# The Lambda deep learning blog

The Lambda deep learning blog

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## RoCE and NVIDIA Quantum InfiniBand, and their quantified tradeoffs

DevFeed: [RoCE and NVIDIA Quantum InfiniBand, and their quantified tradeoffs](<https://devfeed.tech/articles/roce-and-nvidia-quantum-infiniband-and-their-quantified-tradeoffs-146467.md>)

Original publisher: [Read original article](<https://lambda.com/blog/roce-and-nvidia-quantum-infiniband-and-their-quantified-tradeoffs>)

Author: Zach Mueller

Published: 2026-11-20T14:08:48Z

Content type: comparison

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [InfiniBand](<https://devfeed.tech/topics/infiniband.md>), [rdma](<https://devfeed.tech/topics/rdma.md>), [Distributed Systems & Parallel Computing](<https://devfeed.tech/topics/distributed-systems-parallel-computing.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>)

Tags: [1-click-cluster](<https://devfeed.tech/tags/1-click-cluster.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [infiniband](<https://devfeed.tech/tags/infiniband.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [roce](<https://devfeed.tech/tags/roce.md>), [tradeoffs](<https://devfeed.tech/tags/tradeoffs.md>)

### AI overview

RoCEv2 uses Ethernet with UDP/IP and can span subnets, while NVIDIA Quantum InfiniBand uses a dedicated fabric that requires InfiniBand switches and infrastructure. The text gives typical latency ranges of 0.5-2 µs for InfiniBand and 1-3 µs for optimized local RoCE, noting RoCE is more sensitive to configuration, congestion, and topology. In a two-node HGX B200 inference test using GLM-5.2-NVFP4 and SGLang, RoCE showed roughly 70% lower throughput than InfiniBand at about 250 concurrent users and 2.5x slower time to first token. The authors caution that real-world performance varies and recommend tempering expectations for On-Demand nodes; the comparison describes one setup.

### Source excerpt

The GPUs you rent are just part of what determines your cluster's speed and efficiency. This comes into play through interconnect, the network that connects multiple nodes (machines) on the backend.

## Lambda announces the world's first large-scale CPO deployment of NVIDIA Photonics switches

DevFeed: [Lambda announces the world's first large-scale CPO deployment of NVIDIA Photonics switches](<https://devfeed.tech/articles/lambda-announces-the-world-s-first-large-scale-cpo-deployment-of-nvidia-photonics-switches-136925.md>)

Original publisher: [Read original article](<https://lambda.com/blog/worlds-first-large-scale-deployment-of-nvidia-photonics-switches-with-cpo>)

Author: Lambda

Published: 2026-11-16T16:03:52Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Networking](<https://devfeed.tech/topics/ai-networking.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [NVIDIA Blackwell](<https://devfeed.tech/topics/nvidia-blackwell.md>)

Tags: [announcements](<https://devfeed.tech/tags/announcements.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [blackwell-ultra](<https://devfeed.tech/tags/blackwell-ultra.md>), [co-packaged-optics](<https://devfeed.tech/tags/co-packaged-optics.md>), [cooler](<https://devfeed.tech/tags/cooler.md>), [cpo](<https://devfeed.tech/tags/cpo.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [first](<https://devfeed.tech/tags/first.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [integrity](<https://devfeed.tech/tags/integrity.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [large](<https://devfeed.tech/tags/large.md>), [latency](<https://devfeed.tech/tags/latency.md>), [networking](<https://devfeed.tech/tags/networking.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-blackwell](<https://devfeed.tech/tags/nvidia-blackwell.md>), [nvidia-gb300](<https://devfeed.tech/tags/nvidia-gb300.md>), [nvidia-gb300-nvl72](<https://devfeed.tech/tags/nvidia-gb300-nvl72.md>), [optics](<https://devfeed.tech/tags/optics.md>), [photonics](<https://devfeed.tech/tags/photonics.md>), [scale](<https://devfeed.tech/tags/scale.md>), [switches](<https://devfeed.tech/tags/switches.md>), [world](<https://devfeed.tech/tags/world.md>)

### AI overview

Lambda says it deployed NVIDIA Photonics co-packaged optics (CPO) switches to connect more than 10,000 NVIDIA Blackwell Ultra GPUs at its Quebec campus. CPO places optical engines beside the switch ASIC, shortening electrical paths from inches to millimeters; the announcement says this reduces signal loss and removes the need for retimer chips. In benchmarks across rack, scalable-unit and pod configurations, CPO matched conventional pluggable optics in throughput, showed more stable signal integrity, and had lower latency, with the latency advantage increasing from 3.8% at rack to 10.1% at pod. Median optic temperature was 36°C versus 51°C for pluggable optics. The announcement says cooler, more stable optics support up to 15 times higher reliability. The switches have 144 800Gb/s ports, liquid cooling and 18 removable external light sources. These are company-reported findings; no benchmark methodology or independent validation is provided in the excerpt.

### Source excerpt

Deployment connects 10,000+ NVIDIA Blackwell Ultra GPUs and demonstrates lower latency, cooler optics, and improved signal integrity with co-packaged optics networking

## TurboPairFormer: faster, more stable, more accurate OpenFold3 fine-tuning

DevFeed: [TurboPairFormer: faster, more stable, more accurate OpenFold3 fine-tuning](<https://devfeed.tech/articles/turbopairformer-faster-more-stable-more-accurate-openfold3-fine-tuning-133862.md>)

Original publisher: [Read original article](<https://lambda.com/blog/faster-more-stable-protein-folding-training>)

Author: Jianwen Xie

Published: 2026-11-13T15:00:00Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [3d](<https://devfeed.tech/tags/3d.md>), [about](<https://devfeed.tech/tags/about.md>), [adapting](<https://devfeed.tech/tags/adapting.md>), [ai-for-drug-discovery](<https://devfeed.tech/tags/ai-for-drug-discovery.md>), [drug-discovery](<https://devfeed.tech/tags/drug-discovery.md>), [faster](<https://devfeed.tech/tags/faster.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [healthcare](<https://devfeed.tech/tags/healthcare.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [life-sciences](<https://devfeed.tech/tags/life-sciences.md>), [model-training](<https://devfeed.tech/tags/model-training.md>), [more](<https://devfeed.tech/tags/more.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [research](<https://devfeed.tech/tags/research.md>), [source](<https://devfeed.tech/tags/source.md>), [stable](<https://devfeed.tech/tags/stable.md>), [team](<https://devfeed.tech/tags/team.md>), [technology](<https://devfeed.tech/tags/technology.md>), [train](<https://devfeed.tech/tags/train.md>), [up](<https://devfeed.tech/tags/up.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

