Is Google’s Tensor the Answer to Pixel Tension?

Preview of Pixel 6 leaves questions unanswered.

When Google launched its Pixel 5 phone in October 2020, we commented on the move in a piece titled Pixel Miss. Although the Pixel 5 and Pixel 4a weren’t terrible devices by any stretch, they really pinpointed the difficulties facing Google. In short, differentiation had become a losing battle. As we said at the time, “The need for broad distribution of Google services across devices and platforms is a constant tension with the hardware business and the need to set Pixel apart from Android’s long line of licensees. The result is that Pixel lacks clear differentiation in software while rivals are arguably moving faster in hardware design and features”.

We concluded that Google’s smartphone strategy badly needed a reset, and the devices’ lackluster sales in the 10 months that followed only underline the urgency. Google either needs to create highly differentiated, top-end experiences or mass-market products supported by broad distribution. At the moment, it offers neither and is being outmanoeuvred in a cutthroat market.

Yesterday’s announcement of its latest Pixel 6 and Pixel 6 Pro smartphones is the first sign that Google is acting on the need for change with its hardware. As part of this, Google provided a limited preview of its new devices, which it believes have the features to compete toe-to-toe with premium rivals Apple and Samsung. The device isn’t being formally launched until October 2021, so whether this claim holds up remains to be seen. But the basis for Google’s confidence lies in its new custom Tensor chip, which will sit at the heart of its new flagship phones.

On the face of it, Google is following a familiar recipe — invest in custom silicon to allow highly differentiated experiences. This is the strategy that Apple has pursued so effectively, and which has allowed its A series chips to deliver a CPU performance that significantly outstrips the competition. However, Google’s strategy is different. Although Tensor is a system-on-chip, it’s a mix of custom and licensed components. Essentially, Google has custom-designed a Tensor processing unit (TPU), but licensed the CPU and GPU, which we believe to be Samsung Exynos parts, although this is yet to be confirmed.

The question is, what performance does this deliver that Pixel wasn’t capable of with components from its previous supplier Qualcomm? This is impossible to answer without having used the company’s new devices, but it’s unlikely that the CPU and GPU will offer any jumps in performance. Similarly, if Google is using an alternative supplier of modem and radio frequency components, this arguably represents a compromise given Qualcomm’s leadership against rivals Samsung or MediaTek.

Google is essentially doubling down on its artificial intelligence toolset to provide experiences on Pixel devices that address the lack of differentiation that Pixel hardware has struggled with in the past. Again, the jury is out on this, with early examples from Google looking interesting if not groundbreaking — for example, the Pixel 6 can use the TPU to sharpen blurred images and perfectly capture challenging scenes such as low-light sunsets.

Google has always been one of the leaders in computational photography, so it’s a natural place to start with the Pixel 6, but will this be enough to compete with flagship phones from Apple and Samsung? It’s a safe bet that Google’s vision for artificial intelligence goes beyond just photography and image manipulation, and the company’s track record is clear on this front. However, it’s difficult to see this providing dramatic results in the Pixel 6 in such a short space of time. Google’s silicon investment is about gaining significant advantages over the long term, just as it has done with server chips.

Like its efforts in the data centre, cost is also a factor here. Although it didn’t feature in Pixel previews for tech media, the new TPU and its integration into the system-on-chip should allow Google to save on manufacturing costs as it replaces parts from Qualcomm as well as Intel’s Movidius artificial intelligence chip. Speculation that this will result in a major loss for Qualcomm ignores the fact that Pixel’s market share is minimal today and that Qualcomm provides the processing for a large proportion of Android devices.

It’s too early to determine if this is the big reset that Pixel so desperately needed, or another vanity project in custom silicon that falls short. Our hope is that it’s the former but if that’s the case, Google has a host of other questions to answer, not least about the role of Pixel.

While commenting on its latest results, Alphabet suggested that its marketing spending will increase significantly in the holiday quarter, which is perhaps a sign of ambition, but success will need more than vague marketing claims of artificial intelligence smarts. A semi-custom system-on-chip will mean little without distribution, operator support and scale. Google will have its work cut out to overcome these hurdles, and its actions could make for a very interesting holiday season.