The shovel sellers are making the money: Gerard Minack on the AI not-quite bubble

Macro strategy expert Gerard Minack loves a chart. In his presentation at this year's Morgan Stanley Summit, the Minack Partners founder and former Morgan Stanley analyst, cycles through his handcrafted charts at a rate of knots.

But those charts tell a compelling story about where we are in the AI cycle, and where things could go next.

Are AI stocks actually undervalued? 

One interesting data point from Minack's presentation is that AI stocks are actually undervalued on a forward P/E basis compared to the rest of the stock market, due mostly to how their earnings have exploded as part of the AI capex cycle.  

But the AI gold rush prospectors will need to start making money soon if the shovel providers are going to pump the unprecedented levels of capex spend we've already seen enter the sector. 

If we take the hyperscalers - a collection of stocks Minack is calling the "AI 8" - the Mag 7 minus Apple and Tesla but adding Palantir, Oracle and Broadcom - their collective capex in 2026 is expected to jump on sell side forecasts from US$450 billion to more than $800 billion. And that's ignoring the capex spend of the private-for-now AI companies OpenAI, Anthropic and SpaceX.

And we're seeing the same correlation between US tech profits and capex that we saw back in the TMT (Technology, Media and Telecom) bubble of the late 1990s, says Minack. Part of that is the circular economy contained within the AI industry, where Microsoft buying Nvidia chips is fully logged as revenue for Nvidia while Microsoft amortise the cost.

The new shovel sellers

Plenty of money may be getting spent on AI, but it so far hasn't translated to tangible real-world revenue, says Minack. 

"Not many people selling AI services are making any money from it. The people that are making the money are the shovel sellers," he said. 

And now the "AI 8" are being surpassed by what Minack is calling the "Next 11", the second-order AI shovel providers like AMD, Intel, Sandisk, Micron and Cisco. 

Those 11 stocks were up an average of 140% from March to early June and have added more than US$2 trillion in market cap in that time - more than the total value of the ASX. It recalls 1999, where the Nasdaq was up 100% over the calendar year, despite recording five double-digit corrections.

And while US Tech sector earnings forecasts are up 83% since last year, Emerging Markets tech earnings forecasts are up 283%, thanks to two of the three key memory chip providers, Samsung and SK Hynix. 

Along with Micron, those stocks have seen collective earnings go from US$50 billion to $450 billion, and seen their collective market cap go from US$500 billion to $3.5 trillion. 

Exponential usage, declining profits

AI token usage has seen exponential growth in the last few years, but there's a "vicious fight for market share" amongst LLM providers. According to Minack's research, DeepSeek offers the cheapest token prices by far, but its models trail those offered by market leaders Anthropic and OpenAI. 

But he says many users might be happy to make the tradeoff of a non-cutting edge model if prices are cheaper. He makes the analogy to the automotive industry. Different carmakers serve different types of customers and the same could happen with AI. 

He quotes an associate who says that whenever the Chinese enter a sector, profits walks out, and that could be a problem for an industry trying to justify unprecedented levels of investment. 

And if the TMT cycle is anything to go by, equities peaked well before there was any widespread sense of a slow down in cash flow, capex and return on assets. 

"The market will not wait for the hard data to rollover before it senses there's going to be a problem in terms of returns." 

"I certainly know the digital technology that underpinned the TMT cycle was very important," says Minack. "But the market back then made a prediction on how profitable the companies providing the tech would be, and my view is they're making the same mistake again these days." 

And it's a separate argument to the perceived merits of AI as a technology. Ultimately, markets follow the money, and if AI services profits fail to materialise then the shovel sellers will stop making shovels. 

"Whether we find gold in the AI gold rush, or base metals or clay, doesn't depend on the technology. I'm agnostic on whether AI will be transformative." 

"This is not a valuation bubble, it’s clearly an earnings bubble if it proves to be unsustainable."