Microsoft's Maia 300: 2x chips per rack, 2.5x memory bandwidth, up to 7x throughput from 2027
Beth_Kindig · x · 2026-10-30
Analyst Beth Kindig shared specs for Microsoft's next-gen inference chip Maia 300 (from IO Fund's quarterly report; figures are Microsoft disclosures, not independently tested):
- 2x rack density: 96 chips per rack vs 48 in Maia 200
- Power: 167 kW per rack (liquid-cooled), 2.6x Maia 200's 64 kW
- Memory: 288 GB HBM4 with 8 stacks per chip (+33% vs 216 GB HBM3e); bandwidth 18 TB/s, up 2.5x
- Throughput: claimed up to 7x rack-scale tokens vs Maia 200, +40% tokens per megawatt vs "leading GPU-based systems"
- Process & rollout: TSMC 2nm-class N2P with separate compute, I/O and memory dies; production deployments from 2027, targeting long-context reasoning and agentic workloads
More from Infra
- Balanced-ternary K5 inference beats TorchAO FP8 by 3.2× on NVIDIA L4 — Code_MasterCody · 2026-10-30
- Is a $420 Xeon X99 box with 256GB RAM worth it for LLM inference? — Ammoryyy · 2026-10-30
- Meta loses a node every 5 hours; topology-aware scheduling recovers 27% more GPUs — skypilot_org · 2026-10-30
- llmman: run Claude Code, Codex and other agents on any local model with one command — weskambale · 2026-10-30
- How Instagram cut feed load time by 50% and Discord stores 100B+ messages — _jaydeepkarale · 2026-10-30
- YC-backed VirajAero launches turbine that powers datacenters with 40% more power from same fuel — Scobleizer · 2026-10-30