A source details how Google built the Pixel 10's Tensor G5 chip without Samsung's help, using in-house and off-the-shelf IP, and partnering with Arm and others
The Tensor G5 won't be all that different from the previous generations. — • — • — • — TL;DR
Context & Ripple Effects
Google’s Tensor program began as a Google-designed chip for the Pixel 6, then extended into the Pixel 7’s focus on personalized photo, video, security, and speech features. But an assessment of the first four generations flagged weak performance and power efficiency, raising the stakes for greater control over the next design. Google’s original Tensor-based Pixel launch and the later critique of Tensor’s efficiency and performance frame this report as an evolution in ownership rather than a clean architectural reset.
The reported use of Arm and other partners alongside internal and licensed IP suggests Google is assembling a broader chip-development capability while retaining external specialist inputs. Subsequent coverage that Google would move Tensor G5 fabrication from Samsung to TSMC makes the design-side separation especially consequential. The reported TSMC foundry move
First-order effects
- Google takes direct responsibility for Tensor G5’s design integration, using its own IP where available and off-the-shelf blocks where it is not; Arm and other partners gain a more explicit role in the development chain.
- Samsung is no longer described as a design collaborator on this generation, even though the report says the resulting chip will not be substantially different from earlier Tensor designs.
Second-order effects
- Google can more directly align future Tensor revisions with Pixel software and on-device features, but must also absorb more of the integration and validation burden that a partner-led design would share.
- Samsung’s reduced role in design, combined with the reported fabrication change, increases pressure on it to compete for mobile-chip business on manufacturing and ecosystem terms rather than relying on a tightly coupled Tensor relationship.
Third-order effects
- If Google sustains this model, custom mobile silicon is likely to become less about wholly proprietary CPU designs and more about controlling the system integration of licensed, specialized IP around a device maker’s software priorities.
- The near-term continuity of Tensor G5 also shows that internal chip ownership alone does not guarantee an immediate leap in performance or efficiency; differentiation may emerge gradually through successive integration cycles.
The trend: This is one data point in the shift toward vertically integrated AI-device stacks built from a mix of internal design control and external semiconductor IP and manufacturing partners.