While OpenAI and Anthropic strive to create recursively self-improving AI, Google DeepMind is betting on a fundamentally different vector — World Models. This is not just a shift in priorities, but a fundamental divergence in understanding what the next era of artificial intelligence should look like.

On the surface, the Mountain View giant continues to release new models. On July 21, Gemini 3.6 Flash was released, focusing on speed and cost reduction rather than absolute power. According to Google, the new version generates 17% fewer tokens than its predecessor, directly lowering the cost per response. However, in the Artificial Analysis ranking, this model only placed 10th, falling behind all key competitors.

Betting on Physics, Not Text

The key difference in DeepMind's strategy is abandoning the race for records in "next token" benchmarks. Instead, the lab focuses on embodied AI and systems that understand the laws of the physical world: gravity, cause-and-effect relationships, and movement in space. The Genie 3 and Gemini Robotics projects are already listed under the World Models section on the company's website.

In May, DeepMind expanded Project Genie by integrating it with Street View and introduced SIMA 2, an agent learning to interact with virtual 3D worlds. This is a fundamentally different approach compared to the recursive self-improvement chosen by competitors.

Risk or Precise Calculation?

Anthropic co-founder Jack Clark, in his essay dated May 4, estimated the chances that AI could independently manage research by the end of 2028 at 60%. At the same time, he called DeepMind "the most restrained among the big three," citing internal lab safety documents. Restraint in this context means caution bordering on the risk of falling behind.

Meanwhile, Anthropic is already recording cases where Claude writes over 80% of the code deployed in products. In April, the performance of their models increased 52 times. For comparison, an experienced engineer requires four to eight hours of work for a fourfold increase.

Financial Strain

Alphabet continues to generate enormous revenues — $119.8 billion in the second quarter, of which $63.3 billion came from search advertising alone. However, expenses are growing at a faster pace. During the quarter, the company allocated $44.9 billion for data center development and equipment purchases — nearly double the amount from a year ago.

Free cash flow turned negative for the first time, reaching minus $5.86 billion. To cover the deficit, Alphabet issued $49.6 billion in shares and raised $20.3 billion in loans. Long-term liabilities grew from $46.5 billion to $98.2 billion over six months. The expense line related to AI development incurred a loss of $5.79 billion, compared to $3.37 billion a year earlier.

Analyst's Conclusion

Google's strategy is a long-term game. If DeepMind's world models prove their viability, the current "lag" in generative tests will turn out to be a brilliant tactical move. However, with current expense growth rates and negative cash flow, Alphabet has less and less room for error. The release of Gemini 4 will be a key moment, showing whether this risky pivot was justified.