OpenAI Buys Tens of Thousands of Macs to Train Agents, Boosting Apple's AI Hardware Role

According to The Information, citing OpenAI insiders, OpenAI has bought tens of thousands of Mac minis and Mac Studios over the past months for reinforcement learning and computer-use agent tasks, pressing Apple for more supply; Anthropic rents Mac minis via AWS for similar work. The buying spree shows up in Apple's financials—Mac quarterly revenue hit $10.3 billion, up 29%—and has drained high-memory configurations from retail channels. Reddit users still debate the exact use cases, with no firm conclusion beyond the reported agent-training purpose.

Confirmed

• Multiple posts consistently relay The Information's report: OpenAI bought tens of thousands of Mac minis and Mac Studios, leveraging native macOS environments for task execution and trajectory collection for RL training of computer-use agents (@APPSO, @机器之心, @量子位)
• Anthropic rents Mac minis through AWS for related computer-use agent work
• The demand is visible in Apple's results: @rohanpaulai and @Hesamation note Mac quarterly revenue reached $10.3 billion, up 29% year-over-year
• Startups are building Mac-based data centers (per @Hesamation)
• Channel spillover: @kimmonismus notes the buying caused Mac mini shortages; @TheZachMueller adds high-memory configurations are fully delisted with extended shipping for base models
• @eyishazyer notes these machines are not for conventional GPU pretraining but specifically for training agents that autonomously operate computers

Unconfirmed

• Exact quantities, spend, and any special supply arrangements between OpenAI and Apple come from anonymous sourcing; no official response
• Reddit user @FormalAd7367 asked about potential use cases; community guesses include edge computing, environment-specific testing, and dev compatibility—no definitive answer

Why it matters

• With GPU compute scarce, Apple Silicon devices have become an alternative compute source for AI labs due to cost-effectiveness and supply; @kimmonismus argues this worsens the RL compute crunch, while @BLUECOW9 says AI labs are consuming all available compute
• @新智元 and @创业邦 note that despite lower raw compute than Nvidia, Apple's unified memory architecture (UMA) is key for large-memory RL training and local inference—the technical reason Macs are favored
• The spillover into Apple's financials and retail channels could reshape the view that Apple is a weak player in the AI race

2026-08-30 ~ 2026-09-01 · 19 related posts

• Apple's Mac Biz Up 29%; OpenAI/Anthropic Buy Minis for AI Compute — Hesamation · 2026-08-30
• Report: OpenAI bought tens of thousands of Mac minis for agent training — ZeroStateReflex · 2026-08-31
• OpenAI Bought Tens of Thousands of Mac minis for Agents as Unified Memory Wins — APPSO · 2026-08-31
• [source] OpenAI reportedly buying thousands of Macs to train computer-use agents — rohanpaul_ai · 2026-08-31
• Report: OpenAI buying tens of thousands of Mac minis and Studios — ZeYanjie · 2026-08-31
• Discussion on OpenAI's Bulk Purchase of Mac Minis and Potential Use Cases — FormalAd7367 · 2026-09-01

13 near-duplicate retellings: kimmonismus · 量子位 · 新智元 · rohanpaul_ai · 机器之心 · BLUECOW009 · 创业邦 · SumitGup · eyishazyer · BLUECOW009 · rohanpaul_ai · TheZachMueller · 创业邦