Lambda

Can you get JAX-styled functional training without leaving CUDA and Python? We benchmarked a 42M-parameter Transformer on TinyShakespeare and scaled from a single GPU to a 24-GPU distributed cluster. The code changes were minimal: just a shift from @jax.jit to @pmap and jax.distributed.

Today, Leonard Speiser joins Lambda as Chief Operating Officer. Leonard co-founded and served as the CEO of Clover, scaling it into one of the fastest-growing platforms in history. Earlier, he held roles at Intuit, eBay, and Yahoo. Leonard brings more than a decade of experience scaling a mission-critical software and hardware platform. Learn more: https://lnkd.in/giV5R4V5

NVIDIA’s Rubin platform is coming to Lambda Cloud! Access Vera Rubin NVL72 across our 1-Click Cluster and private Supercluster products in the second half of 2026. Read more: https://lnkd.in/gMHqbrVe #CES2026 #NVIDIARubin

2025: The race to superintelligence kicked into high gear. Next stop? 2026. See you in the fast lane.

Deploying frontier AI models demands more than a singular GPU. See our latest tutorial on serving a one-trillion-parameter model on 8x NVIDIA Blackwell GPUs with vLLM: https://lnkd.in/edjF2WNm

We tested Lambda's 1-Click Clusters and managed Kubernetes. The cluster interconnect worked out of the box with no extra setup, which was great to see. If you're using Lambda's clusters, we published a guide on orchestrating distributed workloads on Lambda using dstack. https://lnkd.in/emazKiT8

From bigger models to better intelligence: what NeurIPS 2025 tells us about progress. In his latest piece, Chuan Li, Lambda’s Chief Scientific Officer, unpacks where the field is heading: https://lnkd.in/gtRT2edA

Multi-cloud isn’t a strategy so much as an outcome of constraints. GPUs are scarce, DC capacity is tight, and egress fees complicate moving training data. Here’s a blueprint for running training and inference reliably across multiple clouds: https://lnkd.in/gnbfS8xK

Congratulations to NVIDIA on the launch of their open model Nemotron 3 Nano. Fast, simple, and production-ready. Run on Lambda:

#NeurIPS2025: Our biggest year yet. Six accepted research papers, highest workshop attendance, and an incredible booth presence. Thank you to everyone who stopped by and shared your thoughts with us on what’s next in superintelligence. We can’t wait to see what you build in 2026. https://lnkd.in/gzR7ySgH

The ARChitects raise the bar for human-level reasoning benchmarks under some of the toughest compute limits in AI.

Building superintelligence-scale AI factories takes more than just GPUs. Join Maxx Garrison and Johnson Eung in a webinar on Thursday, December 11th, at 10am PST, to learn how Lambda and Supermicro solve the operational problems of modular, GPU-dense AI factories and validate the architecture, testing, and operational practices needed for mission-critical deployments. Register here: https://lnkd.in/encPnAwp

Until next year, #AWSreInvent!

Today, Heather Planishek joins Lambda as Chief Financial Officer. Most recently, she served as Chief Operating and Financial Officer at Tines, the intelligent workflow platform, and has been Lambda’s Audit Chair since July 2025. Heather brings deep company insight to our leadership team, as we accelerate the deployment of AI factories to meet demand from hyperscalers, enterprises, and frontier labs building superintelligence. Learn more: https://lnkd.in/g9YKftp8

AI can draw a dog and a cat side by side, no problem. But ask it for “one golden retriever half-hidden behind another,” and you’ll get a four-legged blob. Why? Because most layout-to-image models are tested on scenes where objects don’t touch. Real photos don’t work that way. In our #NeurIPS2025 paper, OverLayBench: A Benchmark for Layout-to-Image Generation with Dense Overlaps (https://lnkd.in/e-zgWv5S), Jianwen Xie and Xiang Zhang introduce a new way to address it. The OverLayScore metric measures layout difficulty by combining box overlap and object similarity that you can use to build a benchmark full of crowded, overlapping scenes. The fix? Train models with amodal masks so they learn the full shape of an object even when parts of it are hidden. This improves overlap-region accuracy by 16% with zero extra inference cost. #ComputerVision

Day 3 at #NeurIPS: back to our research roots. NeurIPS is one of the few conferences still focused on real academic work, and that’s been Lambda’s home for the past twelve years. We spent the day meeting with founders who turn state-of-the-art research into products and infrastructure for the next wave of AI. “Superintelligence is something where a computer can beat the smartest humans and actually contribute to our scientific field.” The future feels closer than ever.

