MBW 1044: Live, Love, Re-Org - The Strength of the M5 Mac Studio Running Local Agents
Summary
MacBreak Weekly features Federico Vittici, editor-in-chief of Mac Stories, who reviews the M5 Ultra Mac Studio for local AI performance. He details his setup, including running a 24/7 agent in a storage unit with a 100-meter Ethernet run, and benchmarks the new hardware against the M3 Ultra. The hosts discuss the rapid evolution of MLX backends, the cost-benefit of local AI versus cloud subscriptions, and the potential of clustering Macs via Thunderbolt 5.
Topics discussed
Intro: Federico Viticci joins to discuss M5 Ultra
Podcast banter and the M5 Ultra review unit
Running local AI 24/7 and cooling challenges
Running Ethernet cable to a storage unit
Building a testing harness with Codex
M5 Ultra performance: pre-fill and token speed
Cost analysis: M5 Ultra vs. API subscriptions
Who should buy the M5 Ultra? Business cases
The AI hardware upgrade cycle and addiction
OpenAI Dev Days and Apple's AI strategy
John Ternus reorg and Apple layoffs
Apple's AI lag and the rise of Meta Muse
AI-powered devices: Perplexity and ESP32
Linux on Apple Silicon and AI agents
AI agents managing networks and security
T-Mobile eSIM and Apple Watch cellular
iOS 26.0.1 crash bug and baseband updates
Sponsor: CashFly CDN and show wrap-up
Apple Vision Pro vs. Meta Quest and glasses
Rome food talk and guest pick intro
OpenMinis app and local AI benchmarks
Mac app picks: FruitMenu and keyboards
Upcoming pre-orders and final credits
Listen ad-free on Castria