Apple unveiled the M5 Max MacBook Pro, and I ordered a fully loaded configuration almost immediately: RMB 58,200.
1. A ten-year spending pattern#
Since graduating, my primary machine has always been the same category: a top-tier MacBook Pro, usually maxed out.
- 2015: bought one myself while working at Alibaba.
- 2017: asked for a maxed-out rMBP when joining Tantan.
- 2019: used Apple employee pricing before leaving.
- 2022: bought a fully loaded M1 Max after starting my own company.
- 2026: now the M5 Max, again fully loaded.
The price keeps climbing, but the logic has never changed. If you work ten-plus hours a day on a machine, saving money on your core production tool is usually false economy.
2. My usage can actually justify it#
This is not pure luxury therapy. I do use the machine hard.
Local model experimentation, big codebases, packaging, infrastructure work, multi-VM workflows, and increasingly AI-heavy tooling all push the machine in real ways.
3. What is actually better on M5 Max?#
For me, the meaningful deltas are not just synthetic benchmarks.
Wireless catches up#
Wi-Fi 7 and Bluetooth 6 sound minor, but finally let the laptop stop bottlenecking the network around it.
Memory bandwidth and capacity matter more#
Unified memory and memory bandwidth are the real AI-era specs.
CPU and GPU gains still help#
Compiles, packaging, multi-service local runs, and inference all benefit from the steady CPU/GPU climb.
Neural engine and storage are no longer side notes#
AI throughput and SSD speed increasingly show up in real day-to-day workflows.
AI is the big story#
At around 240 TOPS, the M5 Max starts to feel like a genuinely respectable local inference box for smaller and mid-sized models.
4. What I really want: M5 Ultra#
The truly exciting part is not even the M5 Max. It is the implication of what an M5 Ultra-class machine might become.
If Apple pushes unified memory high enough, a single machine could become the first truly meaningful local frontier-model workstation for serious individual use.
5. Why not just wait for M6?#
Because the point is not endless optimization. The point is buying back time and reducing friction now.
AI tooling is moving too quickly for me to enjoy “waiting one more generation” while sitting on a machine I am already pushing to its limits.
There is also a more interesting long-term direction here:
- iPhone-level devices can already run small models.
- MacBooks can comfortably host mid-sized local models.
- Mac Studio and cluster-scale Apple setups may become serious on-prem inference platforms.
The line between a personal laptop and an AI workstation is getting very thin.
