Qualcomm and Amazon's $60 Billion AI Chip Deal, Explained
On September 8, 2026, Qualcomm and Amazon Web Services announced a multi-generation deal to co-develop custom AI inference chips and fast optical networking for AWS data centres. The headline number, up to $60 billion of purchases through 2036, is a ceiling tied to stock warrants, not a signed order. Updated October 30, 2026: here is what the deal actually commits each side to, and why it matters for the cost of running AI.
⚡ Quick facts
- Ceiling: Amazon must spend up to $60 billion with Qualcomm through 2036 to unlock all of its warrants, per Qualcomm's 8-K as reported
- Warrants: up to 25 million Qualcomm shares at $161.26 each (about $4 billion); about 3.75 million vested immediately
- What's built: custom AI inference accelerators plus 1.6 Tb/s optical interconnects using Qualcomm SerDes and DSP technology
- Two-way: Qualcomm will run more of its chip-design (EDA) work on AWS, including Amazon Bedrock
- Market reaction: Qualcomm shares rose roughly 4–5% on the announcement
What the deal includes
Per HotHardware, which cites Qualcomm's SEC filing, the two companies will co-develop custom neural processing accelerators for AI inference (running trained models, not training them) and 1.6-terabit optical interconnects to link AWS server racks. Qualcomm contributes high-speed SerDes and digital signal processing IP. In return, Qualcomm will move more of its own electronic design automation onto AWS. Neither company has named specific chip models or delivery dates.
The $60 billion figure, unpacked
Qualcomm gave Amazon a warrant to buy up to 25 million shares at $161.26 each, exercisable over about a decade. The shares vest as Amazon spends: about 3.75 million vested at signing based on initial purchase commitments, and the full amount needs up to $60 billion of purchases through 2036. As The Motley Fool notes, that is a procurement ceiling, not committed order volume. Warrant structures like this have become common in AI infrastructure deals because they reward the buyer for actually following through.
Why each side wants it
For Qualcomm, it is the clearest proof yet that its push beyond smartphones into data centres is winning real customers. Coverage, including Digital Watch and Android Headlines, frames it as one of several recent hyperscaler wins, following a custom-chip deal with ByteDance. For AWS, which already designs its own Trainium and Graviton chips, a co-designed inference part adds another supplier and more leverage in pricing talks with NVIDIA, the dominant AI chip vendor.
Our take: how to read big AI chip deals
- Separate ceilings from orders. "Up to $60 billion" is the maximum that unlocks every warrant. Watch Qualcomm's quarterly results for actual data-centre revenue.
- Inference is the battleground. Training needs few, huge clusters; inference runs every chatbot reply, so cheaper inference chips cut the cost of AI products at scale.
- Networking matters as much as chips. The 1.6 Tb/s optical work targets the links between racks, which often limit large AI clusters.
- Timelines are long. Multi-generation means several chip cycles. Do not expect these parts in AWS instances right away.
- Competition should lower prices. More credible suppliers beside NVIDIA usually push compute costs down over time.
For another hardware company moving up the AI stack, see our explainer on NVIDIA's Hugging Face acquisition.
Sources
- HotHardware: Qualcomm strikes major AWS deal for custom AI chips
- The Motley Fool: Qualcomm's massive AI deal with Amazon
- Digital Watch: Qualcomm announces $60 billion AI chip deal with Amazon
- Android Headlines: Qualcomm and Amazon partner on custom AI chips
Frequently asked questions
How much is the Qualcomm–Amazon AI chip deal worth?
Up to $60 billion of purchases through 2036. That is the ceiling needed to unlock all of Amazon's warrants, not a signed order of that size.
What are the Amazon warrants?
Rights to buy up to 25 million Qualcomm shares at $161.26 each, worth about $4 billion. They vest as Amazon spends; about 3.75 million vested at signing.
What will Qualcomm build for AWS?
Custom AI inference accelerators and 1.6 Tb/s optical interconnects for AWS data centres, using Qualcomm's SerDes and DSP technology.
Is this for training or inference?
Inference: running trained AI models to answer requests. That is where most day-to-day AI computing happens.
Does Amazon still make its own chips?
Yes. AWS keeps its in-house Trainium and Graviton chips. Qualcomm adds another option and more pricing leverage against NVIDIA.
When will these chips be used?
Neither company has given dates or model names. It is a multi-generation deal, so deployments will roll out over several years.