Reports
AI-generated structured vendor updates
OpenAI and Broadcom Launch Jalapeño ASIC for LLM Inference, 9-Month Tapeout
OpenAI and Broadcom unveil Jalapeño, a custom ASIC for LLM inference, achieving tapeout in 9 months. The chip reduces data movement and claims superior performance per watt. Deployment planned by end of 2026, marking OpenAI's shift to integrated hardware-software infrastructure.
ReflectionAI Secures $6.3B SpaceX Compute Deal, Open-Source AI Breaks Hardware Lock-in
Open-source AI startup ReflectionAI signs a $6.3B deal with SpaceXAI to lease NVIDIA GB300 compute at Colossus 2 for training open-weight frontier models. This gives open-source labs parity with closed-source giants but creates deep dependency on NVIDIA's proprietary hardware.
NVIDIA Acquires Groq LPU: Inference Architecture Shift from HBM to On-Chip SRAM
NVIDIA signs ~$20B licensing deal with Groq for LPU tech, featuring 230MB on-chip SRAM at 80TB/s bandwidth. This targets Transformer inference decode, replacing HBM bottlenecks with ultra-low latency on-chip storage, potentially reshaping the AI inference chip landscape.
NVIDIA Tops Data Center Ethernet Market: GPU Compute Dictates Network Architecture
IDC reports NVIDIA captured 21.5% of the data center Ethernet switch market in Q1 2026, with $2.1B revenue. This milestone, driven by the Spectrum-X platform using RoCE and NVLink, marks a control shift where GPU compute dictates network architecture, directly challenging Cisco and Arista.
SK Hynix HBM4E Samples: 3nm Logic, 384GB/GPU, Igniting AI Memory Bandwidth Arms Race
SK Hynix has sampled its 12-layer HBM4E, featuring TSMC 3nm logic die and enhanced per-pin bandwidth, targeting Nvidia Rubin Ultra with 384GB per GPU. This marks the start of a sprint with Samsung in next-gen AI memory, where HBM BOM share has surged to 65-70%.
NVIDIA Absorbs Groq LPU: Feynman GPU to Integrate SRAM Inference Tile, Hybrid Architecture by 2028
NVIDIA secures Groq's LPU inference technology via a non-exclusive license and key hires, planning to integrate large SRAM tiles into its 2028 Feynman GPU using TSMC SoIC hybrid bonding. This enables deterministic scheduling and 80TB/s on-chip bandwidth, shifting NVIDIA from a pure GPU vendor to a hybrid inference/training platform.
Z.ai GLM-5.2 Open-Source: 744B MoE, 1M Context, MIT License as Geopolitical Shield
Z.ai releases GLM-5.2: 744B MoE with 40B activated parameters, 1M input and 131K output context, under MIT license. Released one day after Anthropic Fable 5's government takedown, it offers a downloadable, unbanable alternative with Anthropic API compatibility for zero-code migration, giving enterprises a sovereign AI option.
SGLang 0.5.13 Delivers 25x MoE Inference Speedup via Predictive Routing and Sparse KV Cache
SGLang 0.5.13 introduces two-stage MoE routing prediction and sparse KV cache, achieving a 25x inference speedup on NVIDIA GB300 NVL72. Benchmarks on A100 show 65% throughput gain, 40% latency reduction, and 62% lower routing overhead. This optimization directly attacks the core bottleneck of MoE inference, potentially reshaping AI inference economics.
Google TurboQuant: 6x KV Cache Compression, AI Inference Memory Cost Inflection Point
Google releases TurboQuant, a two-stage KV cache compression algorithm (PolarQuant + QJL) achieving 6x memory reduction (3-bit quantization) and 8x attention speedup with no measurable accuracy loss. The announcement triggered a sell-off in memory stocks (Micron -3%, Western Digital -4.7%), signaling a potential structural shift in AI inference memory demand.