Lambda

Lambda has raised over $1.5B in equity to build superintelligence cloud infrastructure. TWG Global, USIT, and existing investors led the Series E round to position Lambda to execute on its mission to give everyone the power of superintelligence. One person, One GPU. Press release: bit.ly/3X3lOOb

Full white paper with reproducible recipes to optimize MFU: lambda.ai/hubfs/4.%20Res…

We tested NVIDIA HGX B200 and GB300 NVL72 with TorchTitan and reproducible config changes: - Llama 3.1 8B on 8x B200: 55% MFU - Llama 3.1 70B on 16x B200: 50% - Llama 3.1 405B on 128 GPUs (GB300): 53%. At 16k–32k sequence, MFU climbs toward 60%.

The xAI "low utilization" story has people mixing up two different metrics. Fleet utilization tells how many GPUs are running. Model FLOPS Utilization (MFU) how much compute each running GPU is actually capturing. Both matter, but they're not the same.

@NagaSaiAbhinay @mulamx Thank you for the shout-out!

@mulamx Try reaching out to @LambdaAPI ! form.typeform.com/to/Wn8YWMIj

I need some credits for GPU. Anyone?

Read the build log: lambda.ai/blog/creating-…

Open-sourced on @huggingface. Downloaded 8.2k times last month.

Agent harnesses reach the many when the models inside them are efficient. Our goal: compress frontier-grade skills into a footprint small enough to run unquantized on lightweight compute.

Register here: web.cvent.com/hub/events/a8a…

The shift to AI factories needs strong infrastructure and strong leadership. Tomorrow in Los Gatos, Lambda joins @nvidia and @aligneddc to discuss women’s leadership, talent trends, and the future of AI infrastructure.

Read the breakdown: lambda.ai/blog/flashatte…

FlashAttention-4 on NVIDIA Blackwell: 1,613 TFLOPs/s, 71% hardware utilization, and up to 2.7× faster than Triton.

Same GPU. More tokens per second. Lower cost per token. The most optimized open-source attention kernel yet.

@vbingliu @scale_AI @Chi_Wang_ @uiuc_aisecure @VirtueAI_co @UofIllinois Wonderful talk by @letscatanate (@LambdaAPI) at #ICLR 2026 “Agents in the Wild: Safety, Security, and Beyond” workshop! 🚀 Devina is now presenting “Securing Agents in the Wild” — great insights on advancing security for real-world agent systems.

If you need some compute call the guys at @LambdaAPI nice marketing @stephenbalaban

.@deepseek_ai v4 Pro's checkpoint is both in FP4 and FP8, depending on the layer. This means that the entire model can fit on a single NVIDIA 8xB200 node without trouble. @vllm_project: "Checkpoint is FP4+FP8 mixed: MoE expert weights are stored in FP4 while the remaining (attention / norm / router) params stay in FP8."

Read the full breakdown: lambda.ai/blog/iclr-2026…

A few highlights: A 7B agent that outperforms GPT-4o, not by scaling up, but by training smarter inside its own loop. Lossless weight compression that speeds up inference by 177%. And a public arena where 23 teams ran 103,000 adversarial battles to stress-test LLM security.

We're at @iclr_conf #ICLR2026 this year with 12 papers and two workshops. Over 20 collaborators across @Stanford, @CarnegieMellon, @BerkeleyRDI, @Google, @nvidia, and @Microsoft helped make this work happen.