AI Cloud
From rapid prototyping to foundation training to scalable inference
Compute
Inference
Verda Inference New
Run frontier open-weight models with performance engineered for long-context prompts.
Run
Serving your own models? Serverless containers
Models
GB300 NVL72
New1x tray to 2+ racks · NVLink v5
Run
Verda Inference Frontier open-weight models per token, through one OpenAI compatible APIServing your own models? Serverless containers
Models
Developers
Quickstart Make your first call in minutes with an OpenAI client Coding agents Set up OpenCode, Pi Code or Codex in minutes Docs Pricing, limits & API referenceSearching…
No resources match your filters.
Meet Verda at Junction 2026 in Espoo, Finland, Nov 13-15. Get free GPU cloud credits, attend our keynote and workshop, and compete for 10,000 EUR in credits.
Meet Verda at Junction 2026 in Espoo, Finland, Nov 13-15. Get free GPU cloud credits, attend our keynote and workshop, and compete for 10,000 EUR in credits.
Verda Inference is now in preview with Kimi K3.1 Pro. Managed inference API for frontier open-weight models, optimized for long-context agentic workloads.
Improve GPU utilization in distributed training: step-by-step guide to streaming sharded data from object storage to NVMe cache with MosaicML Streaming.
Verda Object Storage is now in beta: S3-compatible, all-NVMe storage next to your GPUs for AI training and inference. Free egress and just $0.07/GiB a month.
ExpressAI uses Verda’s Confidential Computing on NVIDIA Blackwell GPUs to keep chats private, encrypted, and unreadable even to ExpressVPN itself.
Epsilon Health runs 5x more experiments and trains radiology AI models 3x faster on a dedicated NVIDIA HGX™ B300 GPU cluster from Verda, with hands-on support.
Verda is a Gold sponsor at NVIDIA GTC Berlin 2026. Meet us at booth 3059 at STATION Berlin. Join the keynote watch party by Supermicro, featuring Verda.
Cryptographically proven protection for your private data and model weights in use, at up to 99% of full speed, on multi-GPU NVIDIA HGX™ B300 and B200.
Sovereign AI, trained in Europe: Aleph Alpha's open-weight Kolibri LLM was trained on Verda's NVIDIA Blackwell clusters with 99.899% cluster uptime.
Verda retains Bronze in ClusterMAX™ being one of the 19 providers with a "recommended AI cloud provider" medallion rating, down from 41 in the previous cycle.
Verda has raised $189 million (€163 million) in the latest funding round led by Emergence Capital. This brings Verda's total funding to over $450M (€390M).
Verda has raised $189 million in new funding, led by Emergence Capital, to accelerate its full-stack AI cloud, from data centre capacity to large-scale serving.
Kubernetes, Slinky, Kueue, layered health checks, audit logs, container registry, and provisioning methods: What's new in Verda’s Instant clusters over 2026.
July recap: audit logs, Google SSO, a new CFO, and Verda on the ground across Paris, Seoul, and Las Vegas.
The Nordic Investment Bank joined Verda’s funding round in July 2026. Additional funding in September 2026 brought Verda’s total funding to over $450M (€390M).
June recap: Activity logs, C5 certification, more capacity - all while crossing $100M revenue run rate
Verda hits a $100M revenue run rate after 10x growth in 18 months, expanding AI cloud infrastructure for instant, on-demand compute across 50+ markets.
May recap: NVIDIA Rubin deployment announced, first APAC office in Taipei, deepened Supermicro collaboration, MLSys 2026 and new research.
As an NVIDIA Preferred Partner, Verda brings first-availability NVIDIA Vera Rubin NVL72 & HGX™ Rubin NVL8 systems to three regions in parallel.
Step-by-step guide to writing an NVFP4 GEMM kernel for B200 in CuTeDSL: tiling, swizzling, block-scaled FP4 quantization, and a persistent 2-CTA pipeline.
April recap: $117M funding round, SOC 2 Type II compliance, Arm AGI CPU deployment, and a packed events calendar including PyTorch Conference Europe
Verda deploys the Arm AGI CPU across its next-gen AI infrastructure, paired with NVIDIA GB300 and the upcoming Vera Rubin fleet. Arm-native, built for agentic AI.
In April 2026, Verda raised an additional $117M in funding. Verda’s latest funding round in September 2026 brought its total funding to over $450M (€390M).
Verda completes SOC 2 Type II audit, expanding its compliance coverage for AI workloads in regulated industries alongside ISO 27001, 27017, 27018, and 27701.
Verda partners with ExpressVPN to power ExpressAI using confidential computing. Learn about our joint R&D and secure GPU infrastructure.
March highlights from Verda: new data center in Sweden, platform updates, SkyPilot integration, new London office
A deep dive into NVFP4 on NVIDIA Blackwell, showing how microscaling and hardware-algorithm co-design make 4-bit floating point viable.
Verda is building a next-generation AI infrastructure hub in Sundsvall, Sweden, combining high-performance compute, sustainable energy, and advanced research to support Europe’s growing demand for AI and strengthen its sovereign AI future.
February highlights from Verda: NVIDIA Preferred Partner, Datacenter expansion plan, ISO/IEC Certificates, and more
Verda deploys NVIDIA GB300 NVL72, powered by Blackwell Ultra, in Europe. Virtualization tested with near bare-metal performance and frontier AI workloads.