TurboPairFormer speeds OpenFold3 training by optimizing triangle attention and triangle multiplication, whose cost grows cubically with token count. Its backward pass reuses each recomputed attention-probability tile across query, key, value, and pair-bias gradients; output-residual compensation reduces rounding error for query, key, and pair-bias gradients; and fixed-order accumulation makes gradients deterministic. On 16 H100 GPUs, it had the lowest measured computation time per optimizer step across three crop sizes; at crop 768 it was up to 1.73x faster than the OpenFold3 Triton backend. Against cuEquivariance, mean errors fell 31.4-44.4% for query, 28.7-42.4% for key, and 16.9-37.9% for pair-bias gradients, while value-gradient accuracy was essentially tied. Gradients were bitwise identical in all 600 fixed-input cases tested. Benchmarks exclude data loading, transfers, and EMA updates. The package supports H100 and H200 GPUs.

### Source excerpt

Lambda, Stevens Institute of Technology, and the OpenFold team built open-source GPU kernels that train OpenFold3 up to 1.73x faster, with more accurate and fully reproducible gradients.

## Lambda has a new address: lambda.com

DevFeed: [Lambda has a new address: lambda.com](<https://devfeed.tech/articles/lambda-has-a-new-address-lambda-com-119765.md>)

Original publisher: [Read original article](<https://lambda.com/blog/lambda-ai-is-now-lambda-com>)

Author: Lambda

Published: 2026-11-05T17:20:10Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [hosting](<https://devfeed.tech/topics/hosting.md>)

Tags: [2025](<https://devfeed.tech/tags/2025.md>), [ai](<https://devfeed.tech/tags/ai.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [compute](<https://devfeed.tech/tags/compute.md>), [customer](<https://devfeed.tech/tags/customer.md>), [domains](<https://devfeed.tech/tags/domains.md>), [email](<https://devfeed.tech/tags/email.md>), [engineers](<https://devfeed.tech/tags/engineers.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [models](<https://devfeed.tech/tags/models.md>), [next](<https://devfeed.tech/tags/next.md>), [prototype](<https://devfeed.tech/tags/prototype.md>), [redirect](<https://devfeed.tech/tags/redirect.md>), [superintelligence](<https://devfeed.tech/tags/superintelligence.md>)

### AI overview

Lambda is moving its primary address from lambda.ai to lambda.com as it presents itself as the Superintelligence Cloud, offering compute from a single GPU to AI campuses with tens of thousands. Customers need take no action: services, accounts, contracts, and support stay the same; existing links redirect, and email addresses continue working during the transition. Lambda says communications will ultimately come from lambda.com and directs customers to lambda.com/support to report broken items.

### Source excerpt

Your path to superintelligence stays the same.

## Introducing identity and access management for Lambda Cloud

DevFeed: [Introducing identity and access management for Lambda Cloud](<https://devfeed.tech/articles/introducing-identity-and-access-management-for-lambda-cloud-118901.md>)

Original publisher: [Read original article](<https://lambda.com/blog/lambda-cloud-iam-rbac>)

Author: Lambda

Published: 2026-11-04T19:45:57Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [identity and access management](<https://devfeed.tech/topics/identity-and-access-management.md>), [rbac](<https://devfeed.tech/topics/rbac.md>), [Access Control](<https://devfeed.tech/topics/access-control.md>)

Tags: [access-control](<https://devfeed.tech/tags/access-control.md>), [access-management](<https://devfeed.tech/tags/access-management.md>), [account](<https://devfeed.tech/tags/account.md>), [admin](<https://devfeed.tech/tags/admin.md>), [api](<https://devfeed.tech/tags/api.md>), [api-key](<https://devfeed.tech/tags/api-key.md>), [api-keys](<https://devfeed.tech/tags/api-keys.md>), [audit](<https://devfeed.tech/tags/audit.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [console](<https://devfeed.tech/tags/console.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-cloud](<https://devfeed.tech/tags/gpu-cloud.md>), [groups](<https://devfeed.tech/tags/groups.md>), [iam](<https://devfeed.tech/tags/iam.md>), [identity](<https://devfeed.tech/tags/identity.md>), [identity-and-access-management](<https://devfeed.tech/tags/identity-and-access-management.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [lambda-cloud](<https://devfeed.tech/tags/lambda-cloud.md>), [product-updates](<https://devfeed.tech/tags/product-updates.md>), [rbac](<https://devfeed.tech/tags/rbac.md>), [role-based-access-control](<https://devfeed.tech/tags/role-based-access-control.md>), [roles](<https://devfeed.tech/tags/roles.md>), [security](<https://devfeed.tech/tags/security.md>), [service-identities](<https://devfeed.tech/tags/service-identities.md>)

### AI overview

Lambda Cloud IAM adds role-based access for users and software, with fine-grained roles at account or workspace scope. Resource actions require both workspace membership and a role that permits the action. Groups can contain users and service identities and pass assigned roles to their members; an effective access view shows direct and inherited roles by workspace. For automation, the article recommends giving scripts their own service identity, granting only needed roles, setting API key expiration, and rotating between its two keys; personal keys remain active until deleted. Only account admins and IAM admins can manage IAM. Existing Admin and Member access maps to Account Admin and Account Member, and current keys are unchanged. Audit logs are retained for six months; accounts can have up to 100 groups and 50 service identities. IAM is free.

### Source excerpt

Role-based access control for your GPU cloud: roles, groups, and service identities are now in the Lambda Cloud console.

