Google's model delay and NotebookLM push split the centre of value

Facts July 17, 2026 · Reading July 17, 2026

Google delayed the broader release of Gemini 3.5 Pro after internal testing, pushing the rollout back by several months. Google also began a global rollout of major NotebookLM upgrades, moving the product to Gemini 3.5 and Antigravity and adding secure per-notebook cloud computers for code execution, agentic reasoning, new output formats and more than 100 curated software skills.

This appears to shift the day's centre of gravity from the model itself to the application wrapped around it. If the flagship model can slip while the user-facing product still expands sharply, that is likely to mean the practical differentiator is becoming orchestration, tools and workflow rather than one more raw model release.

Amazon Web Services said xAI's Grok 4.3 is now available through Bedrock, adding it to Bedrock's third-party model lineup for chat, tool calling, structured outputs, image input and multi-turn conversations. xAI separately launched Grok 4.5 as a new model for coding, agentic tasks and knowledge work, with limited-time free usage through Grok Build and Cursor. Apple, meanwhile, secured approval to launch Apple Intelligence in China and said it will integrate Alibaba's Qwen models and Baidu's AI tools for Chinese users across its operating systems.

Taken together, these moves suggest the model market is becoming easier to redistribute than to own outright. That would likely matter to Microsoft because the same pressure runs through Azure AI and Copilot: when cloud platforms can list outside models and device makers can swap in local partners by country, the advantage moves away from exclusive access and towards packaging, distribution and trust.