The AI Trade Was Never About Chips

For the first time since the AI boom began, investors are no longer betting on the same future…

Since 2023, the AI trade has been remarkably simple to follow:

Everyone won.

But,

That relationship has completely changed.

Markets are now finally asking a question that’s been ignored for years:

AI models are missing on their previously promised revenue targets, so the semiconductor companies will have a harder time delivering for the hyperscalers who backed them.

If that’s the case, capex stops, and that’s why hyperscalers and semiconductors have broken off their relationship.

What used to be one AI Trade has now become four different bets:

That’s exactly when Goldman Sachs started asking questions.

Jim Covello, the bank’s head of equity research and strategy, had something to say about the whole AI trade.

Surprisingly, it is very similar to everything I broke down for you in this three-part AI deep dive series:

Click here to access the report.

Today, I’m going to dissect all of his talking points, and lead you to one of the easiest trades in the age of AI.

Because,

AI was never about chips, it was about enterprise ROI.

Covello’s biggest point wasn’t whether AI will succeed as a technology…

His biggest point is that a return on the trillion-dollar capex wave remains unclear for both hyperscalers and cloud computing companies.

Here are all of his talking points to be covered today:

Let’s start with the first point:

$1.3 trillion in revenue.

This is the consensus projection being given to the markets when it comes to AI model revenue generation.

That includes Anthropic, OpenAI, and some of the smaller ones like Gemini and Copilot.

What’s not being shown in this chart is the amount of commitments that have been racked up by these AI models.

Which, if the current pace is kept, will reach over $5 trillion by the end of 2026.

That’s a more than fivefold gap not accounting for the six years have yet to prove these revenue assumptions correct.

Look,

I’ve talked about this before in my Adobe Deep Dive.

The only way for these commitments to be met is if AI models manage to land enterprise software contracts, bringing in enough adoption, case studies, and revenue to justify the technology.

This is what Covello mentions as one of the biggest risks to AI.

If these revenue targets aren’t met, then hopes for a net ROI in due time go up in smoke…

And so does capex.

In fact, Amazon has recently announced they will raise one last $25 billion round through bonds for capex purposes, making this the last run of 2026.

That’s a lot less aggressive than where expectations were at the beginning of the year.

Even after all the touting from OpenAI and Anthropic, only 18% of employees in this survey report that AI has been implemented into their workflows.

That doesn’t sound like a lot has been done in terms of brining AI into the corporate ecosystem.

And that brings me to another point Covello made on this exact topic:

I think you and I both experience this.

How can we possibly know if the AI output is correct or hallucinating if we ourselves have no knowledge or experience in the topic?

That’s one issue.

The other comes from security and proprietary information concerns, which is why Microsoft has now replaced OpenAI and Anthropic with their own AI labs.

Meta just announced a similar shift this morning as well.

End result?

Little to no adoption currently, and the ones who did attempt to adopt these models are quickly dropping them due to costs and no productivity realization.

That leads me to the next point.

All previous technology has succeeded because it disrupted a previous leader.

AI has not yet replaced any single piece of technology or service.

To be honest, the only one that’s really at risk is replacing search engines, but that’s about it.

In fact, a Citadel research paper shows you less than half of all industry functions are under replacement threat.

Instead, all job functions are benefitting from productivity enhancements from AI, and that’s a completely different story.

Replacing job functions was the biggest selling point in the AI trade and capex justification.

Now companies are realizing this promise was dumbfounded, and at best AI will be like any other piece of software made to increase your productivity.

Which makes me question whether these private AI companies should command the valuations they currently have.

Especially because:

We all know about this circus.

NVIDIA finances its customers so they can afford to buy their chips, book revenues and income, and do the same for about 100 other entities.

The problem here is, no actual cash profit is being made.

Which is why income statements and balance sheets for names like:

Are increasingly being made up of accruals (non-cash items.)

That’s a very dangerous place to be in, because eventually these earnings are adjusted according to GAAP laws.

When that happens, EPS could be cut in half or less, bringing valuation multiples to twice the levels reported today.

This is why Covello says public markets are not in a price bubble, but rather in an earnings bubble.

I agree, and broke it all down for you in this AI deep dive report.

In the private markets, Covello and I can agree there is a pricing bubble now that Anthropic and OpenAI have risen to over $1.5 trillion in valuation each.

Who breaks first is up to debate.

What matters is there isn’t a single bull who is willing or able to tackle these issues head on with more than just an “AI demand” fairytale to defend themselves.

If that’s not a bubble, I don’t know what is.

We all know that the “AI will kill software” narrative is very much alive at this point.

There are those who understand that couldn’t be further from the truth, and they’ve been betting big on it.

Over 1,000 call options have been bought in the IGV Software ETF for $95 strikes this July.

The $110 calls also carry the most open interest.

I believe this bet on a software rally could be due to the recent SK Hynix listing, or perhaps another risk-off rotation back into real economy names.

Either way,

The Offside Portfolio is smartly overweight software in some of the names showing zero disruption at this moment.

All with over 100% upside embedded into them.

Click here to receive live portfolio updates.

For my risk-takers, I’d closely watch this IGV bet for $95 - $110 levels at the end of the month.

Now $92.50.

In the meantime, here’s Howard Marks (one of the market wizards) talking about just how extended this current market and its expectations have become.

Enjoy it over the weekend:

Until next time,

OFFSIDE RESEARCH

Against the Tape, Ahead of the Curve.