On September 8, 2026, Qualcomm announced a multi-generation AI data-center partnership with Amazon/AWS. This is considerably more important than a conventional chip-supply agreement because it gives Qualcomm a credible path into the hyperscale AI market alongside its existing Meta relationship. Qualcomm
What Amazon and Qualcomm are doing
There are really three pieces to the agreement.
1. Custom AI silicon for AWS. Qualcomm and Amazon will jointly develop customized chips for AWS's large-scale AI infrastructure, with a particular emphasis on AI inference—running trained AI models rather than training them. Qualcomm's pitch is power efficiency: generating more AI output per watt, which becomes increasingly important as hyperscalers run into power and cooling constraints. Qualcomm
2. Qualcomm gets into AWS's AI networking infrastructure. This may be almost as interesting as the processors. Qualcomm will provide technology for high-speed optical connectivity, including links extending to 1.6 Tbps and future generations, using its SerDes and optical-DSP technology. In other words, Qualcomm isn't merely trying to sell Amazon an AI processor—it wants silicon inside the networking fabric connecting AI servers and racks. Qualcomm Investor Relations
3. Amazon received warrants tied to potentially enormous purchases. Qualcomm issued Amazon warrants for 25 million QCOM shares at $161.26 per share, representing roughly $4 billion if exercised. Importantly, those warrants vest in stages tied to commercial arrangements and Amazon purchases that could reach $60 billion of Qualcomm server chips and related technology. Reuters
That last point requires some care: this does not mean Qualcomm has booked a guaranteed $60 billion order. The structure gives Amazon incentives as purchases scale toward that amount. Qualcomm's own announcement does not disclose firm unit volumes or a detailed deployment schedule. Qualcomm
Why I think this matters so much for Qualcomm
Smartphones to Edge AI → Physical AI → Data-center AI infrastructure.
And Amazon is now a powerful validation point.
Qualcomm unveiled its Dragonfly data-center portfolio this year, covering CPUs, AI accelerators, custom silicon and connectivity. Meta already agreed to use Qualcomm's Dragonfly C1000 CPUs in next-generation servers. Amazon now adds a second hyperscaler relationship, centered on custom inference silicon and connectivity. Data Center Dynamics
The emerging picture looks something like:
Amazon AWS
↓
Qualcomm custom AI silicon
↓
AI inference
↓
Qualcomm SerDes / optical DSP
↓
1.6T+ data-center networking
That is a dramatically larger addressable market than Qualcomm's traditional Snapdragon handset franchise.
The $60-billion number is especially interesting
For perspective, Reuters reports that Qualcomm is targeting $15 billion of annual data-center chip revenue by 2029. The Amazon arrangement gives some credibility to what previously looked like a very ambitious target. Reuters
The warrant structure also creates unusual alignment. If Qualcomm's technology works well and Amazon dramatically increases purchases, Amazon can become a meaningful Qualcomm shareholder.
There is a downside: if all 25 million warrants ultimately vest and are exercised, existing QCOM shareholders face some dilution. But that dilution would occur in connection with very substantial commercial purchases, so investors need to evaluate the dilution against the revenue generated.
Why Amazon wants Qualcomm
AWS already develops its own silicon—Graviton CPUs, Trainium AI accelerators and Inferentia inference chips. So Amazon isn't abandoning custom silicon and simply buying Qualcomm processors.
Rather, Qualcomm brings intellectual property and engineering expertise in areas Amazon increasingly needs: low-power compute (Robotics), custom silicon, high-speed SerDes, optical DSPs, system integration and inference optimization. Data Center Dynamics
This fits an important change occurring in AI. Training giant models made NVIDIA GPUs indispensable. But as hundreds of millions of users and AI agents continuously run those models, inference becomes an enormous workload of its own.
Qualcomm is effectively betting:
AI inference will eventually dwarf AI training in volume, and electricity efficiency will become one of the industry's most important competitive advantages.
That's precisely where Qualcomm's decades of designing high-performance processors under severe power constraints become relevant.
What it means for the QCOM investment thesis
For someone evaluating Qualcomm primarily as a smartphone-chip company, this agreement changes the picture materially.
The company now has credible exposure to hyperscale AI compute + custom silicon + AI networking + edge AI + automotive + robotics/physical AI.
The principal risk is execution. Amazon has not guaranteed $60 billion of purchases, NVIDIA remains extraordinarily entrenched, and Qualcomm must prove that Dragonfly/custom silicon can move from design wins into large-scale production revenue.
But Amazon joining Meta as a major hyperscaler relationship makes Qualcomm's data-center ambitions substantially harder to dismiss as merely a roadmap.
For our QCOM thesis specifically, I consider the Amazon agreement one of the most important developments we've seen this year. It strengthens the part of the thesis that had previously been the least proven: Qualcomm becoming a serious AI data-center infrastructure supplier, while the edge/robotics opportunity remains intact.
The Edge Edge!
Qualcomm has an “edge at the edge” because much of the next wave of AI will move out of giant data centres and into physical devices—robots, humanoids, vehicles, drones, industrial machines, PCs and smartphones—where AI must operate locally, instantly and with limited battery power.
This is precisely where Qualcomm has spent decades building expertise: highly power-efficient processors that combine CPU, GPU, NPU/AI acceleration, connectivity, cameras, sensors and real-time processing in compact systems.
