November 2025 Summaries

Silicon photonics and co-packaged optics (CPO) are revolutionizing AI compute networks by addressing the growing demands for performance, scale, reliability, and efficiency in AI training and inference clusters. With the increasing scale of AI models and datasets, traditional GPU interconnect networks are being redesigned to improve energy efficiency and reduce signal loss. CPO technology, used in NVIDIA's Quantum-X Photonics InfiniBand and Spectrum-X Photonics Ethernet switches, integrates optical components directly with the switch ASIC, enhancing power efficiency, reliability, and latency while simplifying deployment and reducing potential failure points. Lambda plans to leverage CPO networking in its next-generation GPU clusters, such as the NVIDIA GB300 NVL72 and NVIDIA Vera Rubin NVL144, to provide high-bandwidth, low-latency, and reliable interconnects, streamline deployment, and improve overall cluster efficiency. This shift is crucial for managing the expanding scale and throughput demands of AI workloads, as it ensures that Lambda's infrastructure can support large training runs and distributed inference effectively.

Lambda has secured over $1.5 billion in Series E funding, led by TWG Global and the US Innovative Technology Fund (USIT), to advance its vision of building gigawatt-scale AI factories and supercomputers. This investment aims to address the increasing demand for compute power from hyperscalers, enterprises, and frontier labs focused on developing superintelligence. Founded by machine learning engineers, Lambda seeks to make AI compute as accessible as electricity, advocating a vision of "one person, one GPU." The company's infrastructure, known as the Superintelligence Cloud, is designed to support large-scale AI training and inference, positioning Lambda as a key player in the energy-to-cognition pipeline. Key figures like Thomas Tull of TWG Global and USIT express confidence in Lambda's ability to tackle the significant infrastructure challenges associated with the growing need for AI capabilities.

Prime Data Centers and Lambda have partnered to deploy AI-optimized infrastructure at the LAX01 data center in Vernon, California, designed to support large-scale AI training and inference. This collaboration combines Prime's advanced data center designs with Lambda's AI cloud platform, integrating NVIDIA Blackwell GPUs and robust power solutions to accommodate demanding AI applications across industries such as healthcare, finance, and entertainment. The LAX01 facility, offering 33 MW of critical power within 242,000 square feet, is the first AI-ready data center in the city, marking a significant step in addressing the growing compute demands of modern AI workloads. By focusing on efficient, scalable, and flexible infrastructure, the partnership aims to enable enterprises to build and deploy complex AI models quickly, propelling Southern California as a hub for AI innovation. This strategic collaboration highlights a shift in the industry towards AI-specific infrastructure, as traditional data centers struggle to meet the rapidly evolving needs of AI development.

Lambda's MLPerf Training v5.1 results showcase a significant performance improvement with their NVIDIA GB300 NVL72 cluster, which outperforms previous iterations like the GB200 by 27%. This advancement is attributed to the NVIDIA Blackwell Ultra architecture and enhancements in the software stack, including updates to the NVIDIA driver and CUDA, which collectively contribute to a faster training speed. The new MLPerf version replaces older models with modern workloads such as Meta's Llama 3.1 8B, aligning with the demands of contemporary AI training. Lambda's system demonstrated a 1.6× speedup over the top 64× NVIDIA B200 system and a 1.27× improvement over the best NVIDIA GB200 NVL72 submission, establishing the NVIDIA GB300 NVL72 as a leading platform for large-scale AI training. Additionally, evaluations showed a 13% average speedup using NVFP4 precision over FP8, indicating further potential for reducing training times. This positions Lambda's platform as a versatile choice for enterprise AI teams seeking a balance of performance, cost efficiency, and control.

Lambda has announced a multibillion-dollar partnership with Microsoft to deploy AI infrastructure powered by tens of thousands of NVIDIA GPUs, including NVIDIA GB300 NVL72 systems, under a multi-year contract. This strategic collaboration aims to expand access to cloud-based accelerated computing resources and marks a significant step in the long-standing relationship between the two companies, reflecting the growing global demand for high-performance computing driven by the rise in AI assistant use and enterprise adoption. Lambda, known as The Superintelligence Cloud, focuses on building gigawatt-scale AI factories for training and inference, aspiring to make compute as ubiquitous as electricity. This partnership underscores Lambda's role as a trusted partner in deploying compute infrastructure for leading AI products and highlights a major technological shift in human history.