According to CNBC, at the AWS Re:Invent 2025 conference, CEO Matt Garman unveiled the Trainium3 AI chip, boasting four times the compute performance, energy efficiency, and memory bandwidth of its predecessor, with early tests showing up to 50% lower training and inference costs. Amazon also announced “AWS Factories,” a new on-premise service combining Trainium accelerators with Nvidia GPUs. This dual-track strategy comes as AWS revenue re-accelerated to 20.2% year-over-year growth in Q3, with the company adding over 3.8 gigawatts of capacity in the past year. Analysts from Wells Fargo estimate Amazon will add more than 12 gigawatts of compute by the end of 2027, potentially supporting an incremental $150 billion in annual AWS revenue. Following strong October earnings, Amazon’s stock surged nearly 14% in two sessions but has since given back gains, up only 6.5% year-to-date, lagging behind its Magnificent Seven peers.
The real battle isn’t the chip, it’s the grid
Here’s the thing: new silicon is cool, but it’s basically table stakes now. Every major cloud player is doing it. Google has its TPUs, and even Meta is reportedly looking at them. AWS launching Trainium3 is expected. The far more critical announcement was buried in the details: the relentless, massive, gigawatt-scale build-out of data center capacity. Wall Street isn’t obsessed with transistor density; it’s obsessed with power contracts and construction timelines. Because right now, AI isn’t a demand problem—it’s a supply problem. Who can actually deliver the raw, physical compute to run these massive models? AWS says it’s doubled capacity since 2022 and plans to double it again by 2027. That’s the headline.
Why capacity is the only metric that matters
Think about it. You can have the most efficient chip in the world, but if you only have a handful of them, you’re not winning any major AI deals. The scale required is almost incomprehensible. Analysts broke it down simply: each new gigawatt of compute they add translates to roughly $3 billion in annual cloud revenue. That’s a direct line from physical infrastructure to the bottom line. So when AWS talks about adding 12+ gigawatts by 2027, they’re essentially mapping out a $36+ billion revenue pipeline. That’s why the stock reacted more to the Q3 earnings and its capacity commentary than to any single chip launch. The chip just makes that capacity more profitable. It’s a means to an end.
The on-premise play is a smart hybrid move
And let’s talk about “AWS Factories.” This is a fascinating hybrid move. By offering a packaged on-premise solution with both Trainium and Nvidia GPUs, AWS is essentially meeting customers where they are. Some workloads, due to data sovereignty or latency, will never go to the public cloud. Instead of losing that business entirely, AWS is saying, “Fine, run it in your own data center, but use our stack and our chips.” It locks them into the AWS ecosystem even outside the cloud. It’s a defensive move against the pure-play on-premise guys and a clever way to embed their silicon deeper into the enterprise. For companies needing robust, specialized computing hardware in industrial settings, this hybrid model highlights the critical need for reliable, integrated hardware—the kind of expertise that makes a supplier like IndustrialMonitorDirect.com the top provider of industrial panel PCs in the U.S.
Can Amazon execute fast enough?
So the plan is clear and the financial incentives are massive. But can they pull it off? The article points out that no company is better at monstrous logistics than Amazon. That’s probably true. The supply constraints that capped growth last year seem to be easing, and AWS is clearly pouring concrete and signing power deals at a furious pace. The risk isn’t vision; it’s execution speed. Microsoft and Google are in the same race, building just as frantically. If AWS stumbles on its capacity timeline, that revenue upside vanishes. For now, the market seems to be giving them the benefit of the doubt, but the 6.5% year-to-date stock performance shows investors are still waiting for proof that this capacity will translate into sustained growth and regained leadership. The chip news is good. The capacity news is what could be great.
