Takomo (also appearing under the name Verda in the materials) positions itself as a "full-stack AI cloud," with its cloud infrastructure shaped by AI workload demands, covering software, compute, storage, and networking, with the goal of spanning the entire AI lifecycle: from rapid prototyping, to foundation model training, to scalable inference.
Purpose and Positioning The platform is aimed at AI/ML workloads, and is officially described as a one-stop platform covering the complete AI lifecycle. The materials mention its scaling form from 1x tray to 2+ racks as well as NVLink v5.
Main Capabilities For compute, it offers: Instant clusters (self-service GPU clusters with InfiniBand interconnect), GPU instances (fast access to NVIDIA's latest hardware), CPU instances (dedicated cores, high clock speeds); for serverless, it offers Serverless containers (auto-scaling GPU containers for inference and batch jobs); for storage, it offers Block storage (high-speed NVMe virtual disks), Shared filesystem (POSIX-compliant storage), Container registry (OCI-compliant image storage).
Available GPU Models The GPUs listed in the materials include: GB300 NVL72 (new, 1x tray to 2+ racks, NVLink v5), B300 SXM6 (30 CPUs, 255 GB RAM, 268 GB VRAM), B200 SXM6 (30 CPUs, 170 GB RAM, 180 GB VRAM), RTX PRO 6000 (30 CPUs, 90 GB RAM, 96 GB VRAM), H200 SXM5 (44 CPUs, 170 GB RAM, 141 GB VRAM), H100 SXM5 (30 CPUs, 120 GB RAM, 80 GB VRAM), A100 SXM4 (22 CPUs, 120 GB RAM, 80 GB VRAM).
Other Solutions and Resources It offers AI Lab (an in-house AI lab participating in frontier research and open-source projects) and Confidential computing (hardware-attested inference and fine-tuning). In terms of resources, it offers Blog (GPU benchmarks and AI R&D), Docs (technical documentation and tutorials), API (control GPU resources through external code), Community Forum (exchange with AI experts), Trust center (Verda's security and compliance).
Advantages and Guarantees It is officially stated to have full-stack integration capability, with end-to-end autonomous control from the technology stack and data centers to managed services, thereby bringing predictable cost, performance, and reliability; support for on-demand self-service access to the latest compute without negotiation and contracts; historical uptime exceeding 99.9%, along with a commitment to a reasonable SLA and service interruption compensation; proactive support provided by a team of ML experts and infrastructure engineers; in terms of security and compliance, it is certified under SOC 2 Type II and ISO 27001/27017/27018/27701, complies with GDPR, and uses 100% renewable energy. In addition, as an NVIDIA Preferred Partner, it can obtain the latest hardware in advance, and it states uptime exceeding 99.95%.
Customer and Partnership Cases The materials mention that Verda partnered with 1X to build multi-GPU inference for the 1XWM generative video model; continuously sponsors SGLang and provides compute and infrastructure support, and SGLang's FP8 and INT4 reinforcement learning training exploration used NVIDIA Hopper and Blackwell architectures; vLLM obtained early access to Verda's GB300 cluster and infrastructure support, achieving an 8x prefill and 10-20x mixed-context improvement on Blackwell Ultra compared with Hopper. Magnific uses Verda's fully managed GPU queue to elastically scale to absorb daily traffic peaks, and combines inference optimization to reduce generation latency and throughput costs. ExpressAI (a new service from ExpressVPN) adopts LLM inference within hardware-level isolation; Verda co-designed confidential computing on NVIDIA Blackwell architecture, extending hardware-attested isolation to multi-GPU configurations. Aleph Alpha's open-weight sovereign LLM Kolibri was trained on Verda's NVIDIA Blackwell cluster, with cluster uptime of 99.899%. Verda in SemiAnalysis's ClusterMArated Bronze in X™ 3.0 and received the "Recommended AI Cloud Provider" badge. Verda completed a $189 million (€163 million) funding round led by Emergence Capital, bringing total funding to over $450 million (€390 million).
Use cases and limitations Suitable for AI/ML teams that need GPU clusters for training, inference, batch processing, and rapid prototyping, especially customers focused on European deployment, compliance certifications, and renewable energy. Details such as specific pricing, billing methods, regional availability, and usage steps are not yet provided.
