Google is now the best at AI — but is it enough?

It took just under three years, but Google has finally caught up and overtaken its AI competitors.

You remember how it was. When OpenAI released ChatGPT at the end of 2022 and created a world-wide sensation, Google was caught napping— and panicked. Despite the fact its researchers had introduced the architecture behind the new language models five years earlier, and despite having the now Nobel Prize-winning AI lab Deepmind in the corporate family, it was OpenAI, not Google, that developed the first widely usable model.

There’s a lot of rumor and speculation floating around about what it was like at Google at the time. CEO Sundar Pichai supposedly declared “Code Red” internally, something he later denied. Larry Page and Sergey Brin reportedly stepped back into the company in “founder mode.” Either way, Google reorganized and put Deepmind and its head, Demis Hassabis, in the driver’s seat to focus on the new large language models (LLMs).

Google’s first attempt was Bard; it was released in the spring of 2023 and made serious mistakes from the get-go. Clearly, despite its muscle, Google had a long way to go to catch up with OpenAI and its patron Microsoft.

The journey lasted three years, because when Google released Gemini 3 last month, it was immediately clear Google had — at least for now — taken the lead in AI by a fairly good margin. Gemini 3 crushes rival tools in most measurable parameters, and more importantly, is almost unanimously praised by users.

It is not only Gemini 3 that makes Google right. Competitors are sure to release new top models that will keep the rally at the top of the LLM Arena going. It’s how they’ve done it that puts Google in a unique position.

Gemini 3 is not trained on the same Nvidia GPUs everyone else uses, but on Google’s own TPU chips. TPUs are ASIC chips that have a single use case (or in this case two, AI training and inference). In other words, they’re much more specialized than a GPU used for just about everything.

That makes them more cost-effective, but more importantly, they’re Google’s own chips. This means Google now has the entire vertical in AI: its own chips with its own software layer; its own top models; its own data centers; its own cloud services; its own consumer services;. its own operating systems; its own browsers; and its own mobile devices. It’s basically the same recipe that created the modern Apple. (This is a journey that Microsoft also wants to take, with its own chips and its own models).

Add to this a huge competitive advantage in a data set that no one else comes close to, from years of search indexing, from all of Youtube, from Street View, from Waymo cars — the list goes on.

On the chip front, Google will certainly continue to buy and lease Nvidia chips to its cloud customers. But there are already reports of big customers wanting to use Google TPUs.

Who’s panicking now? OpenAI chief Sam Altman is reported to have sent an internal memo to his staff admitting Google’s progress on Gemini 3 and warning of a “tense atmosphere” and “financial headwinds.” He argued that OpenAI now needs to focus on “catching up fast.”

(For its part, Nvidia went on X and congratulated Google, while pointing out that its own technology is “a generation ahead.”

It’s still important to remember one thing: despite having the full stack in AI, Google is not an “AI company,” in the same way that it’s not really a search service either. Google is still an advertising company, the largest in the world.

Google Cloud, the IT services if you will, is certainly an increasingly important part of the equation. In the last quarter, Cloud revenue grew by 34% and now accounts for 15% of Alphabet’s revenue.

But 72% of Alphabet’s revenue still comes from ads — and it was that revenue that prompted (or didn’t prompt) Pichai to issue that Code Red. The concern was that ChatGPT would compete with Google on search and cut into advertising revenue. And the ambition now is to have the best AI solutions on the market to continue selling ads.

All this means that Google is less sensitive to the bursting of what has been called an “AI bubble.” It is neither dependent on “AI revenue” nor on the most vulnerable AI companies. If I were to make a prediction, Google is probably the company that would hold out best in the event of an AI crash.

However, Google is all the more sensitive to what AI does to the internet economy. Where will companies spend their marketing money when chatbots take over? Will AI summations send enough traffic to make advertisers think it’s worth it? What will be the price per click?

That’s where you should look to see whether Google really will be the “winner” in AI — in the ad market.

This column is taken from CS Weekly, a personalized newsletter with reading tips, link tips and analysis sent directly from Editor-in-Chief Marcus Jerräng’s desk. Would you like to receive the newsletter on Fridays? Sign up for a free subscription here.