Google Tensor for Pixel Smartphones: Wishful thinking or a Wise Strategy by Google?

Google has officially revealed its Google Pixel 6 and Pixel 6 Pro flagship smartphones, due for launch in 3Q 2021, while also confirming that they will be the first to be powered by its own in-house application processor chipset, named Google Tensor.

First rumored in early 2020, Google has been developing its own processors under the codename “Whitechapel”, co-developed with Samsung, for use in Pixel smartphones and Chromebooks. This initiative has now reached fruition with the official announcement that its Tensor chip will feature in its upcoming Pixel flagships, putting Google’s approach in line with that of Apple, who use their own chips in the iPhone and Mac. Google’s announcement means that it will no longer use processor chips made by Qualcomm in its flagship products, although it will no doubt have to rely on the chipset vendor for the supply of 4G/5G modems and they will continue to work closely on future products based on Snapdragon platforms.

By moving to use of its own chipset, Google aims to better integrate its hardware with applications and services, allowing improvements to a number of features and functionalities on its devices. It will also mean that some tasks can be executed partly or entirely on-device rather than solely relying on the cloud. AI workloads and machine learning processes are to benefit mostly from such deep integration. Google outlined several areas that are set to benefit, including improvements in browsing, enhanced Google Assistant, and speech recognition tasks, as well as allowing for the use of larger camera sensors on its smartphones. Other applications that could benefit from having Google Tensor onboard include predictive keying, location-based services, and video services. All these applications will benefit from the use of more distributed AI networks, enabling workloads to be balanced across end-device computing resources and the cloud.

In addition, there is expected to be a retune of the user interface in the latest OS version Android 12, which is likely to arrive alongside the Pixel 6, and is predicted to operate more smoothly through use of Tensor. Google has also assured that Tensor will be backed by its own new security core, the Titan M2, which is an upgrade version of the Titan M Security Coprocessor first used in 2018. While Google Tensor may enhance the user experience of certain applications, this move could potentially compromise advantages brought by the integrated modem-application processor (MODAP) approach the company used in previous generations of Pixel devices. These advantages include increased power efficiencies, cost effectiveness, and a more integrated design.

Google is touting the Tensor processor as a significant leap in technology that will be integrated first into its flagship smartphones. As the chip is not just meant for the high end, time will tell how quickly it manages to extend to Google’s mid-range smartphones and into Chromebooks as expected. Indeed, it will take a few generations before hardware and software are fully optimized. Google will undoubtedly be looking wistfully at what Apple has achieved with its A-series Bionic chipsets once it manages to grasp a greater level of tight integration.

However, from a wider smartphone market perspective, Google’s Pixel smartphones sales are not high volume, selling less than five million units a year. When Google announced it was to make the Pixel, they were always going to be low volume models, as the notion was that they would be used to showcase the best Android implementation, demonstrating to their ecosystem partners what could be achieved when bundling hardware with the latest Android OS and related Google applications. The introduction of Google Tensor should be seen as another step in this strategy. Indeed, with the proliferation of AI applications within the smartphone market, the proof of concept on how AI should be implemented on mobile devices will require a tight collaboration between the device computing resources and the cloud to enable more distributed AI workloads.

While Google Pixels operate under such small volumes, will this move be such a loss to Qualcomm? It was noted that Qualcomm’s shares did fall off the back of the announcement, but the company need not be too fearful as it is undeniable that Pixel flagship volumes are still minimal and Google will need to source a modem from somewhere, probably Qualcomm, much as Apple does at present.

There will obviously need to be some benchmark comparisons of the Tensor with other chipsets before passing judgment. However, if Google can show to its partners that the new chip has demonstrable benefits over the competition, notably when handling AI applications, then perhaps it could be adopted by other Android smartphone vendors. In this instance, Google may wish to sell reference designs to Android OEMs, which would include end-to-end integration from the OS to the chipset level. Such a move would put it in competition with Qualcomm and MediaTek while perhaps also filling a gap left behind in the smartphone market by the departure of Huawei’s Kirin chipsets. If this is Google’s aim, then the company will have to invest in the modem business to provide MODAP chipsets and compete effectively in the marketplace. However, time has shown that building a MODAP strategy from scratch may take several years to materialize, as has been the case for Samsung, Huawei, and now Apple.

In conclusion, success with the Tensor platform may stimulate Google to extend its reach and grow share in the mobile devices market, a strategy that has been restrictive and poorly executed up until now but one that could help it make significant inroads into the smartphone and chipset markets. Alternatively, Google may still not wish to become a fully-fledged smartphone OEM, competing with the likes of Apple and Samsung, and will keep its Pixel sales volumes low. Instead, it will just use Google Tensor for its own consumption, strengthening its strategy of showcasing the Android ecosystem to its partners, while additionally demonstrating how AI could be used within smartphones and the benefits of distributed AI workloads.

However, Google must realize that it could be a long, painful, and costly journey before it achieves its ultimate goal of creating a more distributed computing environment for its AI networks and enabled services. If the advantages brought about by Google Tensor are not enough to compensate for the drawbacks from not using the MODAP approach, then the company may have to face the dreadful reality of aborting this strategy altogether. Such a scenario would not be surprising coming from Google as the company has abandoned many other projects, such as Google Fiber and Google Loon, after realizing the complexity of their commercial implementations. At the end-of the day, the marketplace and end-users will be the ultimate judges for the commercial success – or otherwise – of Google Tensor and its strategy.