Meta signed a multi-year agreement with AMD to build AI computing capacity that can reach roughly 6 gigawatts over time. For AMD, the value is not only the headline size. It is the signal that its accelerators are moving into large-scale use at one of the world’s biggest infrastructure buyers. That is why the market reaction was immediate and positive.
The unusual part is how the deal is designed. AMD is offering Meta an equity-based incentive tied to performance, up to 160 million shares, released in steps as AMD meets delivery milestones. The first portion is linked to delivering the first 1 gigawatt of capacity. For investors, this cuts both ways: it pushes AMD to deliver on time, but it also introduces possible dilution and shows how intense the competition is when hyperscalers negotiate for AI supply.
What does "6 gigawatts" mean and why it's big news
A gigawatt is a unit of power. In this debate, what it means in practice is that $META is not buying "a few thousand units" but planning a giant scaling of computing capacity across data centers. By comparison, even a shift of units of hundreds of megawatts in AI clusters is already considered a significant step. Six gigawatts is therefore a signal to the market that Meta $META is building infrastructure in "industrial production" mode while diversifying suppliers alongside Nvidia $NVDA.
Moreover, both AMD and Meta are talking about multiple generations of chips. The first wave is to consist of MI450 series accelerators to be deployed in the second half of the year in so-called "rack" systems (simplified whole computing enclosures assembled as a single unit). This is important because AI is no longer just about the chip itself, but about how well the cooling, interconnects and software - the whole "package" - are handled.
Why Meta is also paying with AMD stock
The stock instrument (warrant) is set up so that Meta only gets shares if AMD actually delivers what it promises. In practice, this is a "pay for performance": AMD gets a huge subscription and Meta insures that deliveries will be top priority. At the same time, this puts a financial "tether" on Meta to make AMD succeed - if AMD meets the milestones and the stock goes up, Meta can profit.
But for AMD shareholders, there's also the issue of risk. If the entire facility is released, it could mean dilution (more shares outstanding). And it also shows how extremely valuable certainty of supply is today: when companies are building tens of billions of dollars worth of data centers, "not having the chips on time" is more expensive for them than paying substandard terms.
AMD gains validation, Meta breaks down dependence on Nvidia
Just a week ago, Meta announced a multi-year deal with Nvidia to supply millions of Blackwell and Ruby generation GPUs, as well as a large deployment of servers with Grace processors. In sum, this says one thing: Meta doesn't want to stand on one card. Nvidia remains key, but AMD is opening the door to a segment that has long been seen as "Nvidia territory".
In addition to GPUs, processors are also mentioned in the text. Meta mentions CPU purchases from AMD (including new generations) because AI is no longer just about training models, but increasingly about running them on a day-to-day basis. That's the part where CPUs and the whole infrastructure around them are rising in importance alongside accelerators - and AMD has traditionally been strong in CPUs.
Context: Meta raises AI investment to extreme levels
The whole thing makes sense the moment one puts Meta's budget next to it. The company said in earnings that it expects capital expenditures of roughly $115 billion to $135 billion for 2026, primarily due to AI infrastructure. That's a pace that shifts the entire sector - because when one of the biggest players starts spending like that, it forces others to keep up or risk losing ground.
And here's why some investors have become nervous: spending is growing faster than the certainty of return. Meta has held sentiment the best of the big "AI spenders" so far, but the question remains the same for everyone: when will this translate into higher profits, not just bigger data centers.
What to watch next
There are three things that will be crucial with this deal. First, how quickly AMD can realistically deliver the first "1 gigawatt" and unlock the first part of the reward. Second, if the MI450 proves itself in practice to keep the Meta picking up power draws towards 6 gigawatts. And third, how the supplier mix at Meta will evolve alongside Nvidia and its own chips - because diversification is a deal breaker for Meta, but for suppliers it's a competitive pressure on price and performance.