## What Lambda is building with NVIDIA at GTC Berlin 2026

DevFeed: [What Lambda is building with NVIDIA at GTC Berlin 2026](<https://devfeed.tech/articles/what-lambda-is-building-with-nvidia-at-gtc-berlin-2026-92691.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/lambda-with-nvidia-at-gtc-berlin-2026>)

Author: Lambda

Published: 2026-10-21T11:20:47Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [OpenShell](<https://devfeed.tech/topics/openshell.md>), [NVIDIA DSX](<https://devfeed.tech/topics/nvidia-dsx.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-safety](<https://devfeed.tech/tags/agent-safety.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent-security](<https://devfeed.tech/tags/ai-agent-security.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [aiops](<https://devfeed.tech/tags/aiops.md>), [attestation](<https://devfeed.tech/tags/attestation.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [autonomous-agents](<https://devfeed.tech/tags/autonomous-agents.md>), [bare-metal](<https://devfeed.tech/tags/bare-metal.md>), [berlin](<https://devfeed.tech/tags/berlin.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [confidential-computing](<https://devfeed.tech/tags/confidential-computing.md>), [data-centers](<https://devfeed.tech/tags/data-centers.md>), [gtc](<https://devfeed.tech/tags/gtc.md>), [gtc-berlin-2026](<https://devfeed.tech/tags/gtc-berlin-2026.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-dsx](<https://devfeed.tech/tags/nvidia-dsx.md>), [nvidia-gtc](<https://devfeed.tech/tags/nvidia-gtc.md>), [nvidia-gtc-berlin](<https://devfeed.tech/tags/nvidia-gtc-berlin.md>), [nvidia-maxlps](<https://devfeed.tech/tags/nvidia-maxlps.md>), [nvidia-nemotron](<https://devfeed.tech/tags/nvidia-nemotron.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [openshell](<https://devfeed.tech/tags/openshell.md>), [scale](<https://devfeed.tech/tags/scale.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

Lambda describes four efforts with NVIDIA for AI infrastructure: Confidential Bare Metal uses hardware-rooted isolation and cryptographic attestation to protect models and data on dedicated third-party hardware; the NVIDIA OpenShell Reference Architecture places independent security enforcement outside agents' execution environments to govern access; planned NVIDIA Minerva deployment will connect infrastructure telemetry to detect anomalies, predict failures and help identify root causes; and a proof of concept using NVIDIA DSX MaxLPS ran 19 HGX nodes within the power budget of a 16-node baseline, reporting about 24% higher cluster-wide token throughput and 23% higher performance per watt. The article presents these as progress and goals, not measured outcomes for all deployments.

### Source excerpt

Lambda is joining the AI ecosystem at NVIDIA GTC Berlin 2026 around a shared set of priorities: making AI safer, more secure, and more efficient to run at scale.

## What Lambda is building with NVIDIA at GTC Berlin 2026

DevFeed: [What Lambda is building with NVIDIA at GTC Berlin 2026](<https://devfeed.tech/articles/what-lambda-is-building-with-nvidia-at-gtc-berlin-2026-118903.md>)

Original publisher: [Read original article](<https://lambda.com/blog/lambda-with-nvidia-at-gtc-berlin-2026>)

Author: Lambda

Published: 2026-10-21T11:20:47Z

Content type: news

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [OpenShell](<https://devfeed.tech/topics/openshell.md>), [NVIDIA DSX](<https://devfeed.tech/topics/nvidia-dsx.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [virtual machines](<https://devfeed.tech/topics/virtual-machines.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>)

Tags: [agent-safety](<https://devfeed.tech/tags/agent-safety.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent-security](<https://devfeed.tech/tags/ai-agent-security.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-for-it](<https://devfeed.tech/tags/ai-for-it.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [aiops](<https://devfeed.tech/tags/aiops.md>), [attestation](<https://devfeed.tech/tags/attestation.md>), [bare-metal](<https://devfeed.tech/tags/bare-metal.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [confidential-computing](<https://devfeed.tech/tags/confidential-computing.md>), [data-centers](<https://devfeed.tech/tags/data-centers.md>), [dsx](<https://devfeed.tech/tags/dsx.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [gtc-berlin-2026](<https://devfeed.tech/tags/gtc-berlin-2026.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-dsx](<https://devfeed.tech/tags/nvidia-dsx.md>), [nvidia-gtc](<https://devfeed.tech/tags/nvidia-gtc.md>), [nvidia-maxlps](<https://devfeed.tech/tags/nvidia-maxlps.md>), [nvidia-nemotron](<https://devfeed.tech/tags/nvidia-nemotron.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [openshell](<https://devfeed.tech/tags/openshell.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

Lambda describes five initiatives with NVIDIA at GTC Berlin 2026. Confidential Bare Metal uses hardware-rooted isolation and cryptographic attestation to help protect models and data on dedicated third-party infrastructure. The NVIDIA OpenShell Reference Architecture adds governed agent execution and security enforcement outside the agent's runtime, controlling access to tools, data, and systems. Lambda plans to deploy NVIDIA Minerva, which links telemetry across infrastructure and workloads to detect anomalies, anticipate failures, and help identify root causes. In a proof of concept on NVIDIA HGX systems, DSX MaxLPS ran 19 nodes within the aggregate power budget of a 16-node baseline, with about 24% higher cluster-wide token throughput and 23% higher performance per watt.

### Source excerpt

Lambda is joining the AI ecosystem at NVIDIA GTC Berlin 2026 around a shared set of priorities: making AI safer, more secure, and more efficient to run at scale.

## Lambda to integrate open-source NVIDIA Run:ai with its Managed Kubernetes Service

DevFeed: [Lambda to integrate open-source NVIDIA Run:ai with its Managed Kubernetes Service](<https://devfeed.tech/articles/lambda-to-integrate-open-source-nvidia-run-ai-with-its-managed-kubernetes-service-86015.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/runai-on-lambda-kubernetes-service>)

Author: Lambda

Published: 2026-10-13T14:32:00Z

Content type: news

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Kubernetes](<https://devfeed.tech/topics/kubernetes.md>), [Amazon Elastic Kubernetes Service](<https://devfeed.tech/topics/amazon-elastic-kubernetes-service.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [cloud-native-ai-infrastructure](<https://devfeed.tech/tags/cloud-native-ai-infrastructure.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [gpu-scheduling](<https://devfeed.tech/tags/gpu-scheduling.md>), [gpu-sharing](<https://devfeed.tech/tags/gpu-sharing.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [kai-scheduler](<https://devfeed.tech/tags/kai-scheduler.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [lambda-kubernetes-service](<https://devfeed.tech/tags/lambda-kubernetes-service.md>), [managed-kubernetes](<https://devfeed.tech/tags/managed-kubernetes.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-run-ai](<https://devfeed.tech/tags/nvidia-run-ai.md>), [open-source](<https://devfeed.tech/tags/open-source.md>)

### AI overview

Lambda plans to integrate open-source NVIDIA Run:ai as a managed option in Lambda Kubernetes Service. Run:ai will schedule AI workloads and allocate shared GPU capacity, while LKS provides managed Kubernetes and monitors and remediates GPU node faults. The planned integration is intended to help teams use shared GPU infrastructure more efficiently.

