Overview
- Multiple outlets reported Wednesday that Apple has privately spoken with bankers and approached semiconductor startups to gauge interest in acquisitions aimed at boosting its server AI capabilities.
- Apple turned to Nvidia GPUs hosted in Google Cloud after engineers found M2 Ultra–based internal servers could not run large Gemini models at scale while rebuilding Siri.
- An in‑house server chip project known as Baltra has slipped past its expected 2026 debut, and Apple is developing interim M5 Ultra work with an M7 Ultra–class server chip not likely before 2029.
- Apple signaled a greater willingness to spend on AI deals after buying Israeli startup Q.ai for about $2 billion earlier this year and its CFO dropped a formal net‑cash‑neutral target.
- Alongside buys, Apple is exploring model‑shrinking technology such as talks with PrismML to run bigger models on iPhones, a two‑track strategy that could cut cloud costs and change how features like Siri are delivered.