Amazon is about to spend up to $60 billion on Qualcomm’s AI data-centre chips and related products. That might sound normal for a company trying to expand its AI infrastructure, except Amazon already designs and builds its own chips. It’s a bit like someone who bakes their own bread announcing they’ve also started a standing order with the local bakery.
The deal that was confirmed by Qualcomm this week will focus on AI inference and cover multiple generations of chips. So if Amazon is already investing in its own chips, why is it also working with another chipmaker?
Amazon Wants More Control Over AI Hardware
Amazon’s decision to develop its own chips gives it more say over how the hardware used by AWS is designed. Instead of relying entirely on general-purpose chips, Amazon can build silicon around the workloads its cloud customers are actually running which can help the company manage costs, improve performance and make its data centres more energy efficient.
On the AI side, Amazon’s chips designed for AI workloads are called Trainium, while Graviton is designed for general cloud computing. Trainium is built for AI model training and inference, while Amazon also has Inferencia chips that are aimed specifically at inference. Amazon says its custom silicon is designed to deliver better price performance and energy efficiency for the workloads running on AWS. That gives Amazon more flexibility over how its infrastructure is built and which hardware it uses for different types of work.
More from Artificial Intelligence
- Elon Musk Wants To “Solve” Chess, But Isn’t The Struggle The Point Of The Game?
- Anthropic’s Own Alignment Lead Says There Is A 10% Chance AI Kills Everyone Within A Decade
- The Company Teaching Software To Speak: ElevenLabs
- Can AI Stop Ageing? Inside The Clinical Breakthroughs Delivering Concrete Results
- AI Deepfakes Are Forcing Schools To Rethink Child Protection
- Experts Are Sounding The Alarm On AI Animal Translation – Breakthrough Tool Or New Form Of Exploitation?
- The Rise of “Agentic Compliance”: Shifting from Policies To Real-Time Guardrails
- Inside Nairobi’s AI Shock: What Happens After AI Wipes Out A 40,000-Worker Economy?
Why Is Amazon Working With Qualcomm?
The Qualcomm deal is focused on AI inference which is when a trained AI model is used to produce an answer or prediction. That’s different from the training stage, where the model is built and taught using huge amounts of data.
Reuters says Amazon could buy up to $60 billion worth of Qualcomm’s AI data-centre chips and related products. The two companies are also working on optical connectivity for AI data centres, which is designed to help move huge amounts of data around the infrastructure.
Inference is also becoming a bigger part of the AI infrastructure as more AI tools move from training into everyday use. Every chatbot response, generated image or AI-powered search result needs computing power behind it. As more of these services are used every day, the infrastructure needed to handle those requests becomes a really big part of running AI at scale.
Amazon Doesn’t Need To Build Everything Itself
Building its own chips gives Amazon something it wouldn’t get from buying whatever is available. It can decide what those chips need to do and how they fit into the rest of AWS, which can make a difference when you’re running huge data centres and dealing with the cost of powering them.
There’s also a financial side to this, as designing and producing custom silicon needs significant investment – so there’s no point in Amazon trying to replace every other hardware piece on its own.
But designing a chip is a really big job, and there isn’t much reason for Amazon to recreate every piece of technology it needs. Qualcomm already has experience developing hardware for demanding computing workloads, so Amazon can use its own chips in areas where they give AWS an advantage and bring in outside technology where that makes more sense.
Amazon Wants Control, But Not Total Control
Amazon having its own hardware was never really about making every piece of hardware inside AWS itself – it was about having more say over how that infrastructure works, from the performance of its AI systems to the cost and energy needed to run them.
The partnership with Qualcomm means that there’s now another option instead of replacing what Amazon has already built. Its own chips give it one option, while Qualcomm can now bring another type of hardware into the mix.
Therefore, Amazon can develop its own technology while still working with companies like Qualcomm where their hardware makes sense. How those different chips are used across AWS will depend on the workloads they are built to handle.
