Here comes new Siri again

Apple has announced a redesigned Siri at roughly every other WWDC for the past several years, and each cycle has produced a version that underdelivered on reasoning and conversational depth relative to its announcement framing. The question worth asking is not whether this Siri looks better in a demo, but whether Apple has actually solved the infrastructure problem: running capable LLM inference on-device at acceptable latency without the cloud fallback that quietly powers most of its competitors.

The on-device inference angle connects directly to coverage from early June, where Nvidia's RTX Spark pitch positioned local AI agents as newly practical on Windows hardware, and where a separate piece framed Nvidia's entry as a direct challenge to Apple's M-series advantage in edge inference. Apple's privacy-first positioning only holds as a differentiator if its silicon can match the throughput competitors are now promising. Meanwhile, Microsoft's Build announcements the same week signal that every major platform vendor is racing to claim the same 'AI-native assistant' territory simultaneously, which compresses the window Apple has to establish Siri as meaningfully distinct.

Watch whether Apple publishes reproducible benchmark results for on-device reasoning tasks within 60 days of WWDC. If third-party testers find that complex queries still route silently to cloud inference, the privacy narrative collapses and the competitive differentiation story goes with it.