The device that sits on millions of kitchen counters is about to get a new brain. Amazon’s top hardware executive said this week that the company is focused on building its own chips for its most important consumer devices, including the Echo Show and Fire TV, as part of a push to run artificial intelligence directly on the hardware instead of in the cloud. The remarks, reported by financial news service Cailianshe, mark the clearest statement yet of Amazon’s intent to control the silicon inside its own products.
Amazon has been moving in this direction for years, but the pace has accelerated. Last October the company introduced two chips of its own design, the AZ3 and AZ3 Pro, built specifically to run AI models locally on devices. The chips are small by data-center standards, but they represent a significant bet: instead of sending every voice command to Amazon’s cloud servers, the Echo of the future will interpret much of what it hears on the countertop, in the car, or on the living-room shelf.
The logic of on-device AI is partly technical and partly economic. Running models locally cuts latency — the answer comes back in milliseconds rather than after a round trip to a data center. It cuts cost, because every query that stays on the device costs Amazon nothing in compute. And it answers a question that has dogged voice assistants since the first Echo: what happens when the network is slow, or the cloud is down, or the user simply does not want the conversation leaving the room? Local chips address all three.
Amazon’s hardware chief said the company is preparing to swap the processors in its key devices as the technology matures, a process he described as an upgrade cycle for the entire product line. Fire TV, which has sold in the tens of millions, is an obvious candidate: streaming boxes run AI for picture enhancement, voice search, and recommendations, and a custom chip can do all of it more cheaply than a general-purpose processor from Qualcomm or MediaTek. The Echo Show, Amazon’s screen-equipped speaker, is the other flagship.
The strategy echoes what Amazon did in the cloud, and that history is instructive. Inside AWS, Amazon spent years designing its own server chips — Graviton for general computing, Trainium and Inferentia for AI — and the results transformed its cost structure. Today a large share of AWS’s capacity runs on Amazon-designed silicon, and the company regularly points to the price-performance advantage as a reason customers switch. The consumer devices group is now following the same playbook, building on lessons learned in the data center.
The competitive stakes are clear. Apple has designed its own chips for more than a decade and now puts them in everything from phones to laptops to earbuds. Google builds its own Tensor processors for Pixel phones. Microsoft and Meta have in-house silicon programs for their AI infrastructure. Amazon, despite its scale, has been the laggard in consumer silicon, buying off-the-shelf processors for most of its devices. The AZ3 line is the attempt to close that gap, and the hardware chief’s comments suggest the gap will close faster than outsiders expected.
There are reasons to be cautious. Custom chips only make sense if software developers write for them, and Amazon’s consumer devices run on a mix of operating systems and app ecosystems. The company has tried custom silicon in consumer devices before, with mixed results, and switching processors across an entire product line is a multi-year project that can slip. Analysts also noted that Amazon’s devices business has historically been a loss leader — a way to sell subscriptions and services rather than a profit center in its own right — and that cheaper chips improve the economics without necessarily changing the model.
What has changed is the technology. Two years ago, running a useful AI model on a device that costs $50 to manufacture was barely possible. Today the chips that Amazon designed in-house can handle voice recognition, vision, and small language models on the device, with the heavy lifting still sent to the cloud when needed. The AZ3 Pro, aimed at Amazon’s higher-end devices, is designed for exactly this hybrid arrangement, and the company says the division between on-device and cloud processing will shift over time as chips improve.
For Amazon, the chip program is also a hedge. The company spends heavily on AI compute from Nvidia and others, and the same cost pressure that pushed AWS to design its own accelerators is now pushing the devices group to design its own processors. Every inference that happens on a device is an inference that does not need a rented GPU. In that sense, the AZ3 is not just a product decision; it is a cost decision, and Amazon’s history suggests cost decisions get executed with discipline.
The hardware chief’s message was deliberately understated — an acknowledgment that custom silicon is hard, that software matters as much as hardware, and that the payoff comes in years, not quarters. But the direction is unambiguous. The next generation of Echo and Fire TV devices will run on Amazon-designed chips, and the AI that makes those devices useful will increasingly live inside them. The cloud will still be there. It just will not be needed for everything.