LLM alignment typically relies on large and expensive reward models. What if a simple metric could replace them? In a new #NeurIPS2025 paper, Lambda’s Amir Zadeh and Chuan Li introduce BLEUBERI, which uses BLEU scores as the reward for instruction following: https://lnkd.in/eV3XHFQz With high-quality synthetic references, BLEU, a surprisingly simple score, matches human preferences at about 74 percent, which is close to the performance of 20B-scale reward models. BLEUBERI-trained models achieve competitive results on MT-Bench, ArenaHard, and WildBench, and they often produce responses that are more factually grounded. This makes alignment significantly cheaper while maintaining strong output quality.

AI can recognize objects, but it still struggles with simple spatial questions like “Is the water bottle on the left or right of the person?” or “Can the robot reach that?” One of our NeurIPS 2025 papers, co-authored by Lambda researcher Jianwen Xie, introduces SpatialReasoner (https://lnkd.in/eedGehAv), a vision-language model that’s equipped with explicit 3D representation and generalized 3D thinking for spatial reasoning. This opens the door to AI that can move through real spaces, assist in homes, collaborate safely with humans, and understand environments the way people do, rather than as flat images. #NeurIPS2025

Day 2 at #NeurIPS2025 was all about builders talking shop. AI research teams stopped by the Lambda booth to trade notes on multimodal inference for superintelligence, building AI factories, and what reliable NVIDIA GB300 performance looks like when workloads hit production. Real conversations with the researchers and engineers pushing the field toward the next level of AI.

At the atomic scale, running millions of simulations on large-scale datasets is expensive. AI helps, but today’s models still spend most of their time performing heavy computations to ensure their predictions remain accurate regardless of how a molecule is rotated, using the Clebsch-Gordan tensor product. In a new NeurIPS 2025 paper, Yuchao Lin and collaborators introduce Tensor Decomposition Networks (https://lnkd.in/eDy5a2Yz), showing how to significantly reduce this symmetry-related bottleneck with a far more efficient method with more than 2x the throughput while still maintaining accuracy. Faster atomic models = faster discovery cycles, from new materials to better pharmaceuticals. #NeurIPS2025

#NeurIPS2025 opened with a full slate of talks and demos asking the hard questions: multimodal reasoning, training at scale, and what it takes to build systems that behave more like software than static models. "We're from the AI community, building for the community. That's why a cloud should exist.” The Lambda booth stayed packed from open to close. Teams stopped by to compare training runs, debate architecture choices, and dig into what “superintelligence-ready infrastructure” actually looks like in production. More to come this week.

Achieve up to 10× inference speed and efficiency on Mixture of Experts models like DeepSeek-R1 with NVIDIA Blackwell NVL72 systems on Lambda’s cloud: purpose-built for AI teams that need fast, efficient, and seamlessly orchestrated infrastructure at scale, and tightly integrated with NVIDIA’s full-stack, co-designed platform. Learn more: https://lnkd.in/eGb6NgEv

If your bottleneck is data rather than compute, you may want to rethink using standard LLMs. In this latest NeurIPS paper co-authored by our very own Amir Zadeh, “Diffusion Beats Autoregressive in Data-Constrained Settings,” we show that masked diffusion models: - Train for hundreds of epochs on the same corpus without overfitting - Achieve lower validation loss and better downstream accuracy than autoregressive models - Exhibit a predictable compute threshold where they reliably pull ahead We trace this advantage to diffusion’s randomized masking objective, which implicitly augments data by exposing the model to many token orderings. Read the paper here: https://lnkd.in/eyUuTeQV

Everyone knows multimodal models can generate text or images… but few talk about what it takes to bridge the two in a way that’s efficient, aligned, and scalable. One of our accepted papers, Bifrost-1 (co-authored by Chuan Li and Amir Zadeh), tackles that problem head-on by creating a blueprint before generating pixel-level details: https://lnkd.in/evTm5fcF Bifrost-1 introduces a new architecture that connects multimodal LLMs to diffusion models through patch-level CLIP latents, enabling tighter cross-modal alignment and more precise control during image generation. You’ll see how the team built a unified latent interface, improved fine-grained semantic grounding, and enabled more stable multimodal generation — all within a single, end-to-end system. #NeurIPS2025 #AI #DeepLearning #MultimodalAI #DiffusionModels #LLMs #Research

How can language models benefit from explicit reasoning steps rather than relying solely on implicit activations? Join us for a deep dive into Latent Thought Models (LTMs): https://lnkd.in/dgnimKQ2 Jianwen Xie walks through how LTMs infer and refine compact latent thought vectors via variational Bayes before generating text. This creates a structured reasoning space and introduces a new scaling axis: inference-time optimization. He’ll also explain why this matters in practice: LTMs show meaningful gains in efficiency and reasoning quality compared to standard LLMs.