### Source excerpt

We welcome NVIDIA's decision to release NVIDIA Run:ai as open source, and we're working to make it a first-class option on Lambda Kubernetes Service (LKS).

## Lambda to integrate open-source NVIDIA Run:ai with its Managed Kubernetes Service

DevFeed: [Lambda to integrate open-source NVIDIA Run:ai with its Managed Kubernetes Service](<https://devfeed.tech/articles/lambda-to-integrate-open-source-nvidia-run-ai-with-its-managed-kubernetes-service-118907.md>)

Original publisher: [Read original article](<https://lambda.com/blog/runai-on-lambda-kubernetes-service>)

Author: Lambda

Published: 2026-10-13T14:32:00Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Kubernetes](<https://devfeed.tech/topics/kubernetes.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-cloud](<https://devfeed.tech/tags/ai-cloud.md>), [ai-development](<https://devfeed.tech/tags/ai-development.md>), [ai-factory](<https://devfeed.tech/tags/ai-factory.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-software](<https://devfeed.tech/tags/ai-software.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [auto-remediation](<https://devfeed.tech/tags/auto-remediation.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-native-ai-infrastructure](<https://devfeed.tech/tags/cloud-native-ai-infrastructure.md>), [compute](<https://devfeed.tech/tags/compute.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [gpu-scheduling](<https://devfeed.tech/tags/gpu-scheduling.md>), [gpu-sharing](<https://devfeed.tech/tags/gpu-sharing.md>), [kai-scheduler](<https://devfeed.tech/tags/kai-scheduler.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [lambda-kubernetes-service](<https://devfeed.tech/tags/lambda-kubernetes-service.md>), [managed-kubernetes](<https://devfeed.tech/tags/managed-kubernetes.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-run-ai](<https://devfeed.tech/tags/nvidia-run-ai.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [run-ai](<https://devfeed.tech/tags/run-ai.md>)

### AI overview

Lambda plans to offer open-source NVIDIA Run:ai as a managed option in Lambda Kubernetes Service (LKS), building on its support for customer-deployed Run:ai since 2022. Run:ai would extend Kubernetes with GPU-aware scheduling, GPU fractions for sharing devices, and quotas, priorities, and fair-share policies that let teams borrow unused capacity and yield it to higher-priority work. LKS manages the underlying environment: it monitors GPU nodes, cordons nodes when faults occur, drains running work, attempts automatic correction or escalates for hardware repair, and returns nodes after revalidation. Lambda says it will package, validate, and maintain the integration, which is still planned; the text gives no availability date.

### Source excerpt

We welcome NVIDIA's decision to release NVIDIA Run:ai as open source, and we're working to make it a first-class option on Lambda Kubernetes Service (LKS).

## ReLearn at CVPR 2026: can AI still learn from humans?

DevFeed: [ReLearn at CVPR 2026: can AI still learn from humans?](<https://devfeed.tech/articles/relearn-at-cvpr-2026-can-ai-still-learn-from-humans-118906.md>)

Original publisher: [Read original article](<https://lambda.com/blog/relearn-cvpr-2026>)

Author: Jianwen Xie

Published: 2026-10-07T22:14:56Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [3d](<https://devfeed.tech/tags/3d.md>), [ai](<https://devfeed.tech/tags/ai.md>), [cognitive-science](<https://devfeed.tech/tags/cognitive-science.md>), [compute](<https://devfeed.tech/tags/compute.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [cvpr-2026](<https://devfeed.tech/tags/cvpr-2026.md>), [data](<https://devfeed.tech/tags/data.md>), [embodied-ai](<https://devfeed.tech/tags/embodied-ai.md>), [environment](<https://devfeed.tech/tags/environment.md>), [events](<https://devfeed.tech/tags/events.md>), [experience](<https://devfeed.tech/tags/experience.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [generative](<https://devfeed.tech/tags/generative.md>), [global](<https://devfeed.tech/tags/global.md>), [humanobject-interaction](<https://devfeed.tech/tags/humanobject-interaction.md>), [learn](<https://devfeed.tech/tags/learn.md>), [models](<https://devfeed.tech/tags/models.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [research](<https://devfeed.tech/tags/research.md>), [scaling](<https://devfeed.tech/tags/scaling.md>)

### AI overview

ReLearn's speakers argued that visual and physical AI may need structured world models built from continuous interaction, alongside large-scale learning. They identified limits in current models' spatial understanding and persistent world states, and explored complementary approaches: cognitive-developmental accounts of spatial representations, egocentric first-person experience, explicit 3D representations for reasoning, and models of persistent physical objects. The workshop also examined how human assumptions enter learning through language, visual augmentations, and objectives, even without labels. Human knowledge and cognition can provide useful learning principles and diagnostics, but may also constrain what machines discover; the open question is which structures to learn from people and which to learn independently. These are themes and research directions reported in the supplied excerpt, not a single recipe or demonstrated solution.

### Source excerpt

AI has advanced fast by scaling models, data, and compute. As foundation models get more capable, an old question is coming back...

## ReLearn at CVPR 2026: How human intelligence may inform AI learning

DevFeed: [ReLearn at CVPR 2026: How human intelligence may inform AI learning](<https://devfeed.tech/articles/relearn-at-cvpr-2026-can-ai-still-learn-from-humans-84860.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/relearn-cvpr-2026>)

Author: Jianwen Xie

Published: 2026-10-07T22:14:56Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [streaming data analytics](<https://devfeed.tech/topics/streaming-data-analytics.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [3D](<https://devfeed.tech/topics/3d.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cvpr-2026](<https://devfeed.tech/tags/cvpr-2026.md>), [embodied-ai](<https://devfeed.tech/tags/embodied-ai.md>), [events](<https://devfeed.tech/tags/events.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [research](<https://devfeed.tech/tags/research.md>), [scaling](<https://devfeed.tech/tags/scaling.md>)

### AI overview

At the ReLearn workshop at CVPR 2026, researchers examined where human knowledge can guide AI and where systems should discover structure independently. Talks explored spatial representations, continuous first-person experience, persistent world models, and physical structure as directions for visual and embodied AI. The workshop's themes suggest that future progress may depend on structuring intelligence as well as scaling models, data, and compute.

