Jeff Dean, Google AI chief. Photo: News1
Jeff Dean, known as Google’s “legendary engineer,” has left the company after 27 years. Dean, who joined Google as its 30th employee in 1999, is a key figure who built today’s Google, from search infrastructure to Google Brain. In addition to Dean, world-class artificial intelligence (AI) researchers including Sanjay Ghemawat, Oriol Vinyals, and Quoc Le also packed their bags. They plan to establish a startup called “Discovery Loop” that uses AI to automate scientific and engineering research. Alphabet, Google’s parent company, will also invest in the firm.
Google’s remaining AI organization is also undergoing major changes. Demis Hassabis, the Nobel laureate in chemistry who led Google DeepMind, is stepping back from day-to-day organizational management to serve as Alphabet’s chief scientist (CSO) and DeepMind chairman. Instead, Koray Kavukcuoglu, DeepMind’s chief technology officer (CTO), will take responsibility for the actual operation of the AI organization. He will also serve as Google’s chief AI architect and report directly to CEO Sundar Pichai.
Google Created the Transformer but Ceded the Lead to ChatGPT
As figures who symbolize AI research depart or change roles, some observe that Google’s research-centered culture may be weakening. In fact, reflecting such concerns, Google’s stock fell as much as 4% on the 5th (local time), when their departures were decided. However, it is difficult to view this as a sign that Google is stepping away from research. Google is currently conducting research not only in machine learning but also across a wide range of fields including quantum computing, health and bio, robotics, and security. Still, through this reorganization, the company appears to have placed greater weight on the stage of implementing research achievements into products and services and connecting them to actual business results, rather than on the research achievements themselves, in the AI competition.
Photo: Created by ChatGPT
Google is the company that laid the foundation of the modern AI industry. The Transformer, presented in the “Attention Is All You Need” paper published by Google researchers in 2017, became the core technology of generative AI, including ChatGPT and Gemini. But the company that actually dominated the generative AI market first was OpenAI. ChatGPT popularized generative AI, and Anthropic also rapidly penetrated the corporate market with Claude at the forefront. Google itself experienced the fact that having world-class researchers, papers, and computing infrastructure does not naturally mean the market will follow.
Now Google appears set to focus not on whether it can develop new AI technology, but on how quickly it can turn that technology into products used by hundreds of millions of people and connect it to the revenue and profit of its existing businesses. In the second quarter of this year, the monthly active users (MAU) of the Gemini application reached 950 million. The number of developers using Google’s AI models exceeded 9 million per month, and the tokens processed by Google’s model API reached about 22 billion per minute, a 37.5% increase from 16 billion in the previous quarter.
Gemini is also linked to Google’s actual business performance. Alphabet’s revenue in the second quarter of this year was $119.8 billion, exceeding the market forecast of $116.9 billion. In particular, Google Cloud’s revenue recorded $24.8 billion, a surge of 82% from the same period a year earlier. Contrary to concerns that generative AI would replace existing search, search revenue also grew 17%.
While OpenAI and Anthropic must grow AI itself into one massive business, Google can deploy Gemini simultaneously into services already used by billions of people, including search, advertising, cloud, YouTube, Android, and Workspace. Accordingly, Google’s paths to AI monetization are also diverse. If a company uses Gemini-based services, cloud usage may increase, and if consumers use AI search, it connects back to the advertising business. In Workspace, the company can collect corporate subscription fees. This means Google can build a structure in which AI generates more revenue in Google’s existing businesses, not just earning money from Gemini itself.
Google Pours in $200 Billion — Now ‘AI Investment Return’ Is Key
However, AI competition comes with enormous costs. To train and serve AI models, a company must secure not only data centers, servers, and networks but also its own AI semiconductors.
Photo: Created by ChatGPT
Alphabet raised its capital expenditure (CAPEX) forecast for this year to between $195 billion and $205 billion. That is an increase of up to $30 billion from the $175 billion to $185 billion expected earlier this year. The company decided to expand investment in data centers and servers to respond to surging AI and cloud demand. As investment surged, Alphabet’s free cash flow (FCF) in the second quarter of this year turned to a deficit of $5.9 billion. This means that the money actually left over after excluding capital expenditure from the cash the company earned through operating activities turned negative. It is a passage that shows how capital-intensive a business building AI infrastructure is.
Therefore, when investors evaluate Google’s AI business going forward, looking only at Gemini’s benchmark scores will not be enough. They must look at both the money poured into AI and the money AI generates. If the 82% growth rate at Google Cloud continues, it will be a strong signal that large-scale AI infrastructure investment is connecting to increased demand from corporate customers. In AI search, whether the company continues to grow search revenue itself without cannibalizing existing search advertising is important. In this case, Google can protect the advertising business, its largest cash generator, even in the era of generative AI. Gemini’s monetization is also key. More important than MAU surpassing 1 billion is how much additional revenue these users generate in cloud, advertising, and subscriptions. The last is cash flow. If FCF recovers quickly again after executing CAPEX of around $200 billion, the current spending can be evaluated not as a cost but as a forward-looking investment for future growth. Conversely, if users increase but cash-generating power continues to weaken, questions about AI investment efficiency will inevitably grow.