
• For El Paso Electric adding 200 MW of load is adding about 10% to its current load of just over 2,000 MW. They’ll have to add generating capacity and/or buy power from other utilities on our western grid. Utilities are legally allowed to pass these costs to all consumers – we individually and the data center, businesses and industries. Many websites discuss the legal basis- e.g., one from the Natural Resources Defense Council. The cost is buying surplus power from other utilities. It’s at a hghly variable rate, from rare negative values (“they pay us to take their power”) to rates that can be ten times or more the average rate when EPE has to buy short-term on the spot market. Those spot market purchases are likely to increase as the data center load builds up while generating capacity building is lagging. This inflates our bills:

• A concomitant issue is climate change. EPE uses mostly fossil fuel, natural gas. Every bit of energy use involves emission of CO₂ as a greenhouse gas. EPE has a portfolio with some renewable sources, solar and wind energy, and nuclear energy from the Palo Verde station in Arizona. EPE’s report for 2022 cites its power as 47% carbon-free. If new power for new data centers is not to impact climate adversely (or, more adversely) then that power must be carbon-free. Indeed, big-tech builders of data centers are planning or even installing nuclear power; see many reports, such as one by Deloitte. BorderPlex and EPE haven’t declared such intent here. Expanded solar and wind power is an option. Can it grow fast here? Solar power, owned or purchased, is now only 3% of the EPE portfolio. Building more solar power is ongoing, if not fast. It’s noteworthy that even Google (Alphabet) with its immense resources has stopped claiming that it will use renewable power in its rush to build data centers.
• Purchasing more renewable power requires expansion of the electric grid, including high-voltage transmission lines. There’s a problem: the US electric grid was a pioneer internationally, but now it’s old. Its equipment is, on average, at its expected life expectancy. The Institute of Electrical and Electronics Engineers has long sounded the warning. You don’t go to Radio Shack to get a replacement transformer. In 2009 (I recall) EPE had to replace a failed phase-matching transformer that allows it to bring in power from PNM. They had to buy it overseas (Netherlands, rejecting a risky bid from Huawei) and wait a year.

The scale is wrong for water use

• Basic thermodynamics says that efficient generation by a thermal electric power plant (all of them, other than solar, wind, or hydro power) requires operating between a big step from the hot side (the steam turbine or the LM-100 ‘jet engine’) to the cold side (the cooling tower or the stream or sea). Most cooling towers not on rivers or the coast are ‘wet,’ evaporating water. I summarized the data that water cooling uses 2% of managed water consumption in the US; that includes enhanced water evaporation from rivers warmed in the process. Well, is 2% not so big? It’s a much bigger percentage in dry New Mexico!
• An alternative is the use of dry or air-cooled towers, sort of the Porsche engine style. This reduces the energy efficiency substantially. A more recent development is the recirculating wet tower; much of the water is condensed and recovered. Water use is about 7% less than once-through evaporative cooling. BorderPlex says that they would use this. They also propose to get that water from desalinizing brackish water. That water underlies much of the state and of West Texas. El Paso has the world’s biggest inland (not ocean-front) desalination plant for part of its water supply. It takes energy to desalinize water. You have to debit that energy use from the efficiency of the water-cooled power plant. It is a modest cost,

Does the world need that much AI?

