Comment by WarmWash
8 hours ago
The breadcrumbs going back a year or so point to Deepmind wanting to be research heavy, but Google being more interested in hedging AI bets by selling compute. If you ever worked hard on something that you have a lot of conviction about, only to get denied resources for it, it's a pretty big gut punch.
There is another universe where Google is hard AGI forward, freeing up as much compute as possible for Deepmind, turning away OAI and Anthropic, and offering the uncontested premier AI research lab, by probably an order of magnitude or more compute. Gemini 3.5 ultra is limited availability with unreal abilities.
But their balance sheet becomes terrifying, and it's all or nothing that this plays, er pays, out.
Right now though Google is pretty well hedged. Even Chinese models likely just mean more compute sold.
I was really expecting Google to finesse this - not competing head-on with OpenAI and Anthropic to see who can build the largest LLM, but instead being content with an extremely capable Gemini 3.6, good enough to meet their own needs, and putting maximum effort into "more than just an LLM" AGI which would then eclipse OpenAI and Anthropic.
Instead, it seems Pichai has fumbled what they had with DeepMind, and they'll now just be a me-too LLM competitor chasing OpenAI and Anthropic from behind, in a race that will never get to AGI.