Most people remember 2022–2023 as the era when OpenAI and ChatGPT defined generative AI, while Google, caught in the innovator’s dilemma, looked for a while like it was done for.
But starting in mid-2025, Google didn’t just catch up; by the end of the year it had even briefly pulled ahead.
This WSJ piece walks systematically through how Google came back into view. It wasn’t just marketing; it got three things right:
-
More than a decade of research foundations From Google Brain and DeepMind to the Transformer and TPUs, a lot of what everyone uses today grew from trees Google planted years ago, and plenty of cutting-edge research was still sitting in the lab.
-
Willing to spend big and rebuild When Gemini went fully multimodal (text, images, audio, video), Google rewrote search at the same time (AI Mode). That was Google’s answer once it overcame the innovator’s dilemma: if it was going to be disrupted, it would do the disrupting itself.
-
Hardware is the hidden ace Its own TPUs and the Ironwood chip opened up a clear gap in the cost of large-scale inference. There were even reports it would sell chips to Meta, and NVIDIA’s stock fell 7% that day.
And one of the most dramatic, and most worth recording, accidents: an image model casually named “Nano Banana” at 2:30 a.m. shot straight to the top of the leaderboards, made Gemini the number one download on the App Store, and pushed monthly active users past 650 million. After an achievement like that, Sundar Pichai’s internal memo said just one line: “We’re launching at the scale of Google.”
My takeaway after reading it is that in the AI era, it isn’t “whoever gets popular first wins.” It’s whoever lasts longest, has the deepest foundations and is most willing to start over. Google didn’t suddenly get stronger. It finally overcame its internal contradictions and found the rhythm it’s best at.
(This WSJ piece is well worth reading. It lays out the timeline, the inside story, the technology and the power structure all at once. The link is in the comments.)