### Source excerpt

AI has advanced fast by scaling models, data, and compute. As foundation models get more capable, an old question is coming back...

## Lambda closes $1 billion senior secured fixed rate financing

DevFeed: [Lambda closes $1 billion senior secured fixed rate financing](<https://devfeed.tech/articles/lambda-closes-1-billion-senior-secured-fixed-rate-financing-118900.md>)

Original publisher: [Read original article](<https://lambda.com/blog/lambda-closes-1-billion-senior-secured-fixed-rate-financing>)

Author: Lambda

Published: 2026-10-01T23:08:14Z

Content type: news

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [about](<https://devfeed.tech/tags/about.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-cloud](<https://devfeed.tech/tags/ai-cloud.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-infrastructure-financing](<https://devfeed.tech/tags/ai-infrastructure-financing.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [capital](<https://devfeed.tech/tags/capital.md>), [company](<https://devfeed.tech/tags/company.md>), [credit-facility](<https://devfeed.tech/tags/credit-facility.md>), [deployments](<https://devfeed.tech/tags/deployments.md>), [financing](<https://devfeed.tech/tags/financing.md>), [first](<https://devfeed.tech/tags/first.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [insurance](<https://devfeed.tech/tags/insurance.md>), [investors](<https://devfeed.tech/tags/investors.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [marketed-to](<https://devfeed.tech/tags/marketed-to.md>), [plus](<https://devfeed.tech/tags/plus.md>), [superintelligence](<https://devfeed.tech/tags/superintelligence.md>), [support](<https://devfeed.tech/tags/support.md>)

### AI overview

Lambda closed a $1.008 billion investment-grade delayed-draw term loan to fund GPU infrastructure for three committed deployments across multiple data centers. The financing is backed by GPU servers, related infrastructure, and contracted cash flows from two investment-grade customers. Its 6.78% fixed rate is paid semiannually; it matures May 30, 2033, with fully amortizing repayment. The delayed-draw structure aligns funding with infrastructure commissioning, and the facility received ratings of A (low) from Morningstar DBRS and Baa1 from Moody's.

### Source excerpt

Marks Lambda's first $1 billion-plus institutional debt financing marketed to insurance companies and fixed income investors, and second institutional credit facility, expanding access to capital to support committed AI infrastructure deployments

## Lambda closes $1 billion senior secured fixed rate financing

DevFeed: [Lambda closes $1 billion senior secured fixed rate financing](<https://devfeed.tech/articles/lambda-closes-1-billion-senior-secured-fixed-rate-financing-84855.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/lambda-closes-1-billion-senior-secured-fixed-rate-financing>)

Author: Lambda

Published: 2026-10-01T23:08:14Z

Content type: news

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [nuclear energy for AI](<https://devfeed.tech/topics/nuclear-energy-for-ai.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [company](<https://devfeed.tech/tags/company.md>), [financing](<https://devfeed.tech/tags/financing.md>), [funding](<https://devfeed.tech/tags/funding.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [superintelligence](<https://devfeed.tech/tags/superintelligence.md>)

### AI overview

Lambda announced a $1.008 billion delayed-draw term loan to fund GPU infrastructure for three committed customer deployments across multiple data centers. The financing has a 6.78% fixed interest rate, carries investment-grade ratings, and releases funds as infrastructure reaches commissioning milestones.

### Source excerpt

Marks Lambda's first $1 billion-plus institutional debt financing marketed to insurance companies and fixed income investors, and second institutional credit facility, expanding access to capital to support committed AI infrastructure deployments

## From world models to active agents: the next step for physical AI

DevFeed: [From world models to active agents: the next step for physical AI](<https://devfeed.tech/articles/from-world-models-to-active-agents-the-next-step-for-physical-ai-84861.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/world-models-active-agents-physical-ai>)

Author: Jianwen Xie

Published: 2026-10-01T22:37:02Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Computer vision](<https://devfeed.tech/topics/computer-vision.md>), [contact-rich manipulation](<https://devfeed.tech/topics/contact-rich-manipulation.md>)

Tags: [active-sensing](<https://devfeed.tech/tags/active-sensing.md>), [ai](<https://devfeed.tech/tags/ai.md>), [closed-loop-planning](<https://devfeed.tech/tags/closed-loop-planning.md>), [control](<https://devfeed.tech/tags/control.md>), [cvpr-2026](<https://devfeed.tech/tags/cvpr-2026.md>), [embodied-ai](<https://devfeed.tech/tags/embodied-ai.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [world-models](<https://devfeed.tech/tags/world-models.md>)

### AI overview

Research presented at the CVPR 2026 WMAS workshop explores how world models can support agents that sense, plan, and act in physical environments. Benchmarks highlight gaps between visually plausible predictions and executable actions, while GEM-4D reports improved real-world manipulation success using geometry-grounded predictions.

### Source excerpt

Generative AI has made remarkable progress in creating increasingly realistic representations of the world. Image models can synthesize photorealistic scenes, video models can generate complex dynamics over time, and emerging world models are beginning to capture geometry, motion, and interaction.

