Analyst Ethics Warning: Paid Posts About AI Data Centers

This analyst was offered money to, without financial disclosure, undertake paid pro-AI data center posting on behalf of a major AI data center in Wisconsin. If you see pro-data center posts that look fishy, this may very well be why.

Pro-AI Data Center Social Media Posts Are Not to Be Trusted

Yes—we have seen coverage of financial analysts’ disingenuous statements on stocks like SPACE-X, with its valuation unhinged from reality and toxic debt from the Twitter acquisition and from Grok AI rolled in.

What concerns me now though is not only analysts that offer technology and financial insights for a living, but posts on social media. Why am I concerned?

Well, I’ll tell you.

Paid Posts about AI Data Centers Without Disclosure

I got a call recently from a longtime associate who’s been immersed in GOP politics in Wisconsin for more than 20 years.

He’s looking for folks willing to get paid to post on social media in favor of a major AI data center project in Wisconsin. I asked if these people would disclose that they are getting paid, and he said no.

So, we had a frank exchange of ideas. I told him his client had better learn that anything they do is going to wind up on the front page eventually. He told me he didn’t think that would happen.

To be clear—at Rathmann Insights, we write about software and technology, specific to the construction space. In recent years, most of our work has been under our client’s brand or has consisted of consulting with teams internal to the software company. More and more however, we are publishing material under our own brand as objective analyses and yes, software companies pay us for this. But you will always know who is paying us for what.

Always.

And even if a company pays us, well, they can’t pay us enough to lie. We may focus on the relative strengths of one product versus another or talk about trends that may play into the strengths of a client.

Not every software vendor will be comfortable with this. There certainly are vendors who prefer a “State Media” model where they dictate what is said about them. There are some who avoid working with analysts altogether, and they are often the vendors less confident in their value proposition or the modernity of their solutions.

Honest and probing insights are coin of the realm for buyers and specifiers of complex enterprise software but are an anathema to vendors whose statements cannot stand up to the scrutiny of honest brokers.

And this is why AI data center projects have to stoop to corrupt practices like this.

The AI Data Center Bust to Come

I was honestly surprised my associate reached out to me on this matter because I have a certain position on inference models for AI--they are not profitable or fiscally sustainable, and the risk of making major infrastructure decisions around them is extreme. There are better technologies now, and more coming, and we should remember all of the canal projects abandoned when the railroad came through. I am just super concerned about the degree of chicanery around this whole industry.

And while I believe there are broad societal and environmental issues that need to be addressed around AI, my concern focuses less on these and more on the fact that I can read a balance sheet. What do I see?

I see debt. Massive amounts of debt necessary because there is no sufficient cash flow for operations. According to JP Morgan:

“US hyperscaler bond issuance has surged from 2% of total USD investment-grade issuance between 2022 and 2024 to an expected 9% in 2026. Year to date, hyperscalers have issued USD 219 billion of investment-grade bonds, including USD 62 billion equivalent in non-USD currencies such as EUR, CAD, CHF, GBP and JPY.”

The amount of debt is one matter, and the cost of the debt is something else. With returns on treasury bonds spiking on uncontrolled deficit spending, the cost of cash is increasing and borrowers are becoming more selective about what projects they fund regardless of rates as high as 9.75 percent.

On top of that, a lot of this debt is being moved off the balance sheet into special purpose vehicles. This shift is significant enough that it could trigger a global recession when these debts, conveniently held off public balance sheets, go south.

But wait. This is just the start of the structural problems behind AI hyperscalers.

Illusory Demand for AI Hyperscalers

Communities that host AI data centers are not doing so to meet market demand for AI services by end user customers. They are acting at the behest of the entities trying to hide the coming bust cycle, who are lining their pockets now while leaving others to absorb the risk after a collapse.

How do they pull this off?

• Circular Financing: Hyperscalers invest in startup AI labs, which then pay the money back to the hyperscaler in compute capacity, creating illusory demand

• Round Tripping: Naman Goyal points to “OpenAI, Nvidia, Microsoft, Oracle, AMD, Intel, and CoreWeave — all seemingly feeding one another in an endless loop of ‘strategic partnerships,’ ‘cloud credits,’ and ‘GPU deals.’ Actual recurring revenue will need to come from large enterprises that consume the compute capacity, but there is no evidence of this demand. What we are being sold by the hyperscalers is a dream about a story about a myth.

• Massive Losses: The amount of money being shoveled into large-scale AI data center investments is mind-blowing. Yet these investments are tied currently only to red ink. OpenAI has been extremely opaque with investors, but in financials leaked ahead of an IPO we see a $38 billion loss for 2025. Anthropic meanwhile Anthropic carries an estimated $10-15 billion in net losses since 2021.

So just to sum up—AI has tremendous potential, but from what we see, this potential lies with smaller and more targeted models that delve deeply into vertical processes and might not even run in a data center. The greatest potential for business and industry lies in deterministic AI and more efficient technologies than the energy pig that is a large language model (LLM), which essentially acts as an inference engine guessing the next word or the next step or the next image.

These deterministic models may be configured with the help of an LLM, which in this use cases is a tool for human-computer interaction but not for operational AI. Our position is that most mission-critical activities are better handled by deterministic intelligence that can be audited, providing efficient yet responsible means to automate activities around business rules or process requirements.

The Importance of Integrity

A few times over my 30-year career, Ive been asked to do some sketchy stuff. Some of the things that I took exception to others may have taken no issue with. Other things, like a request in one enterprise software marketing job to create a fake LinkedIn profile to post about another company the CEO owned without letting on he was behind it, most other professionals would also refuse.

The point is that for someone who offers objective insight to the market,  a history of stubborn integrity is essential. And in times when powerful forces are trying to hide their motivations or identity from the market, this objectivity will be essential.

Really, paired with a bit of experience working in the enterprise space for construction, it’s all I’ve got to sell.

So, when you see people posting in favor of your local data center project, please realize they may have sold their integrity not as a core component professional services, but as chattel property they gave up for a few bucks.

Maybe they understand the broken economics behind AI data center capacity and construction, maybe they don’t. But don’t take those posts at face value.