What the reaction to Google’s potential entry into selling chips tells us about the AI trade

A report from The Information suggesting that Google is in talks to begin selling its custom TPU chips to Meta has spurred huge market moves, and a ton of commentary about what this means for the major players in the AI trade.

Here’s my contribution:

Huge Margins Are a “Kick Me” Sign

Posting persistently massive profit margins is akin to holding up a big sign to the world saying “you should be in this business, too!”

New entrants into a market generally tend to reduce pricing power for dominant incumbents.

That’s easier said than done when it comes to developing high-powered chips to bolster the growth of a technology that was more science fiction than consumer-facing as recently as three years ago.

The AI boom is marked by both its scale and urgency. Hyperscalers will give you a litany of reasons to invest, from the pursuit of artificial general intelligence  (though you don’t hear as much about that these days!) to the fear of having their existing leading market positions diminished by competitors that are willing to make these large outlays. 

Urgency means you’re willing to pay up for inputs, so if there were a time to be selling AI chips, now would be a good time. That’s the signal from Nvidia, whose adjusted gross margin has mostly been around the mid-70s over the past two years, with the exception of its fiscal Q1 2026.

Perhaps counterintuitively, given the positive market reaction to the reported talks with Meta, for Google, selling TPUs is margin negative compared to renting out access to the computing power provided by those same chips.

But a deal to sell TPUs to Meta would have benefits that could potentially outweigh the drag on margins. It would let the company gain a foothold among customers who would otherwise not use these TPUs and instead run AI tasks on a competitor’s cloud. Striking while the iron is hot would also likely help Google ensure that its software increases in prominence among the developer community — a key contributor to Nvidia’s moat.

Big Endorsements Matter

If these reports bear fruit, what will have been important is not just the fact that Google is selling their TPUs, but who they are selling them to. That Meta is in talks to buy TPUs provides the second strong validation point for their quality in just the last week: Gemini 3 was already a testament to their capabilities, and now they are reportedly closing in on an additional stamp of approval from Mark Zuckerberg.

This is a pattern we’ve seen this before: after AMD went parabolic on the heels of its megadeal with OpenAI, Wedbush analyst Dan Ives said that this was a “huge vote of confidence” and that “any lingering fears around AMD should now be thrown out the window.”

...But Customer Quality Matters, Too

But alas, what appears to be getting thrown out the window now is Advanced Micro Devices. Per the market’s knee-jerk reaction, it is the single biggest loser of Google’s potential foray into selling AI chips.

Loosely, Nvidia is still presumed to get the lion’s share of the pie, but this raises the risk in traders’ eyes that AMD’s slice looks more like scraps.

A vote of confidence from OpenAI simply does not carry the same weight as validation from a trillion-dollar hyperscaler. At this point, if OpenAI hasn’t made an multi-billion dollar commitment to you, are you really an AI company?

Remember, Oracle peaked and rolled over once it was reported that their massive sales backlog was largely being fueled by OpenAI. Their stock price has gone one way since then, soon followed by its credit default swap spreads heading in the opposite direction.

Mini-DeepSeek

In some ways, this negative shock for some parts of the AI trade is an echo of the DeepSeek-induced freakout.

Back in January, the emergence of this Chinese AI model raised the idea that you can do AI on the cheap, casting doubts over the wisdom of hundreds of billions in capex (and counting). Jevons Paradox, in this case, the idea that AI becoming cheaper would ultimately increase overall demand for compute, won the day.

Right now, a potential Google entry is being treated as a zero to negative sum event for AI chip designers.

Unless the cost-savings of Google’s TPUs relative to any performance sacrifices versus GPUs are a game-changer for the economics surrounding AI training, inference, and beyond, this probably isn’t what matters for investors in any of this, or even just people who hold index funds.

As we all soon gather to eat copious amounts of pie, I’ll remind you that the size of the pie is what really matters. And that will be driven by whether AI is or becomes sufficiently cheap to deploy at scale that it generates a sufficient return on investment for the companies making these major outlays, and whether their customers also see enough of a benefit from making use of this computing power.

That’s still a largely unanswered question. Time and again throughout this boom, we’ve seen different regimes dominate: the promise of tomorrow versus the realities of the quarterly corporate reporting cycle today.

Think Beyond Chips

What’s interesting to me is that while AI chips are clearly high in demand, they don’t appear to be the most binding constraint on the boom. Microsoft CEO Satya Nadella recently said his biggest problem today is “not a supply issue of chips; it’s actually the fact that I don’t have warm shelves to plug into.” Nvidia CEO Jensen Huang warned that “China is going to win the AI race” in part because of better access to power. And CoreWeave CEO Michael Intrator told analysts that “across the space,” the issue is a shortage of other physical infrastructure to support data center build outs.

And in a supply-constrained AI world, it’s also fascinating that Google must feel it has the ability to get its hands on enough chips to satisfy not only its own computing needs, but for third parties, as well.