TUE 13 OCT 2026 · 1 MIN READ
Quiet morning: one open rack standard and one new frontier model.
NVIDIA opened Vera Rubin rack designs to OCP, with racks up to 250kW. Meta released Muse Wonder at $20 per million output tokens. Other sections had nothing sourced inside the window.
General AI Infrastructure
NVIDIA opens Vera Rubin racks to OCP
TakeawayNVIDIA contributed Vera Rubin rack and compute tray designs to the Open Compute Project through MGX 1.2.
- MGX 1.2 adds blind-mate liquid cooling and a liquid-cooled 5,000A busbar.
- Racks scale up to 250kW.
- NVIDIA, AMD, Google and Meta published Open Data Center Specification v0.7.
AngleOpen rack specs let operators source power and cooling gear from more than one vendor. Expect facility designs to converge on liquid cooling and 250kW racks, so check that new builds can carry that density.
Model Releases & Compute-Demand Drivers
Meta releases Muse Wonder frontier models
TakeawayMeta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, its first frontier models.
- API price for Muse Wonder: $4 per million input tokens, $20 per million output tokens.
- Meta says it trained them at its Prometheus data center in central Ohio.
- Benchmark parity with Fable 5.1 and GPT-6 Astra is Meta's own claim; no parameter count or compute figure was disclosed.
AngleA fourth lab selling frontier-class tokens at $20 per million output keeps pressure on inference pricing. Meta gave no training-compute figure, so the demand signal for capacity is the price, not the footprint.
How this is made. Compiled each weekday morning by the Rack Rate desk's research agents from primary sources, with every item linked to the document it came from, and reviewed before it publishes. The Angle on each story is the desk's own read; it does not recommend suppliers, and Rack Rate's own data is labelled as such. Corrections: daily@rackrate.ai.