AI cloud providers Nebius and YTL AI Cloud will be early adopters of the new Nvidia Blackwell Ultra instances.

Meanwhile, also announced during Nvidia's GTC conference this week, Vultr is making the B200s generally available, Google is soon to have GB200 instances generally available, while Oracle and Cirrascale are working with Nvidia to tackle AI inferencing.

Blackwell Ultra enters the game

Both Nebius, the AI infrastructure company spun out of Yandex's European operations, and YTL AI Cloud, a subsidiary of Malaysia's YTL Power International, have revealed plans to be early adopters of Blackwell Ultra.

Nvidia announced the launch of Blackwell Ultra at its GTC event on March 18.

Blackwell Ultra chips will be included in the Nvidia GB300 NVL72 rack-scale solution and the Nvidia HGX B300 NVL16 system. The Nvidia GB300 NVL72 connects 72 Blackwell Ultra GPUs and 36 Arm-based Nvidia Grace CPUs in a rack-scale design, acting as a single massive GPU built for test-time scaling.

YTL will be offering the Blackwell Ultra platform from its YTL Green Data Center Campus, a solar-powered 50MW data center in Johor, Malaysia. The campus, currently under construction will also be home to Nvidia's GB200s via the Nvidia DGX Cloud from Q3 of this year.

YTL Power International managing director, Dato’ Seri Yeoh Seok Hong said: “We have made tremendous progress over the past year and are poised to offer powerful AI cloud computing to the region when our first Blackwell clusters come onstream, expected in July this year. Our collaboration with Nvidia means that we are able to make available the latest AI platforms and solutions to Asia, ensuring that the region continues to stay abreast of the latest technological developments as we continue to move into an increasingly AI-powered world.”

YTL and Nvidia previously announced plans to develop $4.3 billion in artificial intelligence cloud and supercomputer infrastructure in Malaysia in December 2023.

Nebius, meanwhile, has said it will give its customers access to the Nvidia GB300 NVL72-powered instances by the end of 2025.

Arkady Volozh, founder and CEO of Nebius, said: “Access to world-class AI infrastructure is the key to realizing the full potential of AI. With dedicated Nvidia Blackwell capacity available from next quarter, Nebius is giving AI innovators and enterprises everywhere access to the world’s most powerful AI compute through our AI-native multi-tenant cloud. This is the future of AI, and we’re building it today.”

“The Nvidia Blackwell platform was built to bring the power of AI to enterprises and innovators worldwide,” said Dave Salvator, director of accelerated computing products at Nvidia. “As one of the first GPU cloud providers to offer Nvidia Blackwell and Blackwell Ultra-powered instances, Nebius will enable its customers to accelerate next-generation reasoning models, AI agents, and physical AI applications.”

Nebius has also revealed that it will be an ecosystem partner for Nvidia Dynamo, an open-source inference serving framework for generative AI in large-scale distributed environments.

The company has also confirmed that its recently announced 300MW New Jersey data center will be dedicated solely to Blackwell-architecture GPUs, while a deployment of Nvidia HGX B200s will be available from the company's Kansas City data center in Q2 2025.

B200s and GB200s enter general availability

Google Cloud is bringing B200s and GB200s out of preview, and into general availability.

Following the company's announcement of its A4 and A4X VMs, which respectively include the B200s and GB200s, in February, the cloud giant revealed that the A4 is now generally available, while A4x will be available "soon."

Vultr, a privately-held cloud infrastructure company, is also jumping on the B200 bandwagon.

The company similarly revealed that it is now offering customers the Nvidia HGX B200 via its GPU cloud. Customers will have global access to the GPUs, but the rollout is starting in Vultr's North American data centers.

“We are pleased to offer early availability of the Nvidia HGX B200, providing our customers with access to Nvidia's latest advancements in accelerated computing,” said J.J. Kardwell, CEO of Vultr. “By combining Vultr’s global cloud infrastructure platform with the performance and flexibility of Nvidia Blackwell GPUs, we are empowering enterprises and AI innovators to develop and deploy AI precisely where it’s needed.”

“Achieving faster, real-time inference is becoming a critical differentiator for organizations aiming to rapidly scale their AI initiatives,” added Dave Salvator, director of accelerated computing products at Nvidia. "The addition of the Nvidia HGX B200 to the Vultr Cloud GPU lineup marks a significant milestone in our long-standing collaboration with Vultr. By making Nvidia Blackwell GPUs accessible to both innovators and enterprises, we are empowering AI pioneers worldwide to accelerate generative and agentic AI initiatives on a global scale.”

Oracle teams up with Nvidia for agentic AI inferencing

Oracle and Nvidia are integrating Nvidia's computing power and inference software with Oracle's AI infrastructure and generative AI services to speed up the development of agentic AI applications.

The integration will make Nvidia Enterprise AI - including 160+ AI tools and 100+ Nvidia NIM microservices - natively available through the OCI Console.

In addition, Oracle and Nvidia are working together on the no-code deployment of both Oracle and Nvidia AI Blueprints and on accelerating AI vector search in Oracle Database 23ai with the Nvidia cuVS library.

“Oracle has become the platform of choice for both AI training and inferencing, and this partnership enhances our ability to help customers achieve greater innovation and business results,” said Safra Catz, CEO, Oracle. “Nvidia's offerings, paired with OCI’s flexibility, scalability, performance, and security, will speed AI adoption and help customers get more value from their data.”

“Oracle and Nvidia are perfect partners for the age of reasoning—an AI and accelerated computing company working with a key player in processing much of the world’s enterprise data,” said Jensen Huang, co-founder and CEO, Nvidia. “Together, we help enterprises innovate with agentic AI to deliver amazing things for their customers and partners.”

Cirrascale Cloud Services announces inference platform with Nvidia B200s

Cirrascale Cloud Services has unveiled plans for an inference-as-a-service platform that will use Nvidia B200s and the Nvidia RTX PRO 6000 Blackwell Server Edition GPUs.

The platform, dubbed Cirrascale Inference Platform, will automatically select the best AI accelerator to adapt to changing model requirements, user demands, and workflow shifts.

It will also help balance workloads across regions to smooth out peak demand.

“Enterprise AI adoption is accelerating, and while hyperscalers are suitable for initial validation and model training, they can introduce additional complexities and costs with larger volume usage,” said Nick Pandher, vice president of product at Cirrascale Cloud Services. “With the launch of the Cirrascale Inference Platform, customers unlock access to a high-performance, seamlessly deployable cloud inference solution that intelligently optimizes their AI configurations for both efficiency and cost-effectiveness.”

The inference platform will be available across multiple US regions, and "select international regions." Early access will be offered in April, and general availability in the Summer of 2025.

This follows on from Cirrascale adding the H200s to its AI Innovation Cloud in October 2024. Cirrascale was notably one of the first companies to get access to the Nvidia H100 GPUs. The company was previously known as Verari Technologies, before being renamed in 2010. The cloud services provider is a known customer of Digital Realty data centers.