Google is changing its AI strategy: why DeepMind is choosing world models over the token race
While OpenAI and Anthropic are making aggressive bets on recursive self-learning and creating AI capable of independently designing the next generations of algorithms, Google DeepMind appears to have chosen a fundamentally different path. Externally, the giant continues to release models and generate revenue, but its focus has shifted from chasing raw power to understanding the physical world.
DeepMind's New Course: From Tokens to Reality
The latest model, Gemini 3.6 Flash, released on July 21, emphasizes speed and cost rather than absolute leadership in benchmarks. According to Google, the new version generates 17% fewer tokens, reducing the cost per response. However, in the Artificial Analysis ranking, this model only placed 10th, lagging behind competitors. This does not indicate a halt in development — the company has already launched the largest training run for the future Gemini 4, and Gemini 3.5 Pro is undergoing testing with partners.
The key difference in strategy was voiced by Google CEO Sundar Pichai: he links the roadmap to the development of personal agents and world models, not to positions in rankings. DeepMind openly declares this course: both Genie 3 and Gemini Robotics are in the sections of world models and embodied AI. These systems analyze gravity, motion, and causal relationships, predicting the development of events in space, rather than simply selecting words.
Risks and Market Reaction
Investors are reacting sharply: in June, Alphabet's shares fell by 6% after two leading researchers left for competitors. Anthropic co-founder Jack Clark published an essay in May, calling DeepMind "the most restrained of the big three," estimating the probability that AI could independently manage research by the end of 2028 at 60%. His arguments rely on documents from DeepMind itself, including an article on AI safety issues published in 2025.
For comparison, Anthropic acts more boldly: by May 2026, Claude had written over 80% of the code that the company integrates into products. In April 2025, their models increased productivity by 52 times, whereas a year earlier this increase was only 2.9 times. For an experienced engineer, achieving a fourfold increase would have required four to eight hours of work.
Why Google Shouldn't Be Counted Out
Despite the apparent lag, there are two important factors. First, Google maintains its lead in machine learning research tests: MLE-Bench, which evaluates models' ability to independently create AI architectures, showed Gemini 3's result at 64.4% in February, and version 3.6 Flash scored 63.9% in July. Second, the IT giant retains enormous reach: the Gemini app audience reached 950 million people per month.
However, the wait-and-see position is costing the corporation increasingly more. In the second quarter of 2024, the company's revenue reached $119.8 billion, but expenses are rising. Over three months, Alphabet allocated $44.9 billion to developing data centers and purchasing equipment — almost double the amount from a year ago. Free cash flow was negative $5.86 billion, and long-term liabilities grew from $46.5 billion to $98.2 billion over six months. The expense item related to joint AI developments resulted in a loss of $5.79 billion, compared to $3.37 billion a year earlier.
What to Track in the Next 30 Days
- Whether the release of Gemini 3.5 Pro will take place and what positions the model will achieve in tests.
- Whether DeepMind will present new data on world models for Gemini 4.
- Whether Alphabet's cash flow will return to positive values in September.
- Whether Demis Hassabis will articulate a clear position regarding self-learning technology.
The launch of Gemini 4 will be a defining moment for the company's strategy. If the approach to creating world models proves itself, the current slowdown will turn out to be a precise calculation. The upcoming financial report will show the real limits of Alphabet's resources in this technological race.
Expert opinion: Google's strategy is not a lag, but a conscious choice. While competitors accelerate development through recursive self-improvement, DeepMind is betting on understanding the physical environment. If their world models can truly simulate reality, rather than just predict text, this could redefine the very paradigm of AI development. However, the high cost of this wait-and-see position and Alphabet's growing debt make the bet extremely risky.