Ready to build the future of AI? Join our Multimodal Superintelligence Workshop on building next-generation multimodal models that observe, think, and act across multiple modalities -> https://lnkd.in/d-RUmCG9 Speakers: Amir Zadeh, Chuan Li, Jason Zhang, Jessica Nicholson, Khushboo Goel. What to expect: - A conversation to explore ideas on the current state of multimodal ML science - Deep dives into developing next-generation multimodal superintelligence - Practical discussions on cross-modal reasoning, alignment, fusion, and co-learning #NeurIPS2025 #AIResearch #Multimodal #ML

Lambda is heading to #NeurIPS2025. We're excited to connect with the AI community, share our work, and talk about what’s next in superintelligence. Here are a few ways to find us: • Join us for a Multimodal Superintelligence Workshop: https://lnkd.in/d-RUmCG9 • Hear our talk on Latent Thought Models: https://lnkd.in/dgnimKQ2 • Stop by Booth 713 to say hello and chat with our team. Not attending? Follow daily recap videos featuring interviews, behind-the-scenes moments, and conversations with fellow engineers.

Terray Therapeutics just launched EMMI, an AI platform that’s turning chemistry’s “impossible” into tomorrow’s drug candidates. Grateful to power their breakthroughs with Lambda’s 1-Click Clusters.

How do we bring language models, agents, and world models together to build the next generation of AI systems? At NeurIPS, Lambda is sponsoring LAW 2025: a workshop exploring how future AI can think, plan, simulate, act, and explain itself in dynamic, partially observed physical and social environments. Attend the workshop and meet our amazing lineup of speakers: https://lnkd.in/dwZk39eH

Interconnect performance is as critical as GPU performance in large AI clusters. That's why Lambda is leading early adoption of Co-packaged Optics (CPO) networking across our next-generation clusters, powered by NVIDIA Quantum-X Photonics. CPO reduces latency, improves power efficiency by 3.5×, and increases network reliability by 10× at scale. This means faster training, steadier long runs, and reliable scaling toward 100k+ GPU clusters. Read more: https://lnkd.in/dPDhJVPv

From an under-the-desk hustle to lighting up the Nasdaq Tower. Thank you Nasdaq for the recognition, and to the hundreds of thousands of customers creating AI breakthroughs on Lambda. Now, back to building the superintelligence cloud. Full announcement: https://lnkd.in/dGmGbB35

What’s new in computer vision research from Lambda?

We're excited to announce that Lambda is leading early adoption of NVIDIA's silicon photonics–based networking: https://lnkd.in/e23zbXpu

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: https://bit.ly/3X3lOOb

MLPerf Training v5.1 results for NVIDIA GB300 NVL72 are in: https://lnkd.in/dmgKYz9v NVIDIA Blackwell Ultra + Lambda engineering = 27% increase in LLM training performance

Lambda and Prime Data Centers partner to deploy new cloud AI infrastructure in Southern California. Our LA data center features over 12,000 NVIDIA GPUs already serving Superintelligence customers. Learn more: https://bit.ly/448ZypX

Scaling AI means rethinking the backbone of infrastructure, from power and networking to data center capacity. Lambda’s AI factories are engineered for speed, density, and cooling to maximize intelligence per watt. Join Rebecca Naughton at IgniteGTM alongside industry leaders to discuss what’s powering the next wave of AI infrastructure. Learn more → https://lnkd.in/dfNJtQmk

Scaling AI from prototyping to production can be hard: distributed jobs, reliability, cost. Ray makes it practical. Lambda makes it run. Read more → https://lnkd.in/dguF_w-f Meet us at Anyscale's Ray Summit, booth S2.

Lambda announces a multibillion-dollar, multi-year agreement with Microsoft to deploy mission-critical AI infrastructure. Press release: https://bit.ly/4i02CL7 https://lnkd.in/gAgM8w8Z