## From world models to active agents: the next step for physical AI

DevFeed: [From world models to active agents: the next step for physical AI](<https://devfeed.tech/articles/from-world-models-to-active-agents-the-next-step-for-physical-ai-118908.md>)

Original publisher: [Read original article](<https://lambda.com/blog/world-models-active-agents-physical-ai>)

Author: Jianwen Xie

Published: 2026-10-01T22:37:02Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Computer vision](<https://devfeed.tech/topics/computer-vision.md>)

Tags: [active-sensing](<https://devfeed.tech/tags/active-sensing.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [closed-loop-planning](<https://devfeed.tech/tags/closed-loop-planning.md>), [cvpr-2026](<https://devfeed.tech/tags/cvpr-2026.md>), [embodied-ai](<https://devfeed.tech/tags/embodied-ai.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [robotics](<https://devfeed.tech/tags/robotics.md>), [world-models](<https://devfeed.tech/tags/world-models.md>)

### AI overview

Physical AI needs world models that support sensing, planning, and actions that succeed in the real world. RoboWM-Bench finds that visually plausible generated videos can imply manipulation behaviors that fail because of spatial reasoning, contact prediction, or physical consistency errors. GEM-4D adds dense 4D correspondence supervision and uses inverse dynamics to turn generated rollouts into executable 6-DoF trajectories; its authors report real-world manipulation success rising from 61% to 81%. SAW-Bench highlights weaknesses in situated spatial awareness, reporting a 37.66% gap between humans and the best of 24 evaluated models. Together, the work points toward a closed loop in which agents use world models to imagine outcomes, actively gather useful observations, act, and revise plans based on what happens. The article presents this as an emerging direction, with open questions about uncertainty, when to sense, and adaptation to unfamiliar environments.

### Source excerpt

Generative AI has made remarkable progress in creating increasingly realistic representations of the world. Image models can synthesize photorealistic scenes, video models can generate complex dynamics over time, and emerging world models are beginning to capture geometry, motion, and interaction.

## Open Jarvis: making local LLMs work as agents

DevFeed: [Open Jarvis: making local LLMs work as agents](<https://devfeed.tech/articles/open-jarvis-making-local-llms-work-as-agents-118905.md>)

Original publisher: [Read original article](<https://lambda.com/blog/open-jarvis-local-llm-agents>)

Author: caia.costello@lambdal.com (Caia Costello)

Published: 2026-09-30T14:49:50Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [llama.cpp](<https://devfeed.tech/topics/llama-cpp.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agent-harnesses](<https://devfeed.tech/tags/agent-harnesses.md>), [agent-tools](<https://devfeed.tech/tags/agent-tools.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-benchmarks](<https://devfeed.tech/tags/ai-benchmarks.md>), [api](<https://devfeed.tech/tags/api.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [code](<https://devfeed.tech/tags/code.md>), [code-execution](<https://devfeed.tech/tags/code-execution.md>), [config](<https://devfeed.tech/tags/config.md>), [configuration](<https://devfeed.tech/tags/configuration.md>), [control](<https://devfeed.tech/tags/control.md>), [cost](<https://devfeed.tech/tags/cost.md>), [harness](<https://devfeed.tech/tags/harness.md>), [kimi-k2-6](<https://devfeed.tech/tags/kimi-k2-6.md>), [lambda-gpu-cloud](<https://devfeed.tech/tags/lambda-gpu-cloud.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [llms](<https://devfeed.tech/tags/llms.md>), [local](<https://devfeed.tech/tags/local.md>), [local-inference](<https://devfeed.tech/tags/local-inference.md>), [local-llms](<https://devfeed.tech/tags/local-llms.md>), [open-weight-models](<https://devfeed.tech/tags/open-weight-models.md>), [pinchbench](<https://devfeed.tech/tags/pinchbench.md>)

### AI overview

Open Jarvis argues that an agent's harness must be retuned for its model and hardware: prompts and tool strategies built for cloud models can limit open-weight models. Its spec separates model settings, inference engine, agent logic, tools and memory, and learning; automated search uses a frontier model to diagnose failures and accept improvements across four editable areas. On PinchBench, retargeting a Qwen3.5-9B harness raised accuracy from 62.3% to 88.4%, recovering 77% of the gap to Claude Opus 4.6. The excerpt also reports broader local-model results, but notes a workload-specific tradeoff: on 75 shared PinchBench tasks using the native OpenHands harness, Kimi K2.6 scored below Claude Fable 5.1 (0.7301 vs. 0.8216) while estimated cost was about 34% lower ($2.07 vs. $3.14 per task). Self-hosting may suit workloads where Kimi's quality meets requirements; the excerpt provides setup commands for trying Open Jarvis.

### Source excerpt

Running an open-weight model is only half the job. The other half is the harness: what tells the model what to do, which tools it can use, and how to learn from its mistakes.

## Open Jarvis: making local LLMs work as agents

DevFeed: [Open Jarvis: making local LLMs work as agents](<https://devfeed.tech/articles/open-jarvis-making-local-llms-work-as-agents-84859.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/open-jarvis-local-llm-agents>)

Author: caia.costello@lambdal.com (Caia Costello)

Published: 2026-09-30T14:49:50Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [OpenClaw](<https://devfeed.tech/topics/openclaw.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [agent-harnesses](<https://devfeed.tech/tags/agent-harnesses.md>), [agent-tools](<https://devfeed.tech/tags/agent-tools.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [code](<https://devfeed.tech/tags/code.md>), [kimi-k2-6](<https://devfeed.tech/tags/kimi-k2-6.md>), [lambda-gpu-cloud](<https://devfeed.tech/tags/lambda-gpu-cloud.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [local-inference](<https://devfeed.tech/tags/local-inference.md>), [local-llms](<https://devfeed.tech/tags/local-llms.md>), [open-weight-models](<https://devfeed.tech/tags/open-weight-models.md>), [pinchbench](<https://devfeed.tech/tags/pinchbench.md>)

### AI overview

Open Jarvis retunes an AI agent harness--including its model, inference engine, agent logic, tools, and memory--for a particular local model and machine. The article reports that this recovered much of the accuracy lost when replacing a cloud model with a smaller open-weight model on PinchBench, and describes harness configuration and optimization as key to making local models useful for agentic tasks. It also presents Lambda infrastructure benchmarks and cost comparisons, with results that vary by model, benchmark, and workload.

### Source excerpt

Running an open-weight model is only half the job. The other half is the harness: what tells the model what to do, which tools it can use, and how to learn from its mistakes.

