How long 'til we go full Jensen?

Jensen Huang is uber-bullish on the AI revolution. But there are cracks emerging in the AI investment boom narrative.
Greg Canavan

Fat Tail Investment Research

Strangely, Nvidia founder Jenson Huang might actually be the contrarian in the room right now.

As you would expect, Jensen is uber bullish on the AI revolution. In a recent interview on the BG2 Podcast, he said he thinks OpenAI will be the first $10 trillion dollar company.

Given that he’s just committed to a long-term investment of US$100 billion in the company, he would say that.

His views contrast sharply with those of many others in the market. As the Chanticleer column in the Financial Review wrote on Monday:

There’s a lot of bubble talk swirling around markets. Monday morning desk notes were full of it.’

Hmmm…is it a bubble if the consensus call is that we’re in a bubble?

I don’t think that's how it works.

That share prices are expensive is not in question. However, a bubble is as much about investor psychology as it is about prices and valuations.

It’s when formerly rational commentators and analysts talk about new eras and paradigms. When the old rules no longer apply. And that if you don’t get on board, you’ll miss out for good.

It’s not when Wall Street strategists are warning about a bubble.

You’ll know we’re near the end when everyone goes full Jensen Huang.

What could turn everyone into Jenson Huang?

Simply, for the ‘bubble’ to keep expanding. 

The worst thing for many investors, professional and amateur alike, is to be on the sidelines, or out of the 'hot stocks' while everyone else is making money.

They may have moved to the sidelines because of concerns about a bubble and valuations…but find it increasingly hard to stay out as prices keep moving higher.

The final blow-off top phase of bubbles is usually the result of formerly cautious investors throwing in the towel and joining the speculative frenzy.

If that is going to occur (and there are no guarantees it will) it hasn’t happened yet.

Data centre boom needs commodities

Everyone knows about the huge pipeline of data centres to be built around the world. That investment will take place alongside massive investments in grid expansions. As Reuters reports:

U.S.-based utility PG&E Corp said on Monday it plans to spend $73 billion by 2030 for transmission upgrades to meet the data centre-led surge in electricity demand.

That’s just for California. It’s the tip of the iceberg.

These grid expansions will use lots of copper, aluminium and steel, so even as an AI backwater, Australia will capture a slice of this investment spend.

The point is, the AI data centre expansion is going ahead with full force and grid upgrades will occur alongside it. This capex boom could have years to run.

But I haven’t swallowed Jensen’s Kool-Aid just yet.

Here’s where I get confused about his grand vision for an AI world…

Capital destruction looming?

An AI factory, or data centre, is stuffed full of the latest ‘GPUs’, or Nvidia chips. By the time the factory is complete, Nvidia will have probably designed a newer and more powerful chip.

What happens when a data centre is five years old? Will it need a complete GPU replacement?

I don’t know. But these things are expensive. That’s why Nvidia is the most valuable company in the world.

Thinking about this issue, I decided to have a look at Next DC’s (ASX: NXT) financial accounts.

What I saw shocked me…

For FY25, its depreciation charge was nearly 70% of gross profit!

Let me explain what that means…

In FY25, NXT earned revenue from existing data centres of $427.2 million.

To get the gross profit, I subtract ‘direct costs’ of $77 million (basically the cost of powering the data centres) and data centre facility costs of $46.6 million.

That leaves gross profit of $303.6 million, a solid margin of 70%.

From this gross margin the business has to cover other costs like labour, interest expense, head office costs and depreciation.

Depreciation is an estimate of how much a company’s fixed assets are ‘used up’ each year.

The depreciation charge for NXT was $208.4 million, a whopping 70% of gross profit. For comparison, a quality capital-lite business might be around 5%.

If you know of any other business where depreciation eats into profits so much, let me know.

The company did acknowledge this in its FY25 annual report:

While the Group generated a statutory loss after tax during FY25, this remains consistent with the Company incurring upfront costs linked to its rapid expansion program, such as depreciation and finance expense, ahead of generating increased revenues from those investments.

Maybe I’m missing something. It wouldn’t be the first time. But I didn’t think you depreciated assets that weren’t generating revenues yet!

A look at NXT’s cash flow statement below gives another insight into its capital-intensive nature.

Operating cash flow of $222.6 million covered only around 14% of the total capital expenditure bill of $1.57 billion.

It had to raise $678 million in equity and draw down on existing cash reserves to fund its growth, ending the year with $243.7 million in the bank.

NextDC's FY25 financial results (Source: NextDC)
NextDC's FY25 financial results (Source: NextDC)

This seems like a lot, but it isn’t for a capital-hungry data centre builder. It will be increasing debt or issuing new equity in no time. After all, its capex budget for FY26 is around $2 billion.

Not a good investment

I don’t mean to criticise NXT. It’s building essential infrastructure for the age of AI.

But investors need to understand that that doesn’t make it a good investment. At best, this will be an average return on capital business over the long term.

One of the characteristics of infrastructure assets is that they are generally long-life, low-returning assets with predictable income streams. That’s why financial engineers love them. They stuff them full of debt to goose the return on equity.

NXT may have predictable income streams, but they’re not owners of long-life assets. The buildings are, but the racks of chips that live inside them will need constant reinvestment/replacement to keep up with Nvidia’s innovation cycle.

If this is any indication of how the global data centre rollout will unfold in the years ahead, then we’re all in trouble.

Investors will soon learn that depreciation is not just an accounting entry. It is a real-world estimate for the actual depletion of assets. And in the world of AI, asset (chip) value depletion happens very quickly indeed.

This bubble might keep expanding until it sucks in more believers into Jensen’s orbit, but it's on very shaky foundations.

Want more like this?

Enjoy unfiltered investment analysis? That's just a taste of the daily insights at Fat Tail Daily, covering everything from tech to miners to economic trends.

Expand your perspective - subscribe for free here.
........
All advice is general in nature and has not taken into account your personal circumstances. Please seek independent financial advice regarding your own situation, or if in doubt about the suitability of an investment. Any actual or potential gains in these reports may not include taxes, brokerage commissions, or associated fees.

3 topics

2 stocks mentioned

Greg Canavan
Editorial Director
Fat Tail Investment Research

Fat Tail is Australia’s largest independent financial publisher. Greg is Editor of its flagship newsletter, The Fat Tail Investment Advisory, where he writes market commentary and looks for out-of-favour ASX 200 stocks on the cusp of a...

I would like to

Only to be used for sending genuine email enquiries to the Contributor. Livewire Markets Pty Ltd reserves its right to take any legal or other appropriate action in relation to misuse of this service.

Personal Information Collection Statement
Your personal information will be passed to the Contributor and/or its authorised service provider to assist the Contributor to contact you about your investment enquiry. They are required not to use your information for any other purpose. Our privacy policy explains how we store personal information and how you may access, correct or complain about the handling of personal information.

Comments

Sign In or Join Free to comment