Scaleway’s recent announcements have been somewhat overshadowed by the high-profile launch of Kyutai, Europe’s first open research laboratory in artificial intelligence, co-founded by Xavier Niel. During the ai-PULSE conference on November 17, the cloud subsidiary of the iliad group nevertheless gave details of its strategy in the field of AI.
Scaleway has confirmed the acquisition of “ the largest AI cloud computing power deployed to date in Europe “. Called Nebuchadnezzar, this supercomputer is based on Nvidia’s DGX SuperPOD platform. The latter includes 127 DGX H100 systems which each integrate eight Nvidia H100 Tensor Core GPUs, for a total of 1016 latest generation graphics chips dedicated to large language models (LLM) of generative AI.
“The DGX SuperPOD supercomputer can train large language models up to four times faster than computers using previous versions of GPUs », says iliad in a press release. For the training of smaller and targeted models, the cloud provider offers, with another offer called Jeroboam, less computing power, made up of two interconnected DGX H100 systems, i.e. 16 Nvidia H100 GPUs.
Cover all AI use cases
To accelerate the development and deployment of AI models, enterprise customers rely on Nvidia’s range of dedicated tools such as Nvidia AI Enterprise software, including more than a hundred already trained AI models. Startups can benefit from cloud credits to access this cloud power via the Nvidia Inception program.
To complete its system in the field of AI, Scaleway has formed a partnership with Ampere Computing. Founded in 2018, this California-based semiconductor manufacturer is the originator of the industry’s first cloud-native processors. Based on the ARM architecture, Ampere Altra is an alternative to traditional x86-type processors.
“More efficient and less energy intensive “, this processor constitutes, according to Scaleway, a cost-effective solution for the inference (prediction operation) of AI models, alongside GPU graphics processors. The supplier cites, as use cases, the implementation of chatbots, real-time analysis or video content analysis.
Quantum computing as a service
After AI, Scaleway is moving towards another technological revolution: quantum computing. This time, the provider has entered into a partnership with Quandela, a French startup which recently obtained 50 million euros in funding and delivered its first quantum computer.
While awaiting the arrival of quantum supremacy – i.e. computing power greater than traditional supercomputers – Scaleway will offer, through this partnership, a Quantum Computing as a Service (QCaaS) offering. This type of solution allows companies to experiment with quantum computing and carry out their first experiments in risk analysis in the financial field, the simulation of the design of a prototype or the modeling of a new molecule.
In a hybrid approach, both classical and quantum, Scaleway’s QCaaS service is powered by an exclusive Quandela circuit simulator, and accelerated by the famous Nvidia H100 Tensor Core GPUs. “ This technological advancement offers a powerful alternative while awaiting the maturity of large-scale quantum processing units (QPUs), expected by 2027-2030 », Specifies the supplier.
With these announcements and by playing the sovereignty card, Scaleway is clearly showing its muscles against other French providers, starting with Outscale, the cloud subsidiary of Dassault Systèmes, but especially OVHcloud. The similarities between the roadmaps of the two cloud providers are also disturbing.
In September, OVHcloud expanded its portfolio of AI solutions by offering the latest generations of Nvidia GPUs, up to the Tensor Core H100 and A100 GPUs. A few months earlier, the first European cloud provider acquired its first quantum machine, MosaiQ designed by… Quellela.
Photo credit: Alexis Anice