## Lambda to build new data center in Mayes County, Oklahoma, generating half a billion dollars in tax revenue over next decade

DevFeed: [Lambda to build new data center in Mayes County, Oklahoma, generating half a billion dollars in tax revenue over next decade](<https://devfeed.tech/articles/lambda-to-build-new-data-center-in-mayes-county-oklahoma-generating-half-a-billion-dollars-in-tax-revenue-over-next-decade-84858.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/new-lambda-data-center-in-oklahoma>)

Author: Lambda

Published: 2026-09-27T14:01:36Z

Content type: news

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [clean energy data centers](<https://devfeed.tech/topics/clean-energy-data-centers.md>), [Sustainable Computing](<https://devfeed.tech/topics/sustainable-computing.md>)

Tags: [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [company](<https://devfeed.tech/tags/company.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [cooling](<https://devfeed.tech/tags/cooling.md>), [costs](<https://devfeed.tech/tags/costs.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [data-centers](<https://devfeed.tech/tags/data-centers.md>), [energy](<https://devfeed.tech/tags/energy.md>), [superintelligence](<https://devfeed.tech/tags/superintelligence.md>), [water](<https://devfeed.tech/tags/water.md>)

### AI overview

Lambda announced plans to build an AI data center in MidAmerica Industrial Park in Mayes County, Oklahoma. The project is expected to create up to 1,000 construction jobs and 100 permanent positions. Lambda says it will pay all facility energy costs, fund required infrastructure, and use closed-loop cooling to reduce water use. It also plans to shift or reduce some computing loads during peak electricity demand.

### Source excerpt

Facility will be built in MidAmerica Industrial Park & deploy closed-loop cooling technology to preserve and recycle water; all energy costs paid in compliance with Oklahoma's ratepayer protections

## Lambda to build new data center in Mayes County, Oklahoma, generating half a billion dollars in tax revenue over next decade

DevFeed: [Lambda to build new data center in Mayes County, Oklahoma, generating half a billion dollars in tax revenue over next decade](<https://devfeed.tech/articles/lambda-to-build-new-data-center-in-mayes-county-oklahoma-generating-half-a-billion-dollars-in-tax-revenue-over-next-decade-118904.md>)

Original publisher: [Read original article](<https://lambda.com/blog/new-lambda-data-center-in-oklahoma>)

Author: Lambda

Published: 2026-09-27T14:01:36Z

Content type: release

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [clean energy data centers](<https://devfeed.tech/topics/clean-energy-data-centers.md>), [Sustainable Computing](<https://devfeed.tech/topics/sustainable-computing.md>)

Tags: [ai-cloud](<https://devfeed.tech/tags/ai-cloud.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-training-and-inference](<https://devfeed.tech/tags/ai-training-and-inference.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [build](<https://devfeed.tech/tags/build.md>), [company](<https://devfeed.tech/tags/company.md>), [construction](<https://devfeed.tech/tags/construction.md>), [cooling](<https://devfeed.tech/tags/cooling.md>), [costs](<https://devfeed.tech/tags/costs.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [data-centers](<https://devfeed.tech/tags/data-centers.md>), [energy-costs](<https://devfeed.tech/tags/energy-costs.md>), [lambda](<https://devfeed.tech/tags/lambda.md>), [revenue](<https://devfeed.tech/tags/revenue.md>), [superintelligence](<https://devfeed.tech/tags/superintelligence.md>), [tax](<https://devfeed.tech/tags/tax.md>), [water](<https://devfeed.tech/tags/water.md>)

### AI overview

Lambda plans to build a data center at MidAmerica Industrial Park in Mayes County, Oklahoma. The project is estimated to generate $500 million in taxes over 10 years, create up to 1,000 construction jobs and up to 100 permanent positions. Lambda says it will pay all facility energy costs and fund the upfront infrastructure needed to serve the site, and is designing the facility to shift or reduce some computing loads during peak demand. It also plans to use closed-loop cooling to reduce water use and site the facility away from residential areas. The project will sit within the Chouteau-Mazie Public School District tax boundary. The text says local leaders and residents will be consulted as planning proceeds; a project timeline and further economic details are still to come.

### Source excerpt

Facility will be built in MidAmerica Industrial Park & deploy closed-loop cooling technology to preserve and recycle water; all energy costs paid in compliance with Oklahoma's ratepayer protections

## Building multimodal models for spatial reasoning

DevFeed: [Building multimodal models for spatial reasoning](<https://devfeed.tech/articles/building-multimodal-models-for-spatial-reasoning-84852.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/cvp-spatial-reasoning>)

Author: Jianwen Xie

Published: 2026-09-24T19:58:32Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [human computer interaction](<https://devfeed.tech/topics/human-computer-interaction.md>), [World models](<https://devfeed.tech/topics/world-models.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [3d-scene-understanding](<https://devfeed.tech/tags/3d-scene-understanding.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [computer-vision-research](<https://devfeed.tech/tags/computer-vision-research.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [next](<https://devfeed.tech/tags/next.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [robots](<https://devfeed.tech/tags/robots.md>), [spacial-reasoning](<https://devfeed.tech/tags/spacial-reasoning.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

The article describes CVP, a multimodal model that combines a target-affinity token for focusing on task-relevant objects with an allocentric grid for representing the broader 3D scene. It reports improvements over a baseline across five 3D vision-language benchmarks and outlines the compute used to train the model.

### Source excerpt

What is spatial reasoning? AI is moving beyond the digital world. The next generation of AI systems needs to understand and interact with physical environments, from robots to autonomous systems navigating complex spaces.

## Building multimodal models for spatial reasoning

DevFeed: [Building multimodal models for spatial reasoning](<https://devfeed.tech/articles/building-multimodal-models-for-spatial-reasoning-118899.md>)

Original publisher: [Read original article](<https://lambda.com/blog/cvp-spatial-reasoning>)

Author: Jianwen Xie

Published: 2026-09-24T19:58:32Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [3D](<https://devfeed.tech/topics/3d.md>)

Tags: [3d-scene-understanding](<https://devfeed.tech/tags/3d-scene-understanding.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [computer-vision-research](<https://devfeed.tech/tags/computer-vision-research.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [spacial-reasoning](<https://devfeed.tech/tags/spacial-reasoning.md>), [spatial-reasoning](<https://devfeed.tech/tags/spatial-reasoning.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

CVP combines task-focused object selection with a world-centered scene map to improve 3D spatial reasoning. Its target-affinity token learns which objects matter to a query, while a compact allocentric grid gives the language model broader spatial context; the default 6 x 6 grid performed strongly, and higher resolution added no benefit. Evaluated against Video-3D-LLM on five benchmarks, CVP improved reported scores for question answering, visual grounding, and dense captioning. Ablations attribute larger grounding gains to the target-affinity token and greater question-answering and captioning benefits to the grid. Researchers fine-tuned a 7B model across five datasets using eight NVIDIA A100 GPUs.

