Data Centre Takeaways From 'The Woodstock of AI'

San Jose, US, transformed into the epicentre of the technological universe once again, hosting the NVIDIA GTC summit in March 2026.

For data centre operators and infrastructure providers, this event signals the most significant shift in facility design, power delivery and cooling architecture in decades. Dubbed the "Woodstock of AI" by industry insiders, GTC has evolved from a developer conference into a high-stakes summit where the world's most powerful CEOs outline the future requirements for the computational backbone that will underpin the global economy.

As NVIDIA moves toward its ambitious goal of becoming the world's first US$1tn sales organisation by 2027, the implications for data centre infrastructure have moved to centre stage.

The March 2025 event focused heavily on physical AI and the transition to the Vera Rubin architecture, presenting data centre operators with unprecedented challenges in power density, cooling capacity and facility integration. For technology companies, GTC is no longer just about the chips that power AI – it shifts to the fundamental infrastructure required to make AI work at scale within modern data centre environments.

The challenges facing data centre operators define the next decade of computing. Finding ways to power and cool the massive, high-density AI infrastructure with integrated energy systems becomes critical. Developing "smart" AI systems that can reason, plan and act autonomously via agentic software platforms requires new approaches to data centre management. Testing and proving these new AI systems across huge, production-ready, global cloud networks demands infrastructure that can scale rapidly while maintaining reliability.

Liquid cooling becomes production standard

Microsoft wasted no time in asserting its position at the front of the queue for NVIDIA's latest hardware. Satya Nadella, CEO at Microsoft, confirmed that the firm is the first cloud provider to operationalise the new Vera Rubin NVL72 system. This rack-scale solution, which is 100% liquid-cooled, represents a massive leap over the previous Blackwell generation, offering up to five times the inference performance. The shift to mandatory liquid cooling represents a fundamental change in facility design and operational requirements for data centre operators.

"We're the first cloud to bring up an NVIDIA Vera Rubin NVL72 system for validation, another big step in building the next generation of AI infrastructure with NVIDIA," Satya said.

By moving these systems into a "lab" environment for immediate validation, Microsoft aims to roll out Rubin-based instances to Azure data centres within months. This collaboration ensures that Microsoft's cloud customers will be among the first to access the 36-CPU, 72-GPU configuration designed for the most demanding LLM workloads. The infrastructure requirements for these deployments set new benchmarks for power delivery and cooling capacity across the industry.

Despite industry rumours that Tesla and SpaceX might pivot entirely to in-house silicon, Elon Musk, CEO at Tesla and SpaceX, used the "Woodstock of AI" to clarify his long-term reliance on NVIDIA. While Tesla continues to develop its own processors, Musk played down the idea that they would fully replace NVIDIA's hardware in the immediate future.

"I am a huge admirer of NVIDIA and Jensen [Huang]," Elon wrote in a post on X.

"That market cap is well-deserved."

He further solidified the partnership, stating: "SpaceX AI and Tesla expect to continue ordering Nvidia chips at scale." 

For AWS, the March 2025 GTC marks a massive expansion of its 15-year partnership with NVIDIA. David Brown, Vice President of Compute and Networking at AWS, detailed a roadmap that included the deployment of more than one million NVIDIA GPUs across AWS Regions starting in 2025. This scale of deployment presents enormous challenges for data centre infrastructure, from power procurement to cooling system design. AWS positions itself as the premier destination for agentic AI, which are systems capable of autonomous reasoning and planning.

Scaling infrastructure for production AI

"AI is moving fast and, for most of our customers, the real opportunity isn't in experimenting with it – it's in running AI in production where it drives meaningful business outcomes," David said. To support this, AWS integrates the NVIDIA Inference Xfer Library to speed up data movement between nodes, alongside offering new EC2 instances powered by the RTX PRO 4500 Blackwell Server Edition. The data centre implications of running AI at this scale require entirely new approaches to facility design and energy management.

As compute density skyrockets, the conversation at GTC shifted from the GPU to the power grid.

Olivier Blum, Executive Vice President at Schneider Electric, highlighted that AI success now depends as much on energy systems as it does on silicon. Schneider worked closely with NVIDIA on the Vera Rubin NVL72 reference design to ensure that high-density compute can scale sustainably within existing and future data centre facilities.

"One of the clearest signals coming out of NVIDIA GTC this week is how quickly the industry is moving toward a more connected model for AI infrastructure," Olivier said. He emphasised that the industry moves away from fragmented tracks toward "coordinated energy-centric systems".

By using digital twins and agentic AI in operations, Olivier believes facilities can finally reduce manual risk in these increasingly complex environments. For data centre operators, this means fundamentally rethinking how power, cooling and compute are integrated at the facility level.

Adrian McDonald, President of Dell Technologies EMEA, echoed the sentiment that GTC lives up to its "Woodstock" billing. For Dell, the focus is on the practical economics of AI, where the "new currency" is defined by cost per token and tokens per watt. "The world is moving to AI Inference at pace and NVIDIA intends on leading the way," Adrian said. He noted that as AI becomes central to government and business services, "tokens as a service" is likely to become a dominant business model.

Dell continues to work as a primary infrastructure partner, helping organisations turn "power into revenue" by deploying NVIDIA's full-stack capabilities across the global enterprise. For data centre providers, this shift toward measuring efficiency in tokens per watt becomes the defining metric for facility performance in the AI era.