Not to be outdone by Apple and Google, Amazon has reportedly begun development of its own artificial intelligence chip.
According to The Information, Amazon has started designing a custom AI chip for Echo devices, with the purpose of improving Alexa and the quality of her responses as well as her response time. The AI chip would allow Alexa-powered products to handle more on-device processing, rather than having to communicate with the cloud. Amazon reportedly has about 450 people with chip expertise working on the effort.
These experts come from Amazon's recent acquisitions and hires. For instance, the company acquired Israeli chipmaker Annapurna Labs in 2015. But keep in mind one of the more interesting aspects about this latest move by Amazon is that both Apple and Google have embarked on similar paths. They've also developed custom AI hardware that can handle a device's computationally-intensive AI tasks.
- Is Apple making its own chips for Macs so it can eventually ditch Intel?
- Google Assistant might come to all Chromebooks, leaked code shows
- Apple might bring Animoji to iPad and FaceTime in future iOS update
Apple has started developing its own chips for the iPhone, for instance, such as the phone’s graphics processor and power management unit. The Cupertino-based company also designed a new “neural engine” with the A11 Bionic chip, which does all the on-device processing for machine learning algorithms that enable some of the iPhone's more smarter features, whether that be Face ID or ARKit and AR apps.
Meanwhile, Mountain View-based Google has its custom ASIC processor, the Tensor Processing Unit, which is for the company's TensorFlow AI training platform, the TPU. The TPU is the foundation of Google subsidiary DeepMind’s AlphaGo system. Google also started developing custom AI chips to power smart devices like its Clips AI camera. It even has a new custom image processor inside the Pixel 2.
In other words, if Amazon wants to stay competitive in the smart home race, it needs to control the design of its own AI chip, even if it won't actually product the chip itself, considering it doesn't yet have much experience in fabrication.