OEM channel check: Vera Rubin slips to Q1 2027, neocloud MOUs look shaky
tengyanAI · x · 2026-11-10
tengyanAI shares the key takeaways from a paid Wukong Substack channel check with a global AI-server OEM expert:
- GB300: the OEM hit 700-rack GB300 NVL72 capacity in Q3 2026, delivered 600 racks, 90% to CSPs, largely on schedule.
- Vera Rubin delays: December 2026 output (<100 racks) shifts to a Q1 2027 volume ramp, held back by data-center power scheduling bottlenecks and sub-target HBM4 yields; this OEM projects 5,000-6,000 racks in 2027 at $8M per rack.
- Power queues: connection scheduling backlogs delay deployments by 1-3 months across all buyer tiers.
- Neocloud risk: emerging GPU-rental operators' MOU volume grew <50% QoQ; only 30-40% of Phase-1 MOUs convert to binding POs, with the rest facing financing and unproven-unit-economics risk. Many lease commodity capacity, creating price-war risk, while hyperscalers self-fund with high conversion rates.
Related event: Memory Price Surge Hits Demand as OEM Channel Checks Signal Cycle Peak(2 posts)→
More from Venture
- Teams will track cost per resolved task once agent bills sit next to payroll — LadyAshBorg · 2026-11-10
- Ex-Leica survey operator: aerial mapping and flight planning software is ripe for AI rebuild — RileyRalmuto · 2026-11-10
- Anthropic targets Nov. 19 Nasdaq debut under ticker ANT, investors push valuation below $1.5T — josh_bickett · 2026-11-10
- Anthropic retracts botched IPO prospectus after leak to Reuters; listing set for tomorrow — GaryMarcus · 2026-11-10
- AI capex needs to grow 5-10x before oversupply, argues Rob LeClerc — robleclerc · 2026-11-10
- Morph launches Ultra: unlimited Kimi and DeepSeek for $200/month — zack_overflow · 2026-11-10