NVIDIA AI Cloud Expands Globally to Meet AI Demand

What Happened?

NVIDIA has announced the expansion of its AI Cloud ecosystem to six continents, incorporating new regional partners such as Cassava (Africa) and Claro (South America). This ecosystem, combining compute acceleration, networking, and AI software, is designed to meet the growing demand for inference, training, and agentic AI applications. According to NVIDIA's official blog, existing partners like CoreWeave, Firmus, IREN, Nebius, and Nscale are also expanding capacity to serve enterprises, startups, nations, AI labs, and developers. The geographic expansion now includes partners in North America, South America, Europe, Africa, Asia, and Oceania, marking a milestone in the decentralization of AI infrastructure.

Why Is This Important?

The expansion addresses the critical need for AI infrastructure close to users and data, reducing latency and meeting digital sovereignty requirements. Jensen Huang, CEO of NVIDIA, emphasized that "every enterprise and every country needs AI factory infrastructure to turn data into intelligence." This move accelerates AI adoption in emerging regions where access to compute capacity was limited. For example, in Africa, Cassava will bring clusters of NVIDIA H100 and H200 GPUs, enabling local startups to train models without relying on overseas data centers. In South America, Claro will offer low-latency inference services for real-time applications like chatbots and video analytics. Compared to past events, such as AWS's expansion to local regions in 2016, NVIDIA is betting on a specialized partner model rather than building its own cloud, reducing costs and speeding up deployment.

According to the blog, NVIDIA's AI Clouds are "co-designed with NVIDIA's full-stack AI infrastructure" to meet the explosive demand for tokens behind the most popular AI applications. The combination of compute acceleration, networking (like NVLink and Spectrum-X), and software (like NeMo and Triton Inference Server) allows partners to offer the best performance per watt and the lowest cost per token. This is crucial for agentic AI applications that require real-time inference with low latency. Additionally, the expansion strengthens NVIDIA's position against competitors like AMD and Intel, which do not yet have such a broad and geographically diverse ecosystem.

Consequences and Outlook

The expansion will have several far-reaching consequences:

• Increased access to compute capacity: Local startups and enterprises in regions like Africa and South America can now train and run AI models without importing foreign cloud services. This will drive regional innovation, especially in sectors like agriculture, healthcare, and finance.
• Reduced dependence on concentrated data centers: Currently, over 60% of AI compute capacity is in the US and Europe. This expansion decentralizes infrastructure, reducing geopolitical risks and latency. For example, a user in Nairobi can get inference with less than 10 ms latency, instead of 100+ ms to a data center in Virginia.
• Strengthening sovereign AI programs: Countries like India, Japan, and Singapore already have sovereign AI initiatives. NVIDIA can now support governments in Africa and South America to build their own AI factories, ensuring sensitive data stays within the country. This is key for sectors like defense, public health, and government administration.
• Increased competition in the cloud market: Hyperscalers (AWS, Azure, GCP) currently dominate the AI infrastructure market. However, NVIDIA's AI Clouds offer a specialized alternative with better performance for AI workloads. This could pressure hyperscalers to lower prices or improve their GPU offerings. According to analysts, the cost per hour of an H100 GPU on an NVIDIA AI Cloud can be up to 20% lower than on AWS.
• Regulatory and sustainability challenges: Installing high-energy data centers in regions with weak electrical infrastructure could create tensions. For example, in South Africa, where power outages are frequent, Cassava will need to invest in renewable generation and storage. Additionally, data regulations in Africa vary by country, requiring local compliance. NVIDIA and its partners must address these challenges to avoid delays.

Compared to Google's cloud expansion to regions like Chile in 2021, NVIDIA does not offer direct services but enables local partners. This reduces NVIDIA's capital investment but also limits its control over service quality. However, the partner model has proven successful in other segments, such as gaming with custom GPUs from ASUS and MSI.

What Readers Should Know

The NVIDIA AI Cloud ecosystem is not a direct NVIDIA service but a network of partners using its technology. End users should evaluate providers based on their specific workload: training, fine-tuning, inference, or agentic AI. For example, for large model training, a partner with H100 clusters interconnected via NVLink, like CoreWeave, is recommended. For real-time inference, a partner with low-latency servers like Claro in Brazil is more suitable. The geographic expansion offers alternatives to hyperscalers, especially for applications requiring low latency (like autonomous vehicles or voice assistants) or local regulatory compliance (like GDPR in Europe or the Data Protection Act in South Africa).

Additionally, NVIDIA has detailed that the AI Clouds are optimized for "the best economics: the lowest token cost and the best performance per watt," making them attractive for startups with tight budgets. However, users should consider that not all partners offer the same level of service; some focus on training, others on inference. Therefore, it is advisable to test multiple providers before committing. Finally, NVIDIA's expansion could accelerate AI adoption in sectors like healthcare (diagnostic imaging in Africa) and agriculture (crop optimization in South America), where low latency and data sovereignty are critical.

"Every enterprise and every country needs AI factory infrastructure to turn data into intelligence," said Jensen Huang.