Nvidia Powers Apple AI: Nemotron 3 Ultra Reshapes Stack

Apple will rely on Google's fleet of Nvidia chips to power its overhauled Siri when it launches in September. This means queries needing cloud processing will use Google's large Gemini models, under an existing agreement. Apple will specifically use Google's Nvidia Blackwell B200 data center chips. User data will be encrypted with Nvidia's hardware-based confidential compute feature. Apple reportedly found running a version of Google Gemini under its own Private Cloud Compute too slow for consumers. What's interesting is Nvidia recently unveiled its most powerful AI model, Nemotron 3 Ultra, at Computex. This open-source model has approximately 500 to 550 billion parameters and is designed for advanced reasoning. It's part of a three-tier family, offering developers flexible options. Nvidia says it delivers up to five times faster inference and can reduce costs for complex tasks by 30%. The broader Nemotron 3 family has seen over 50 million downloads. Confidential compute, a security feature in Nvidia GPUs, encrypts data and AI models during cloud processing. While it slightly slows AI queries, it helps Apple maintain user privacy. This move by Apple marks a divergence from its usual strategy of controlling all product ingredients. This development positions Nvidia as a fundamental infrastructure provider for a major consumer AI launch.