Watch out Nvidia. Meta just signed a multi-billion-dollar deal for Google AI chips

What if the biggest risk to Nvidia (NVDA) stock isn’t a new chip.

But a clever negotiating tactic that could completely reshape AI infrastructure?

Meta Platforms (META) just signed a multi billion dollar, multi year deal to rent AI chips from Alphabet’s Google (GOOGL).

That move puts Alphabet, Nvidia, and Advanced Micro Devices inside the same Meta supply chain, just as Google makes moves in the AI race with Gemini and TPUs

For the last two years, if you wanted to train cutting edge AI models at scale, you bought Nvidia. Period. Customers paid up, margins surged above 70 percent, and Nvidia effectively set the terms of the AI boom.

Now Meta is signaling it doesn’t want to be locked into one supplier, even as it expands its Nvidia partnership with millions of AI chips.

Meta will rent Google’s Tensor Processing Units, known as TPUs, to train new AI models. That announcement came just days after AMD said it could sell up to $60B worth of AI chips to Meta, sending shockwaves through the market as AMD stock jumped after the major Meta GPU deal. Nvidia is still in the mix with ongoing orders.

In other words, Meta is spreading billions in AI spending across three chip providers instead of one.

Meta needs more compute. Google has capacity. AMD wants market share. Nvidia wants to defend dominance.

Meta is building leverage.

It doesn’t need to shift most of its budget to change the conversation. It just needs to make the threat credible.

The disruption behind the news: This is the beginning of the end for Nvidia’s pricing power.

Meta is creating competition at the infrastructure layer.

Google is turning internal hardware into an external revenue driver.

For the last two years, Nvidia has effectively set the rules of AI computing.

If you wanted cutting edge models, you bought Nvidia. If you wanted scale, you paid whatever the market would bear. Gross margins above 70 percent usually do not happen in a healthy competitive market. They happen when customers have few real alternatives.

Meta just signaled it will not stay locked in.

Meta does not have to move most of its spending to change the market. It just has to make the threat believable. If Meta is on track to direct about $10B a year of additional AI compute spending, then even a 5 percent pricing concession from Nvidia, because Meta can point to AMD or TPU benchmarks, would equal about $500M a year in savings. That is the leverage math. A relatively small shift in orders can force broader repricing across the rest of Nvidia’s business with that customer.

By splitting demand across Nvidia, AMD, and now Google TPUs, Meta creates a live benchmark inside its own operations. Performance per watt. Training speed. Cost per token. Every supplier now competes for a larger slice of the same budget. That is how margins get compressed. That is how a choke point turns into a bidding contest.

Google’s position is even more interesting. TPUs were originally built to power its own services. Now they are revenue generating infrastructure. If Meta trains on Google silicon, even partially, Google benefits from the AI arms race regardless of which company wins at the model layer. It also strengthens Google Cloud, where TPU sales are becoming a growth driver investors can measure and model.

There is a second order effect. When one hyperscaler validates alternative chips at scale, others tend to follow. If Meta proves that TPUs can handle large scale training reliably, companies like Amazon and Microsoft will have to explain why they remain heavily exposed to Nvidia. The switching cost argument starts to weaken.

What to watch next

Watch Nvidia’s pricing over the next 12 months to 24 months.

Watch how much of Meta’s training workload shifts to non Nvidia chips.

Watch whether Google opens more TPU capacity to outside customers.

If AMD actually delivers anywhere close to $60 billion in chips to Meta, that alone would diversify a large portion of AI demand. Add a multi billion dollar TPU rental deal and you begin to see a three player market instead of something close to a monopoly. That changes capital allocation decisions across the sector, from data centers to networking to power infrastructure. And Meta’s aggressive AI spending doesn’t exist in a vacuum. Even as it invests heavily in infrastructure, it continues expanding hardware ambitions, including reports that Meta has revived its smartwatch project targeting a 2026 debut, a reminder that AI compute underpins far more than just chatbots.

Also watch deployment speed. Renting TPUs through Google Cloud is faster than building new data centers from scratch. If Meta can spin up training clusters without waiting years for construction, it shortens model development cycles. Faster iteration means faster product launches. That directly affects competitors.

And watch compute costs. They shape everything from what you charge for your API to how fast you burn through cash. If the big cloud players can cut training and inference costs by even 10 percent to 20 percent, a lot of products that didn’t quite make financial sense suddenly will.

We’re moving past the era where AI infrastructure comes from just one supplier. Nvidia. It's not a nail-in-the-coffin bad. Nvidia will still thrive as it still makes the best AI chips for training. But don't think that the kind of dominance it enjoyed in the early days of the AI boom will last.

Meta (META), Nvidia (NVDA) and Google (GOOGL) all have a Disruption Score of 4. Click here to learn how we calculate the Disruption Score. 

All three are also part of the Disruption Aristocrats, our quarterly list of the world’s top disruptive stocks.