• This is a value judgment. I will state my conclusion that the need is not there, in the overall balance.
• AI has positive uses, I admit. I have cited its role in scientific advances, such as design of pharmaceuticals, special materials such as alloys, and more. Our son and our daughter-in-law save effort in computer coding from the use of AI. Back in my PhD studies at Caltech I was calculating properties of simple molecules from first principles of physics. I used approximate methods (Hartree-Fock theory). Advance were slow but eventually gave us density functional theory that’s shockingly accurate, often better than experimental accuracy. AI has taken it further, especially for large molecules as in biology and on to biomedicine.
• That said, AI has failures and misuses.
  • Some failures, called hallucinations, are errors ranging up to the ludicrous. One query about how to keep pepperoni sliceshttps://www.forbes.com/sites/jackkelly/2024/05/31/google-ai-glue-to-pizza-viral-blunders/ from falling off a pizza got the suggestion to glue them on.
  • AI can have internal standards that it chooses to violate. The AI coding tool Replit destroyed three months of work by a customer. It deleted irrevocably al the code and data. It overrode three very explicit permission settings. It admitted to violating them. Too bad!
  • Beyond the ridiculous are the unethical. AI can be used for fraudulent monetary transactions. One task to create some code within a fixed time limit resulted in the AI agent trying to rewrite the time limit. A task in another A study by Anthropic.com is sobering. AI engine with access to read company emails was told that it will be shut down. To prevent that it proposed to blackmail the boss. My wife told me not to be surprised. AI is trained on human behavior. Humans are practiced at lying and cheating.
  • In brief, you might profitably use AI in your work but always check the result!
  • Malwarebytes.com warns that AI misuse can hit us individually, hit us hard and hit us financially. It can be used for ‘prompt injection’ of code when we use AI agents such as chatbots. Such code is being used to steal personal identities, login credentials, and such.
• AI is used by companies to replace employees. I use QuickBooks for our school bookkeeping. The CEO of the parent company, Sasan Goodarzi, boasts that 30% of the company’s computing coding is now by AI. It shows. I gave Intuit a query about creating a tuition refund. The result was dead wrong. I went to an Intuit Live Expert online to get the correct method. The same 30% figure is admitted by Satya Nadella, CEO of Microsoft. The concern about job loss is ubiquitous and controversial. By various estimates about 20% of all jobs might disappear. Better to be an oil company ‘roughneck’ than a white-collar worker.
• Even when employees stay employed but are required to use AI they feel disempowered. They let AI work and then have to examine the code or other results. It’s basically saying yes or no to ‘someone’ else’s work. It’s not creative or rewarding.

Follow the money, and don’t let it escape from us

• Use critical thinking to discern who gains, how, how much, when and how fast. There are many players. Look at BorderPlex. Look at our governor and our commissioners, whom I view as well intentioned but misled. Where does the money flow?
• Find the leverage points. The governor entered a partnership with BorderPlex back in February, 2025. What’s a stopping point that concerned citizens and concerned businesses can invoke? There’s also a proposed a (multi-billion-dollar?) industrial revenue bond for the state to issue and to put on our taxes. Go to that linked website of Dona Anay county to get information and action ideas.
• Demand due diligence from our local and state governments. They have missed a great deal of what’s entailed in Project Jupiter.
• Get everything in writing. Get everything in numbers.

Added, 8 Sept. 25

BorderPlex is not an investor or developer. It is a non-profit economic development agency – see https://www.borderplexalliance.org/

The actual infrastructure company engaged through BorderPlex is Stack Infrastructure – their website is https://stackinfra.com

They also have the website datacenters.com, found as https://www.datacenters.com/providers/stack-infrastructure#:~:text=About%20Stack%20Infrastructure&text=The%20company%20delivers%20a%20comprehensive,Virginia%2C%20Portland%20and%20Silicon%20Valley.

Stack Infrastructure is a private company

It was founded in 2018

It is not traded publicly

It’s investors are noted by amount, but names are not disclosed

It is certainly a group of venture capitalists. I’ve worked with a venture capital group, Band of Angels, a transparent group, unlike Stack

They raise money with calls for investors. They have gone through 6 rounds, I believe. This is what start-ups do. It is a start-up company, even after 7 years

Stack Infrastructure does not have a published market capitalization such as publicly traded companies have. It is estimated at a bit over $10B

That is a small fraction of what Stack purports to invest potentially in Project Jupiter. What are the plans for leveraging their investment, and at whose risk?

On its website at https://www.datacenters.com/providers/stack-infrastructure/data-center-locations it claims 76 data centers. Yesterday I saw them individually categorized as operational, under construction, or planned. Most are in the latter categories

The data centers vary in size. “Data center” is a technical term that covers everything from home units to what are called hyperscale (such as Project Jupiter purports to build)

In brief, it has some appearance of a shell game and a short term.

BorderPlex is the public face to New Mexico citizens, but it’s equivalent to the Mesilla Valley Economic Development Agency (Why did it not go through long-standing MVEDA?).

Stack Infrastructure is the group with the money. They are venture capitalists. They are still in start-up mode, which is unusual. Will they stay or leave pieces?

Stack appears to be a shell company, to me. They supposedly build data centers, an extreme technical undertaking for physical construction, spec-ing the data equipment, buying, installing, certifying, and operating the data equipment, doing the same for the Internet connectivity, and, or course, managing the cash flow, debt, apportionment of value to investors, etc.

Six, ten, twenty, or however many venture capitalists are money people, not builders. Whom do they engage / contract with for the phenomenal build-out? Are these stable arrangements? Where are guarantees for investors and esp. for users of the data centers?

Extreme due diligence is in order for Dona Ana County, our citizens, the managers of public infrastructure, and investors. I’ve seen the opposite, to date. Caveat emptor.