How Google’s Weird Chip Strategy Won The AI War

Ajouté : 2026-06-02

[00:00:00]Google's tensor chips have a weird reputation. They overheat on using cameras. They throttle during gaming.

[00:00:06]Performance is nowhere near other flagships. And somehow Google kept making them anyway. Which gets really interesting once you notice Google may have been never trying to win the normal chip race in the first place. Because while everyone else was building faster smartphones, Google was quietly building AI phones years before the industry fully understood what they even meant.

[00:00:26]Let me explain. When Tensor first launched inside the Pixel 6, the reaction was mixed almost immediately.

[00:00:33]People expected Google's first custom chip to be their Apple silicon moment.

[00:00:37]Instead, reviewers found a bunch of major issues ranging from heating to small battery life. And honestly, a lot of those complaints were well valid.

[00:00:45]Google still trails flagship chips pretty heavily in raw performance. A Snapdragon 8 Elite device can absolutely demolish Tensor G5 in sustained gaming.

[00:00:54]Apple's A series chips still dominate single core performance. You can open almost any benchmark chart and tensor ends up sitting awkwardly below the phones it directly competes against and gaming especially exposed the gap fast.

[00:01:07]High-end Snapdragon phones were running games at crazy frame rates while tensor devices were getting hot through Genshin Impact sessions and throttling down to survive. This confused people because these were in accidental weaknesses Google somehow failed to notice for five story generations. You don't accidentally build a chip architecture centered around a giant machine learning pipeline and then forget gamers exist.

[00:01:29]Tensor was strange from the beginning because Google prioritized completely different workloads than the rest of the industry. Apple and Qualcomm were obsessed with traditional flagship metrics CPU, GPU and battery. Google cared about those two. Well, obviously they just didn't treat them as a main event and that's where their real focus lies. That's why Tensor launched with a massive emphasis on ondevice machine learning even before the current AI boom. Back in 2021, most companies still treated AI like a background feature.

[00:02:00]Google treated it like the future operating system, adding a ton of AI features to Pixels. Object removal, life translation, filtering spam calls. Those features sound normal now because the industry caught up. Back then, they stood out immediately on Pixels. And Google wanted these AI tasks happening directly on the phone instead of constantly bouncing everything to the cloud. That changes privacy and speed both at once. A normal smartphone chip can technically run AI workloads too.

[00:02:26]Snapdragon chips already had NPUs. Apple had neural engines, but Google had a different advantage. They controlled everything from Android to the camera pipeline to now the silicon 2. That let them optimize tensor specifically around Google's own machine learning systems instead of building a general purpose chip meant to satisfy every Android on earth. And that's why cancer starts making a whole lot of sense even with all its faults. Google's objective was never really to build the fastest Android chip. They were trying to build the most Google chip possible. That's a huge difference cuz before Tensor, Pixels used Snapdragon processors like almost every Android flagship, which also meant Google had the same limitations as everyone else. They had to wait for Qualcomm's road map, work around their priorities, and share similar hardware foundations with Samsung, OnePlus, and basically the entire Android ecosystem. Tensor changed that. Now Google could build features around hardware they specifically designed for them. That level of vertical integration is exactly why Apple's ecosystem feels so tightly connected. Google clearly wanted a version of that control and you can actually see the shift happening over time. Pixel phones slowly stopped competing on traditional spec sheet flexing and leaned harder into software intelligence instead. Once you look at it that way, a lot of Google decisions stopped looking weird. Hell, even the compromises started making sense. But for a while, the industry didn't fully buy into that vision. Then 2023 happened. Suddenly every tech company on earth started screaming about AI. The two most major players Samsung and Apple started racing towards AI phones. Every keynote had more and more discussion about AI features. And weirdly enough, Google unusually prepared for it because they' already spent years building phones around exactly those workloads.

[00:04:11]And that's why Tensor aged in such an interesting way because it was built for AI way before the trend. If you like this video, consider subscribing, liking, sharing, and commenting too. And I'll see you in the next one.