### Source excerpt

What is spatial reasoning? AI is moving beyond the digital world. The next generation of AI systems needs to understand and interact with physical environments, from robots to autonomous systems navigating complex spaces.

## From structures to dynamics: scaling AI for molecular dynamics in drug discovery

DevFeed: [From structures to dynamics: scaling AI for molecular dynamics in drug discovery](<https://devfeed.tech/articles/from-structures-to-dynamics-scaling-ai-for-molecular-dynamics-in-drug-discovery-84854.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/eginterpolator-molecular-dynamics-drug-discovery>)

Author: Jianwen Xie

Published: 2026-09-21T17:26:34Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [Molecular Dynamics](<https://devfeed.tech/topics/molecular-dynamics.md>), [AI for science](<https://devfeed.tech/topics/ai-for-science.md>), [BioNeMo](<https://devfeed.tech/topics/bionemo.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-for-drug-discovery](<https://devfeed.tech/tags/ai-for-drug-discovery.md>), [biotech](<https://devfeed.tech/tags/biotech.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [drug-discovery](<https://devfeed.tech/tags/drug-discovery.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [gpu-computing](<https://devfeed.tech/tags/gpu-computing.md>), [md-trajectories](<https://devfeed.tech/tags/md-trajectories.md>), [molecular-dynamics](<https://devfeed.tech/tags/molecular-dynamics.md>), [molecular-simulation](<https://devfeed.tech/tags/molecular-simulation.md>), [molecular-structure-prediction](<https://devfeed.tech/tags/molecular-structure-prediction.md>), [scaling-ai](<https://devfeed.tech/tags/scaling-ai.md>), [structure-pretraining](<https://devfeed.tech/tags/structure-pretraining.md>)

### AI overview

EGInterpolator uses a two-stage approach to generate molecular dynamics trajectories: it first pretrains a diffusion model on abundant molecular conformer data, then fine-tunes a trajectory interpolator on scarcer molecular dynamics data. On the DRUGS benchmark, it outperformed GeoTDM on reported measures of molecular geometry and motion. The article describes the work as a way to make AI-driven molecular dynamics more data-efficient for drug discovery.

### Source excerpt

Why molecular dynamics matters for drug discovery Molecular dynamics (MD) is an important tool in modern drug discovery. While structure prediction and molecular docking can show how a potential drug might fit into a protein, they provide only a snapshot. Molecules are constantly moving. MD simulations allow researchers to see how a drug and its target behave over time: whether the interaction remains stable, how the protein changes shape, and how the molecular system evolves.

## Closing the loop: agentic evaluation for image editing foundation models

DevFeed: [Closing the loop: agentic evaluation for image editing foundation models](<https://devfeed.tech/articles/closing-the-loop-agentic-evaluation-for-image-editing-foundation-models-84853.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/edival-agent-agentic-evaluation-image-editing>)

Author: Jianwen Xie

Published: 2026-09-17T16:29:33Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [ai code review](<https://devfeed.tech/topics/ai-code-review.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-evaluation](<https://devfeed.tech/tags/ai-evaluation.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [iclr-2026](<https://devfeed.tech/tags/iclr-2026.md>), [image-editing](<https://devfeed.tech/tags/image-editing.md>), [model-development](<https://devfeed.tech/tags/model-development.md>), [multi-turn-editing](<https://devfeed.tech/tags/multi-turn-editing.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>)

### AI overview

Lambda researchers describe EdiVal-Agent, a framework that uses specialized AI models and visual tools to evaluate image-editing foundation models. It assesses instruction following, content consistency across editing turns, and visual quality. The article reports 81.3% agreement with human judgments for its instruction-following component, compared with 75.2% for a vision-language-model-only evaluator and 68.9% for a thresholded CLIP-based metric. The framework is presented as a way to identify model failure modes and support iterative development.

### Source excerpt

Why evaluating image editing models is both critical and challenging Instruction-based image editing is becoming a core capability of multimodal foundation models. Users can increasingly edit images simply by describing what they want: "remove the person in the background," "make the car red," or "move the chair next to the table."

## MLPerf Inference v6.1: pioneering agent, VLM benchmarks

DevFeed: [MLPerf Inference v6.1: pioneering agent, VLM benchmarks](<https://devfeed.tech/articles/mlperf-inference-v6-1-pioneering-agent-vlm-benchmarks-84857.md>)

Original publisher: [Read original article](<https://lambda.ai/blog/mlperf-inference-v6.1>)

Author: Lambda

Published: 2026-09-16T15:02:49Z

Content type: article

Language: en

Sources: [The Lambda deep learning blog](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-inference-benchmark](<https://devfeed.tech/tags/agentic-inference-benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [datacenter](<https://devfeed.tech/tags/datacenter.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gguf](<https://devfeed.tech/tags/gguf.md>), [gpt-oss-120b-throughput](<https://devfeed.tech/tags/gpt-oss-120b-throughput.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [inference](<https://devfeed.tech/tags/inference.md>), [mlperf](<https://devfeed.tech/tags/mlperf.md>), [mlperf-inference-v6-1](<https://devfeed.tech/tags/mlperf-inference-v6-1.md>), [model](<https://devfeed.tech/tags/model.md>), [qwen3-vl-inference](<https://devfeed.tech/tags/qwen3-vl-inference.md>), [trillion-parameter-inference](<https://devfeed.tech/tags/trillion-parameter-inference.md>)

### AI overview

Lambda presents its MLPerf Inference v6.1 results, reporting throughput gains on unchanged hardware, VLM benchmark results, and an open agentic inference run using a model with over one trillion parameters. The article describes workload setups and performance measurements across NVIDIA Blackwell Ultra and HGX B200 systems, while noting that the agentic results are not directly comparable with closed division edge submissions.

### Source excerpt

First agentic workload on datacenter hardware in MLPerf, and the first model over a trillion parameters. Plus 8.85% more throughput on identical hardware since v6.0.

[Next page](<https://devfeed.tech/sources/the-lambda-deep-learning-blog.md?cursor=WyIyMDI2LTA5LTE2VDE1OjAyOjQ5KzAwOjAwIiwgIjlhMGVmMjIwLWJkMjgtNGRjYy1iODYxLTU3MDVkZmFlYzVkNyJd>)