Google AI Timeline: Transformers, Gemini, and AI Search

An advertising company with a search engine attached

Google is a subsidiary of Alphabet Inc., which Sundar Pichai has run as chief executive since 2019, having led Google itself since 2015. The business is still, by revenue, an advertising company: roughly three-quarters of Alphabet's revenue comes from ads sold against Search, YouTube, and its ad network, with Google Cloud, subscriptions, and hardware such as Pixel and Nest making up most of the rest. Search itself holds close to ninety percent of global search share, which is the distribution advantage every other line of business, including AI, gets to lean on. [2]

That scale has drawn regulatory scrutiny that predates and now overlaps with the AI race: a US federal court found in August 2024 that Google held an illegal monopoly in search and search advertising. Google's AI products, including Gemini in the Gemini app and AI Overviews and AI Mode in Search, now sit inside the same products a court has already ruled anticompetitive, which is a separate and ongoing legal exposure from anything specific to its AI work. [2]

The paper and the acquisition that set up everyone else

Google's deepest claim on AI history predates ChatGPT by more than five years. "Attention Is All You Need," published by Google researchers in 2017, introduced the transformer architecture that every major model since, GPT included, is built on. Google also open sourced TensorFlow in 2015 and published BERT in 2018, both of which shaped how the field trained and shared models years before the current wave of products existed. [3]

Atlas interpretation: The uncomfortable fact for Google is that it did the foundational research and still lost the first round of the product race. Buying DeepMind in 2014 for a reported £400 million, when DeepMind had about fifty employees and no shipped product, gave it the lab that would eventually produce Gemini. But ChatGPT still shipped from a smaller competitor first, in November 2022, and forced Google to run its own version of the same emergency Sam Altman would later declare against Gemini. [4]

Google on the timeline

The 2014 acquisition of DeepMind and the 2017 transformer paper are the two turning points on the timeline from Google's research history. Its third turning point is recent and different in kind: Apple basing its next foundation models on Gemini, announced in January 2026. Apple said its intelligence features would keep running on-device and through its own Private Cloud Compute, but a multi-year licensing deal for the underlying models put Google's technology inside the one major device ecosystem it does not control, and analysts were already comparing the arrangement to the Microsoft antitrust cases of a generation earlier. [5]

Google's product response to ChatGPT shows up across several major events: Gemini's December 2023 launch, introduced as three model sizes with native multimodality, was undercut on day one by a demo video later reported to have been edited for pace, a credibility cost the launch never fully shed. AI Overviews arrived in Search in May 2024, putting a generated answer above the links in the product that funds the company, over publisher objections about referral traffic. By August 2026, Google reported the Gemini app had passed a billion monthly users, its fastest product ever to reach that mark. [6][7]

Google also used the same period to buy its way into AI-adjacent security: it agreed to acquire the cloud security company Wiz for $32 billion in March 2025 and closed the deal a year later, folding Wiz into Google Cloud. It was, at signing, the largest acquisition in Alphabet's history, and it was a security company rather than a model lab, a reminder that Google's AI-era spending is not confined to compute and talent. [8]