What runs underneath?

Compute & infrastructure

Silicon, inference systems, AI factories and the physical constraints on intelligence.

14 research briefs 84 distinct sources

Research briefs

NVL72 rack-scale systems, 4-bit floating point (FP4) Tensor Cores, and 5th-gen NVLink interconnects

Groq LPUs, deterministic tensor streaming, SRAM-first architectures, and ultra-high-speed inference

4-trillion transistor monoliths, 44GB on-wafer SRAM, 21 PB/s bandwidth, and cluster-scale supercomputing

100MW–1GW datacenter topologies, liquid cooling CDUs, high-voltage power distribution, and PUE optimization

InfiniBand Quantum-X, RoCEv2 (Ultra Ethernet), non-blocking fat-tree topologies, and Optical Circuit Switches

Bare-metal GPU nodes, non-blocking RoCEv2 fabrics, distributed data sovereignty, and multi-cloud interconnects

Small Modular Reactors (SMRs), geothermal, grid queues, carbon-free baseload, and gigawatt power purchase agreements

Apple Neural Engine, Qualcomm Snapdragon X, Intel Core Ultra, and 4-bit edge inference runtimes

Wormhole, Blackhole, open-source ISA architectures, chiplet scaling, and decoupling AI from closed hardware ecosystems

Hardwired transformer architectures, zero general-purpose overhead, and 10x throughput per dollar

HNSW, DiskANN, IVF-PQ, GPU-accelerated similarity search, and hybrid vector-relational engines

vLLM, TensorRT-LLM, SGLang, PagedAttention, continuous batching, chunked prefill, and speculative decoding

Trusted Execution Environments (TEEs), NVIDIA Hopper/Blackwell CC, remote attestation, and data clean rooms

GPUDirect Storage (GDS), NVMe-over-Fabrics, high-throughput checkpointing, and parallel filesystems (Lustre, WEKA, VAST)