Google Lost $2.7 Billion In Talent This Week. The Real Reason Isn't Money.

Nate B Jones offers a counterintuitive reading of the week: OpenAI may look like the winner, but Anthropic may be stronger than the headlines suggest. The key signals are talent flows, the state of pre-trained models, and a major move outside the OpenAI-Anthropic rivalry: Midjourney’s push into preventive medical imaging.

What the talent moves say

OpenAI drew attention with rumors of a new model cycle and the hire of Noam Shazeer, one of the authors of “Attention Is All You Need.” Anthropic, however, also gained a major figure: John Jumper, known for his AlphaFold work at Google.

For Jones, these moves matter because they show where top researchers believe recursive model improvement may be starting to work. The story is not simply which company had the louder news cycle.

The Anthropic argument

Jones reads Anthropic’s Fable and Methuselah as evidence of fresh pre-trained models. OpenAI has leaned heavily into reasoning, post-training, memory and product harnesses inside ChatGPT and Codex. Anthropic may instead have an advantage if it currently owns the freshest, strongest pre-trained model.

That kind of advantage is expensive because scaling pre-training requires more compute. But if the model is strong enough, it can help train the next generation and create a compounding lead.

Why Midjourney may matter more

The largest signal may be outside the model race. Midjourney is using the profits from its AI product to pursue a preventive medical imaging device: a fast ultrasound-based scan that could be cheaper and more accessible than MRI.

The goal is population-scale preventive imaging, potentially tracking cardiovascular changes, cancer signals and other bodily changes over time. Doctors may worry about false positives, but Jones argues that society has never really seen what happens when imaging becomes cheap, fast and widely available.

Takeaways

• Google is framed as the week’s talent loser, with major figures moving to OpenAI and Anthropic.
• Anthropic may be underestimated if its edge is in pre-training rather than product narrative.
• OpenAI remains strong in reasoning, post-training and integrated user experiences.
• Midjourney shows how profitable AI companies can fund hardware and health innovation directly.
• The next important AI story may be less about benchmark leadership and more about concrete applications.

Source

• Chaîne: AI News & Strategy Daily | Nate B Jones
• Vidéo source: https://www.youtube.com/watch?v=h1MxhfZSTjo