AI competition shifts toward integrated stacks and personal models

AI competition is moving beyond benchmark gains toward control of the full product stack. SpaceX completed its $60 billion all-stock acquisition of Anysphere, issuing about 389 million Class A shares and folding Cursor into SpaceXAI, while Grok 4.6 launched for coding and agentic work with a 500K context window and $2/$6 per million tokens below 200K prompt tokens.

Anthropic is reportedly in talks to acquire Decart for about $6 billion, a move aimed at model infrastructure, world models and compute optimization. Z.ai released GLM-5.3, built through post-training on the same 743B base as GLM-5.2, with a focus on coding and cyber defense. NVIDIA introduced Nemotron 3.5 Lightning and NeMo Switchyard, while DeepSeek-V4-Pro-0813 reached general availability across app, web and API.

River AI, founded by former xAI co-founder Igor Babuschkin, raised $1.1 billion to build tools for personally owned models trained on users’ own data, rewards and preferences. Research updates highlighted Microsoft’s full-bandwidth transformer, Salesforce’s DarwinX approach to evolving agent harnesses, and work from Anthropic on multiagent systems and verified mathematics.