Nvidia GTC Berlin Preview: Rubin Platform, AI Agents and European Sovereign Compute Take Center Stage

Europe is expected to be the most prominent focus of this GTC. Jensen Huang previously stated that Nvidia will announce approximately $2 billion in new investments or partnership projects in Europe. Combined with the roughly $8 billion invested over the past year, the company's total investments and commitments in Europe could reach nearly $10 billion.

Nvidia has already invested in French healthtech company Doctolib, robotics firm Genesis AI, enterprise software company Pigment, and materials R&D platform Entalpic, while expanding support for AI, biotech, and quantum computing startups in the UK. These investments cultivate developers using Nvidia's platforms while also helping expand European demand for GPUs, networking equipment, and AI cloud services.

Germany is likely to be the focus of the next phase of investment. The German government is expected to advance industrial AI cooperation with Nvidia, focusing on areas including Physical AI tools required for robotics and autonomous driving training, as well as large AI models for manufacturing. German Chancellor Merz is scheduled to meet with Jensen Huang on Monday evening, and the collaboration may involve German companies, research institutions, and other technology partners.

Nvidia's strongest fundamental driver currently remains its data center business. In the second quarter of fiscal 2027, the company's total revenue reached $96.2 billion, up 106% year-over-year; of that, data center revenue accounted for $89.0 billion, up 117% year-over-year and representing over 92% of total revenue.

Blackwell remains in the large-scale revenue realization phase, while Vera Rubin carries the mission of driving the next platform upgrade. Unlike in the past when the emphasis was primarily on single-GPU performance, Nvidia is integrating CPUs, GPUs, networking, storage, software, and data center design into complete 'AI factories'.

This shift is related to the migration of AI workloads from training to inference. While model training places greater emphasis on cluster scale and peak performance, large-scale inference requires balancing latency, throughput, energy efficiency, and cost per token. Ultimately, customers are no longer purchasing just GPUs, but a computing system that can reduce long-term operational costs.

Therefore, the most noteworthy aspect regarding Rubin at this GTC is not parameter upgrades, but the actual pace of deployment. The market needs to understand when major cloud providers and enterprise customers will scale up purchases, how much the new platform can reduce inference costs, and whether existing Blackwell clusters will impact customers' willingness to upgrade.

AI agents were another core topic at this GTC. Traditional chatbots primarily handle one-off Q&As, whereas agents need to continuously invoke models, retrieve information, use tools, run code, and execute multi-step tasks. Even if the cost per single inference declines, continuously running agents could still significantly increase overall compute volume.

This is also why Nvidia has tightly linked Rubin with agentic AI. The company hopes to lower the technical barrier for enterprises deploying agents by providing chips, interconnect devices, inference software, and development tools, while expanding compute demand from a small number of AI labs to scenarios such as customer service, software development, finance, healthcare, and enterprise operations.

However, growth in compute consumption does not necessarily equal an increase in commercial value. If agents remain stuck in the pilot stage for a long time, enterprises may find it difficult to justify their return on investment and will not infinitely expand their data center budgets. Therefore, the market needs to see real deployment scales, sustained call volumes, and efficiency gains achieved by customers using agents, rather than just new demos and partnership lists.

Physical AI and robotics represent a longer-term growth opportunity. Nvidia hopes to apply models, simulation platforms, and edge computing technology to robotics, autonomous driving, and smart manufacturing. Germany boasts a massive industrial ecosystem, making Berlin a suitable location to promote this strategy.

However, the commercialization cycle for robotics and autonomous driving is significantly longer than that for data center chips. Product reliability, regulatory approval, and customer validation could all affect the pace of revenue realization, making it difficult for these segments to replace data centers as Nvidia's primary growth driver in the short term.

This content was translated using AI and reviewed for clarity. It is for informational purposes only.