One morning in August, at exactly 2:30 a.m., an artificial intelligence project manager at Google uploaded a new tool from the DeepMind lab to a platform that evaluates AI models. This tool, a fast image generator, was intended to help Google catch up in an area where OpenAI's ChatGPT had been successful. Manager Naina Raisinghani, known for working night shifts, came up with the name Nano Banana—a combination of two nicknames given to her by friends. Within days, Nano Banana reached the top of the rankings on the LM Arena platform, trended on X, and exceeded Google's usage expectations. By September, the Gemini app had become the most downloaded app in Apple's store.
Early Days and Investment in AI
Google, owned by Alphabet, previously found itself at a disadvantage when OpenAI attracted hundreds of millions of users to ChatGPT. But thanks to deep roots in science, billions in investments in its own hardware, and leadership changes, the situation turned around. In November, Google launched its most powerful model, Gemini 3, which surpassed its competitors and became the most capable chatbot. This success sent Alphabet's shares higher and triggered panic at OpenAI, which launched "Code Red"—an emergency effort to improve ChatGPT. OpenAI then responded with a more powerful version of ChatGPT, which still has more users than Gemini.
Google CEO Sundar Pichai emphasized the scale of this effort in an internal message to employees: "I love that we're launching at Google scale." When Pichai took over as CEO in 2015, artificial intelligence was primarily of interest to scientists. A year later, he declared that the company, known for its search engine, maps, and productivity tools, would fully focus on AI. In a blog post, he wrote that the previous decade had belonged to smartphones, but the next decade would be about a world where computing is available everywhere thanks to AI.
Google already had the foundations in place: the Google Brain division, founded in 2011 by Jeff Dean, who helped develop the neural networks used by today's large language models. Several years later, it acquired DeepMind, a London-based lab founded by Demis Hassabis, a chess prodigy who later received a Nobel Prize for an AI system that assists with biomolecular research. Google also began designing its own AI chips, called tensor processing units (TPUs), which consume less energy than conventional processors or graphics cards. These chips proved crucial for both Google and the entire industry.
A Cautious Approach
Initially, Google proceeded cautiously with chatbot development. Some executives and researchers were concerned about risks such as the spread of inaccurate or biased information. Julia Winn, a former Google Brain employee, said that chatbots were not considered central to the company's ambitions and that they easily produced racist or sexist responses in tests. This caution frustrated some researchers, who left the company.
In August 2022, Google introduced the LaMDA model with conversational capabilities, making it available to a limited number of people through the AI Test Kitchen app. Users could test features such as "Imagine It," "List It," or "Talk About It," where the conversation focused solely on dogs.
Three months later, OpenAI launched ChatGPT, which not only held conversations but also wrote stories, jokes, and code. Within five days, one million users had signed up without significant restrictions. Google employees who had worked on similar technology for years were angry. Analysts wondered whether Google had missed a major wave and whether chatbots might threaten its search and advertising business, which generated $254 billion in 2022 (approximately CZK 5.8 trillion).
Joining Forces and Project Magi
Jeff Dean, Demis Hassabis, and James Manyika, who joined in 2022, merged the DeepMind and Brain divisions. In January 2023, they presented a plan for the most powerful model to Alphabet's board of directors. In February, they launched the LaMDA-based Bard AI, but the introduction failed: a video showed an incorrect answer about the James Webb Space Telescope, causing Alphabet's shares to fall by 8%.
At that point, co-founder Sergey Brin returned after a conversation with OpenAI's Daniel Selsam at a party. Brin, who had retired, decided to return and work on AI. In 2023, the leadership sought to align the teams' efforts, even though the divisions had different cultures—Brain focused on research, while DeepMind focused on products.
Google had an advantage: it financed development with billions in profits, unlike OpenAI, which had to seek investors. However, it also had to protect its 90% share of the search market. It therefore launched Project Magi under the leadership of Liz Reid, who became vice president of search in 2024. The goal was to redesign search so that it could quickly provide answers from multiple sources.
Liz Reid explained in an interview: "People don't just use search; they depend on it. If you mess it up, you'll hear about it from your mother, a friend, or your child."
The Success of the Gemini Model
The end of 2023 brought the first version of Gemini, trained on text, code, audio, images, and video—which is why it took longer than ChatGPT, which initially focused mainly on text. The first Gemini lagged behind, but the ambitious approach paid off. At a Stanford University event in December, Sergey Brin said: "I think we were still benefiting from that long history."
Since returning, Brin has worked on problems with Gemini and helped bring back researchers Daniel De Freitas and Noam Shazeer by acquiring their startup for $2.7 billion (approximately CZK 62 billion). In May 2024, Google introduced AI Overviews—brief AI summaries at the top of search results, which led users to submit more complex queries.
Then came AI Mode, a conversational search feature launched in May. Liz Reid described how, after developing many versions, the team began using the tool themselves rather than merely testing it. In August 2024, a judge ruled that Google had an illegal monopoly in search, but its $20 billion annual agreement with Apple (approximately CZK 460 billion) could continue with minor changes, thanks to the argument that AI chatbots had ended the monopoly.
Nano Banana boosted the share price in August 2025. Josh Woodward, who oversees the Gemini app and Google Labs, called the launch a "successful disaster"—billions of images were generated, requiring emergency loans of computing time. By October, Gemini had more than 650 million monthly users, up from 450 million in July.
November's Gemini 3 caused another shortage of computing capacity, but chips such as Ironwood reduced costs. A report about negotiations with Meta to sell it billions of dollars' worth of chips caused Nvidia's shares to fall by 7%. In December, Sundar Pichai wrote to employees: "We are ending 2025 in a great position. When I look back one year, it is amazing to see the progress."
Source: wsj.com