Verda advances within the NVIDIA Partner Network (NPN) and becomes an NVIDIA Preferred Partner, as we build out sovereign AI infrastructure across Europe.
NVIDIA Blackwell Ultra (GB300) vs Blackwell (B200). Explore per-GPU specifications, NVL72 rack performance, Grace CPU unified memory, and NCCL test results.
January highlights from Verda: GB300 NVL72 early access, a new Terraform provider, RTX PRO 6000 capacity, FLUX.2 endpoints, and open-source AI research.
Step-by-step guide for DeepSeek-R1 NVFP4 inference with SGLang and Terraform, covering cloud B300 GPU provisioning, NVMe storage, and reproducible benchmarks.
Meta description: Manage compute and storage resources for AI workloads using Terraform with the official provider from Verda and OpenTofu compatibility.
December highlights from Verda including H200 Instant Clusters returning, ICE-01 migration, new sovereign AI ecosystems with Siili Solutions and SGLang, and our EurIPS side event in Copenhagen.
NVIDIA Blackwell and Blackwell Ultra on Verda: B200 and B300 SXM6 virtual machines with a pre-configured CUDA toolkit and the latest driver versions.
Verda and Siili Solutions have launched a strategic partnership to equip European enterprises and public-sector organizations easy access to sovereign AI models.
November was a high-momentum month for Verda, with rebrand progress, expanded GPU capacity, new pricing updates, and fresh collaborations. Catch all the highlights in our November Digest.
October was a milestone month for Verda with new hardware launches, product upgrades, events, and our rebrand from DataCrunch. Catch up on the highlights shaping the future of sustainable AI infrastructure in Europe.
Powered by 16x-128x NVIDIA B200 GPUs and InfiniBand interconnect, Instant Clusters by Verda (formerly DataCrunch) received the bronze ranking from SemiAnalysis.
DataCrunch is changing its name to Verda. Derived from the Spanish word for “true” and the Esperanto word for “green”, this name reinforces our values of technological integrity and sustainable innovation.
Join the Verda Developer Forum and Discord: two communities where AI engineers share knowledge, optimize GPUs, and collaborate on machine learning.
Performance implications and other differences between NVIDIA B300 and B200 GPUs based on the latest information available.
In September 2025, Verda raised $64M in Series A funding. Verda’s latest funding round in September 2026 brought its total funding to over $450M (€390M).
Learn how we built and optimized our AI Cloud Platform for production-grade use cases and peak efficiency across compute, networking, storage, and software.
Discover how Happywhale revolutionized marine mammal research with AI-powered whale fluke (tail) and dorsal fin photo identification on DataCrunch
Discover how Simli powers life-like AI avatar API with DataCrunch’s GPU clusters and instances, achieving sub-300ms latency and up to 50% faster startup times.
Verda, working with the Republic of Latvia, has expressed interest to the European Commission to construct an AI GigaFactory. This proposal builds on Verda's expertise with high-performance and GPU infrastructure.
Benchmark insights on the unified memory in the NVIDIA GH200 Grace Hopper system: Testing asynchronous memory copies, a memory-bound GPU kernel, and cache flushing.
Step-by-step configuration and testing of multi-node distributed workloads using TorchTitan and by integrating with Pyxis and Enroot.
Cost-efficient and low-latency image generation without compromising on model output with high GPU utilization, elastic scaling, and near-zero cold starts.
Multi-Head Latent Attention (MLA) vs. Group Query Attention (GQA): Transformer inference optimization in DeepSeek V3 with lower KV cache and higher FLOPs/s.
How WaveSpeedAI and DataCrunch achieved an up to 6x faster image inference by optimizing FLUX-dev's latency and efficiency: NVIDIA B200 vs. H100 benchmark.
A comprehensive overview of optimization techniques applied by the SGLang team for DeepSeek-V3 inference with GitHub commits, benchmarks, and results.
Multi-Head Latent Attention (MLA) improves upon Group Query Attention (GQA), enabling long-context reasoning models and wider adoption across open-source LLMs.
## Datacenter-as-a-Computer (DaaC) and the Path to Global-DaaC for LLM Training Modern AI workloads, especially large language model (LLM) training, treat an entire data center as a single computer – a concept popularized by [Barroso et al. as "Datacenter as a Computer" (DaaC)](https://research...
A step-by-step guide for deploying and benchmarking DeepSeek-R1 on 8x H200 NVIDIA GPUs, using SGLang as the inference engine and DataCrunch.
Inference benchmark DeepSeek V3 SOTA LLM using single-node and multi-node NVIDIA H200 GPUs, BF16 and FP8 quantization, and SGLang.
DataCrunch announces the deployment of NVIDIA H200 GPUs in Finland, making it one of the first in Europe to offer this next-generation AI training and inference technology with unmatched performance and sustainability.
In October 2024, Verda raised $13M in its seed funding round. Verda’s latest funding round in September 2026 brought its total funding to over $450M (€390M).
In this article we compare options for deploying cloud GPU instances, including Google Cloud Platform, Amazon AWS, Microsoft Azure, and NeoClouds such as Nebius, Lambda, RunPod and CoreWeave