News
Nvidia LPU shipments projected to grow tenfold by 2027 with new 256-unit rack architecture
Monday, March 16, 2026 at 06:35 AM
A new industry survey indicates Nvidia is significantly revising its LPU shipment forecasts upward following its investment in Groq. Shipments for 2026-2027 are projected at 4-5 million units, representing a tenfold increase. To handle low-latency inference and expanded KV cache for long-context windows, Nvidia plans to increase LPU density per rack from 64 to 256 units. Production for these new racks is expected between Q4 2026 and Q1 2027. WUS Printed Circuit is identified as a key PCB supplier, marking the first large-scale adoption of CCL M9 materials and advanced high-layer board processing for this architecture.
Context
Supply chain analyst Ming-Chi Kuo reports that Nvidia is significantly upwardly revising its LPU (Language Processing Unit) shipment forecasts following its $20 billion acquisition of Groq assets. Kuo projects total shipments of 4 million to 5 million units across 2026–2027, representing a tenfold increase over previous annual volumes. This surge is driven by the rise of AI agents and real-time applications that require the ultra-low latency provided by LPU architectures, which Nvidia is now integrating into its CUDA and NIM software ecosystems to lower deployment barriers.
To meet increasing memory demands for long-context inference, Nvidia is shifting from 64-unit to 256-unit rack architectures (LPX), with mass production expected by Q4 2026. This transition initiates a major growth cycle for the PCB supply chain, specifically benefiting WUS Printed Circuit. The new racks are the first to undergo large-scale adoption of CCL M9 materials, with WUS also beginning early testing on next-generation M10 materials for the upcoming Rubin Ultra and Feynman platforms slated for 2027.
Sources (10)
Groq and Nvidia Enter Non-Exclusive Inference Technology Licensing Agreement to Accelerate AI Inference at Global Scale | Groq is fast, low cost inference.Nvidia buying AI chip startup Groq for about $20 billion, biggest dealChina’s Moore Threads rockets 400% in IPO: Is this the next DeepSeek to challenge Nvidia? - CapacityBlog | Ben Pouladian — AI & Semiconductor Investment Research – Ben Pouladian's BlogNVIDIA GTC 2025 Preview: B300 Chip Revolutionizes AI ... - Lemon8Nvidia to buy assets from Groq for $20B cash | Hacker News华泰| 再谈NV的下一个Mellanox：Groq LPU的整合_搜狐网华泰 | 再谈NV的下一个Mellanox：Groq LPU的整合|英伟达_新浪财经_新浪网
Related Companies
Nvidia
NVDA