While NVIDIA dominates centralized AI training, Qualcomm's particular opportunity is AI inference at the physical edge, where sending every decision back to the cloud is too slow, expensive or unreliable.
If Physical AI develops as expected, Qualcomm's competitive advantage could therefore be its ability to put a powerful, energy-efficient “AI brain” directly inside millions of intelligent machines—an “edge at the edge.”
Robotics:
Qualcomm now has several concrete relationships with robotics manufacturers, including humanoid companies. The most significant are Figure and NEURA Robotics, and Qualcomm says its processors are already powering robots from other OEMs including Booster and VinMotion. Qualcomm
The relationships break down like this:
| Robotics company | Qualcomm relationship | Significance |
|---|---|---|
| Figure AI | Collaborating to define the next generation of compute architecture for Figure's humanoids | Very important |
| NEURA Robotics | Long-term strategic collaboration; Dragonwing processors + NEURA humanoids/robots | Very important |
| Booster Robotics | Dragonwing processors already powering humanoid platforms | Commercial ecosystem |
| VinMotion | Dragonwing processors powering humanoid platforms | Commercial ecosystem |
| KUKA Robotics | Discussions/collaboration around next-generation robotics solutions | Industrial robotics opportunity |
| Advantech / AutoCore | AMR and industrial robotics development | Industrial Physical AI |
| PickNik Robotics | Qualcomm agreed this month to acquire it; owns/stewards MoveIt robotics software | Strategically important |
Qualcomm describes the Dragonwing IQ10 as essentially the "brain of the robot"—a processor designed specifically for advanced AMRs and full-sized humanoids. It integrates CPU, NPU/AI acceleration, vision, sensor processing, connectivity and real-time control. Qualcomm
1. Figure is the one I watch most closely
Figure AI and Qualcomm announced in January that they are collaborating on the next generation of compute architecture as Figure scales its humanoid platforms.
Figure CEO Brett Adcock specifically highlighted Qualcomm's combination of compute performance and energy efficiency as a building block for scaling Figure's humanoids. Qualcomm
This is particularly interesting because Figure is trying to manufacture humanoids at very large scale.
The relationship suggests Qualcomm wants to occupy a position analogous to:
NVIDIA → AI servers
Qualcomm → autonomous robots / Physical AI
It is far too early to say Qualcomm will achieve that position, but that is clearly the strategic direction.
2. NEURA Robotics is an even broader agreement
NEURA Robotics and Qualcomm announced a long-term strategic collaboration in March 2026.
This goes considerably beyond supplying a processor.
They're developing what the companies call a "Brain + Nervous System" reference architecture for robots. Qualcomm supplies Dragonwing compute, AI acceleration, connectivity and software; NEURA contributes the robot hardware and embodied-AI stack. The architecture is intended for robotic arms, AMRs, service robots and humanoids. Qualcomm
That is exactly the type of relationship I'd want to see if the thesis is that Qualcomm becomes a major picks-and-shovels supplier to Physical AI.
3. Qualcomm silicon is already inside humanoid platforms
Qualcomm says its Dragonwing industrial processors already power humanoid robots from Booster Robotics and VinMotion, among other robotics manufacturers. Its newer IQ10 reference-design program also includes NEURA, Booster and VinMotion as early-access ecosystem partners. Qualcomm
So this isn't entirely a 2028–2030 story. There is hardware being developed and deployed now.
4. And Qualcomm just made another very interesting move
Only five days ago—September 23—Qualcomm announced an agreement to acquire PickNik.
PickNik is the longtime steward of MoveIt, one of the most widely used open-source software frameworks for robotic motion planning and manipulation.
Qualcomm intends to integrate MoveIt much more tightly with Dragonwing. That potentially gives Qualcomm both:
Robot hardware compute: Dragonwing
+
Robot software/motion layer: MoveIt
+
AI inference: Qualcomm NPU
+
Connectivity: Wi-Fi/5G/6G
+
Real-time control: Dragonwing robotics architecture**
That's strategically much more interesting than merely selling chips. Qualcomm
But there are some conspicuous absences
I don't find evidence of announced Qualcomm compute agreements with Tesla Optimus, Boston Dynamics or Apptronik.
For example, Apptronik's disclosed AI relationship is with Google DeepMind; its latest Apollo 2 announcement describes fleets collecting real-world data to develop humanoid AI models in partnership with DeepMind. Apptronik
And I would not assume Tesla is a future Qualcomm customer. Tesla has strong incentives to develop its own Optimus compute architecture.
Why this strengthens the QCOM thesis we've been discussing
This is where our interest in Qualcomm as a Physical AI investment becomes increasingly relevant. Qualcomm is building an actual robotics ecosystem rather than simply claiming that Snapdragon-type processors could someday be used in robots. Qualcomm
And the timing is interesting because Qualcomm's Dragonwing IQ10 can deliver up to 700 TOPS, while supporting as many as 20 cameras plus lidar/radar and real-time control—all designed around relatively power-efficient edge computation. Qualcomm
That potentially puts Qualcomm into three enormous AI markets simultaneously:
Amazon/Meta → Data-center AI
PCs/phones/automotive → Edge AI
Figure/NEURA/Booster/VinMotion → Physical AI & humanoids
That third category was considerably more speculative when we first discussed the "sleeping dragon" QCOM thesis. As of September 2026, it is becoming much more tangible.
