Apple to skip high-end M6 Mac chips in favor of AI-focused M7 line

Original Article ↗ Hacker News Discussion ↗

Scope of the M6/M7 Change

  • Article is interpreted as: Apple will ship only low-end M6 variants (e.g., base chips), skipping M6 Pro/Max/Ultra and jumping high-end directly to M7 Pro/Max/Ultra in 2027–2028.
  • Several commenters note this is similar to Apple skipping “Ultra” variants in some past generations, but broader this time.
  • Some see the naming as a marketing ploy; others argue it simply reflects real architectural generations and internal roadmap shifts.

Impact on Mac Lineup and Timing

  • Likely no M6 Pro/Max MacBook Pro; some expect no major MBP redesign (OLED, thinner) until M7-era, disappointing people on M1 Pro/Max waiting to upgrade.
  • Others note Apple could still ship a redesigned MBP with base M6, but marketing optics would be awkward if it’s slower than older high-end models.
  • Mac Studio: expectation of an M5 Ultra version later this year; Mac mini “Pro” seen as neglected after skipped Pro variants.
  • Overall product matrix (Air, Pro, Mini, Studio, iMac) is viewed as increasingly confusing across chip generations.

Memory Bandwidth, Capacity, and Local AI

  • Discussed bandwidth targets: base M7 around 240 GB/s; extrapolated M7 Ultra could exceed 1 TB/s, making local LLM inference more compelling.
  • Enthusiasts anticipate 192–512+ GB unified memory Macs as powerful local AI boxes; skeptics highlight extreme RAM costs and note Apple already killed 256/512 GB M3 Ultra configs.
  • Debate over whether very high-RAM Macs (e.g., 768 GB–1 TB) make economic sense vs using the same DRAM for tens of iPhones.

DRAM Shortage, Pricing, and Apple Margins

  • Many see this strategy as partly driven by DRAM scarcity and high prices, not just “AI focus.”
  • Argument: with limited RAM supply, Apple maximizes profit by prioritizing iPhones and lower-RAM Macs; high-RAM desktops would need huge price tags to match iPhone margins.
  • Some expect RAM prices to normalize in a few years; others assume elevated pricing for longer.

Apple’s AI Strategy and Competitive Position

  • Some claim Apple is “late to the AI party” and will be outclassed by Nvidia GPUs plus used high-end cards.
  • Counterpoint: Apple is pursuing a different goal—mass-market, low-power, on-device AI rather than hyperscale training, which may align better with privacy and offline use.
  • Several predict local AI will become mainstream and see Apple’s vertical integration and unified memory as a strong long-term position.

Local vs Cloud Models

  • Thread explores whether local models can replace or just complement frontier cloud models.
  • Optimists think consumer hardware plus model/quantization advances will handle most everyday tasks locally.
  • Skeptics argue data-center accelerators will remain far ahead for large, long-context, or high-accuracy workloads and that cloud inference may still be cheaper when electricity and hardware costs are included.

Foundry and Packaging Rumors

  • Rumors cited that M7 might use Intel’s 18A process; viewed as risky but plausible given US policy pressure and TSMC capacity constraints.
  • Discussion of Apple moving to chiplet-based designs and advanced packaging (InFO, CoWoS-style techniques); some confusion over blog-level claims vs realistic DRAM/process integration.