Apple has been promising a revolution in its voice assistant for years, but reality falls far short of expectations. Technical issues with the new Siri, based on large language models (LLM), have forced the company to acknowledge significant delays. We analyze what is going wrong, how long it will take, and why Apple has had to turn to Google to try to resolve the situation. The question of whether Apple is still setting the pace for technological innovation is no longer just rhetorical. For the third consecutive time, the company has admitted that its big bet on artificial intelligence is not yet ready. Craig Federighi, Apple's software chief, acknowledged in an interview with the Wall Street Journal: “We had functional prototypes, but we couldn't make them reliable as quickly as we hoped.” A striking statement for one of the world's most valuable companies. The development of the new Siri began in 2022 and was presented at the WWDC in June 2024 with a demonstration that, in retrospect, turned out to be largely simulated. Apple promised its launch for early 2025 but did not deliver. It later set March 2026 alongside iOS 26.4 as a new date, but that also did not happen. Engineers testing internal versions of iOS 26.4 and 26.5 report that Siri still struggles with complex queries, responds slowly, and in many cases, defers answers to ChatGPT. For a company that has always prided itself on controlling the entire user experience, this represents a serious setback. The main problem is not economic but technical. Apple wants Siri to process data on the device itself to protect privacy, but LLMs require computational power that current iPhone chips cannot consistently guarantee. Siri's architecture has been divided into two layers: an old one for basic functions and a new one for AI. This coexistence generates failures: interruptions if the user speaks quickly, errors in summaries, and issues with third-party apps. A team of hundreds of analysts in Texas, Spain, and Ireland manually reviews the results, highlighting how far total automation is from being achieved. At the end of 2025, Apple signed an agreement with Google to integrate the Gemini models as the foundation for the new Siri, paying nearly $1 billion annually. Paradoxically, the company that has maintained a historic rivalry with Google now depends on its technology to compete in artificial intelligence. This turn confirms that Apple has lost the first battle of generative AI. While OpenAI, Google, and Meta launched conversational assistants in 2023 and 2024, Apple chose to develop its own architecture, accumulating delays and losing trust among analysts and users. Some experts believe that Apple could present "Project Campos" at the WWDC in June 2026, a second phase of Apple Intelligence that would transform Siri into a more advanced chatbot. The first features would arrive with iOS 26.5 in May, but evaluators warn that performance remains inconsistent. The advice for users is clear: do not buy a new device expecting the new Siri. If you have a well-functioning iPhone 15 or 16, wait at least until fall 2026 before upgrading. Apple will eventually release a robust version, but the gap with Google in AI is real and will take time to close.