Intel Gaudi 3 now in the IBM Cloud: Cooperation with many options, but little new

As part of Intel Vision 2025, IBM has officially announced that Intel’s Gaudi 3 accelerators are now available in selected IBM Cloud regions. Specifically, this currently concerns the Frankfurt (eu-de) and Washington D.C. (us-east) locations, with a planned expansion to Dallas (us-south) in the second quarter of 2025. By adding Gaudi 3 to its own infrastructure portfolio, IBM aims to provide a more cost-efficient alternative for companies that want to scale Generative AI workloads – at least in theory.

Goal: Cheaper AI infrastructure with the same scalability

The strategic direction of this cooperation is clear: IBM and Intel want to provide companies with a more cost-effective platform for testing, developing and productively using AI models. In view of rising infrastructure costs and growing demand for generative AI, this also appears to be necessary. According to IBM’s AI in Action 2024 report, 67% of executives surveyed reported revenue increases of 25% or more through the use of AI – with a corresponding hunger for infrastructure.

Technical implementation: more than just one instance

The modular structure of the offer is interesting. Customers have various options for using Gaudi 3:

1. Stand-alone server in the IBM Cloud VPC
   Gaudi 3 can be operated as a dedicated instance in the Virtual Private Cloud (VPC) model. This variant is particularly suitable for companies that need full control over their software setup or have special security requirements. Among other things, Red Hat Enterprise Linux with AI-specific images is also supported.
2. Containerized worker nodes
   From the second quarter of 2025, Gaudi 3 will also be available as a worker node for Red Hat OpenShift AI clusters on IBM Cloud. For companies already investing in containerized environments, this could be a seamless extension.
3. Bring-Your-Own-watsonx license model
   Customers with existing licenses for IBM’s watsonx.ai can use them on Gaudi 3 instances in the cloud. This allows them to build their own development environments – an important point for organizations with sensitive data or individual toolchains.
4. Deployable Architectures (DAs)
   To facilitate the introduction, IBM offers ready-made architecture modules for various scenarios – including virtual servers for VPC, watsonx-based DAs and OpenShift integrations. These are to follow in the second half of 2025.

Ecosystem support: Red Hat and Wipro on board

In addition to IBM and Intel, Red Hat and Wipro are also involved in the project. Red Hat particularly emphasizes the open source approach and sees the combination of Red Hat AI and Gaudi 3 as an opportunity to accelerate AI adoption in companies. Wipro, on the other hand, refers to its own GenAI Center of Excellence and the integration into existing cloud partnerships with IBM – classic statements in the style of strategic partnerships that promise a great deal of collaboration but reveal little concrete technical depth.

Security and confidentiality: Intel TDX already integrated

Apart from performance issues, IBM has also responded to the security concerns of its customers. Intel TDX (Trust Domain Extensions), which offers hardware-based functions for confidential computing, is already integrated into the IBM Cloud Virtual Server for VPC. This allows confidential data isolation and integrity to be realized at hardware level – an increasingly relevant topic, especially in regulated environments.

Broad options, limited differentiation

The availability of Intel Gaudi 3 in the IBM Cloud is technically a sensible move, especially for customers looking for alternatives to NVIDIA’s ecosystem. While the multiple deployment options – from dedicated servers to containers to pre-built deployments – offer flexibility, the real value depends heavily on performance, software integration and market adoption. While IBM is selling Gaudi 3 integration as part of a long-term multi-vendor strategy, the question remains whether customers beyond the experimental faction will be willing to move their production environments to a platform that currently has little widespread support.

Source: IBM