While Google's Tensor-class SoCs have always represented an intricate balancing act between performance and cost, the upcoming Tensor G6 chip - which bears the internal codename Malibu - appears to be tilting that balance ever so slightly in the favor of cost controls, as evidenced by Google's apparent decision to use a refreshed version of a 5-year-old GPU.
Of course, in the age of relentless memory-focused 'chipflation,' Google's quirky gambit might well yield an optimal mobile-focused SoC, especially if the Tensor G6's TPU steps up, and manages to paper over any GPU-led shortcomings.
Google Tensor G6 chip's architecture and specifications
TSMC's process node
Just like Apple's A20 chip, Google's Tensor G6 is expected to leverage TSMC's 2nm-based N2 fabrication process, and eschew the more expensive N2P process, which reportedly bestows a performance boost of between 5 percent and 10 percent relative to the N2 node.
At any rate, the N2 process will still entail material improvements in efficiency for the Tensor G6 chip vs. its Tensor G5 counterpart, which was based on TSMC's 3nm process.
CPU
The Tensor G6 is now expected to come with a 7-core CPU vs. an 8-core one on its predecessor, consisting of:
- 1x ARM C1-Ultra core clocked at 4.11GHz
- 4x ARM C-1 Pro cores clocked at 3.38GHz
- 2x ARM C-1 Pro cores clocked at 2.65GHz
As is typical with most of Google's Tensor-related decisions, we can peg this regression to a 7-core CPU on the tech giant's ever-present cost considerations.
GPU
Back in late April 2026, Mystic Leaks spurred somewhat of a shock by disclosing that Google's Tensor G6 chip would sport the PowerVR CXT-48-1536 GPU that launched all the way back in 2021.
Thankfully, some much-needed clarity arrived soon thereafter, when it emerged that the GPU would be a CXTP variant, where the "P" presumably denotes improved power efficiency - akin to Imagination's DXTP-class GPU that launched in 2025.
As such, we can now conclude with a reasonable amount of certainty that the Tensor G6 chip will feature Imagination's PowerVR CXTP-48-1536 GPU.
At any rate, the fact remains that the Tensor G6's GPU will be based on a fundamental design that launched around 5 years back, with this quirky design choice again likely prompted by Google's penny-pinching ways.
TPU and other specifications
The Tensor G6 chip is expected to sport a dual-TPU design, bearing the codename "Santafe":
- A bespoke TPU to handle major AI workloads.
- A nano-TPU to handle relatively simple AI tasks much more efficiently.
Google's next-gen Tensor SoC will also feature the new Titan M3 security chip. For the benefit of those who might not be aware, Google's Titan-class security coprocessors provide hardware-level protection for user data, including encryption keys and biometric information.
The chip is also expected to sport a new Image Signal Processor (ISP), codenamed "Metis," as well as a Graphics eXtension Processor (GXP) unit. This ISP/GXP combo will work with the Tensor G6's TPU to allow for superior computational photography, video processing, and hardware-accelerated image handling.
Elsewhere, the SoC is expected to come equipped with the LPDDR5X RAM. Also the chip will likely eschew the adoption of UFS 5.0 storage and retain support for UFS 3.1 and UFS 4.0, depending on the Pixel 11 variant. There is a slim chance, however, that Google might onboard support for UFS 4.1.
Cost and expected launch date
The cost of Google's Tensor G6 is not known at this time. Even so, we do know that the Tensor G5 cost Google roughly $65 per unit. Of course, given the ongoing memory 'chipflation,' the Tensor G6 is likely to cost a bit more than its predecessor, especially as a typical mobile-geared LPDDR5 module cost around $10/GB in Q1 2026, and is expected to average between $19.3/GB and $19.8/GB in the second quarter of 2026.
As far as its launch date is concerned, the Tensor G6 chip will debut within the Pixel 11 series, which is tentatively set to come out in August 2026.
