Apple Preps 'Core AI' to Replace Core ML
Apple is reportedly set to replace its long-standing Core ML framework with a new 'Core AI' in iOS 27, a move signaling a major shift toward on-device generative AI and agentic workflows. The new framework, detailed by 9to5Mac, promises better model orchestration and memory sharing. The change coincides with the launch of the new $599 iPhone 17e, which features the A19 chip with advanced on-chip ML accelerators.
The move from Core ML originates with hardware decisions made nearly a decade ago. The first Apple Neural Engine (ANE) debuted in the 2017 A11 Bionic chip, a dedicated processor for AI that powered features like Face ID by executing 600 billion operations per second. This established Apple's long-term strategy of tight integration between custom silicon and machine learning frameworks. That hardware capability has scaled exponentially, creating the performance headroom for more advanced AI. The ANE in the A12 Bionic jumped to 8 cores and 5 trillion operations per second, while the A18 Pro's 16-core Neural Engine hits 35 trillion operations per second, running Apple Intelligence features up to 15% faster than the prior generation. This relentless performance curve is fundamental to running generative models locally. Core ML was built to harness this power, allowing developers to run models trained in libraries like PyTorch or TensorFlow directly on-device. This approach prioritizes privacy and responsiveness by avoiding network calls. Recent updates to Core ML have focused on advanced compression techniques to shrink large language models (LLMs) and diffusion models for efficient execution on Apple silicon. The shift to 'Core AI' signals a strategic repositioning beyond a tool for developers toward a platform for generative AI and autonomous agents. It follows the introduction of the Foundation Models framework in iOS 26, which provided direct API access to Apple's own pre-trained LLMs, distinct from Core ML's function as an engine for running other models. This new framework's emphasis on model orchestration is a direct response to the complexity of managing multiple AI models on a single device. It necessitates deep hardware-software co-design, influencing everything from the A19's ML accelerators to how iOS 27 will manage memory and schedule tasks between the Neural Engine, GPU, and CPU. The on-device AI strategy also mirrors Apple's operational playbook. The company uses AI and predictive analytics to manage its complex global supply chain, from demand forecasting to optimizing logistics. Owning the entire stack, from the 'Core AI' framework down to the custom silicon and its manufacturing, creates a powerful feedback loop for innovation and operational control.
