A couple of weeks ago I had the pleasure of attending Nvidia’s annual GTC2026 (Global Technology Conference) which both expanded growth opportunities and set numerous challenges for data centre operators. It’s taken me those two weeks to fully decompress from what was an AI extravaganza.

I came away with one essential phrase: Scale in Every Direction, and evidence of this was not in short supply. Over 1,000 sessions of keynotes, presentations, workshops and technology showcases were witnessed by more than 30,000 in person attendees with hundreds of thousands more joining online, with subject titles including GPUs in space.

Those present engaged with partners from hyperscalers and neoclouds, enterprise tech and hundreds of others from across the software and AI ecosystems. For the data centre industry GTC2026 brought many opportunities closer, such as inference at scale, while hardening the challenges for some around their ability to execute.

Density is destiny

Following the raft of GPU architecture roadmap announcements at GTC2025, this year’s event was less about product and more about strategy. But, details emerged of the impressive Vera Rubin platform, a vertically integrated technology stack first announced in January 2026.

This platform delivered advancements for accelerated AI MoE (Mixture of Experts) models through inference context memory storage, super-fast low power photonic networking and AI factory scalability to tens and hundreds of thousands of GPUs. 

Also announced was the Nvidia Groq 3 Language Processing Unit, or LPU. Speaking with customers about current GPU deployment requirements confirmed what Kao Data already knows about the direction for M+E infrastructure within the data hall: the move to ultra-high rack densities is real and accelerating. Liquid cooled technology can no longer be considered innovation, it is a baseline requirement for any data centre white space, and is central to our latest offering in Harlow.

 

The rise of the neocloud

The shape and make-up of GTC events and the discussions within is evolving fast. This is in no small part due to the rise of the neoclouds who are grabbing hold of the AI challenge with impressive speed, agility and determination.

For effectively early-stage companies, organisations like Nebius and Coreweave are making tremendous progress with long-term AI infrastructure supply deals agreed, investment secured from Nvidia themselves and a deployment roadmap that highlights both ambition matched with delivery.

Power is the product

On power, over the last 12 months there have been many conversations within the data centre sector about energy access as the critical constraint on AI growth ambitions. GTC2026 showed that such conversations are even more immediate and relevant than previously considered. 

GPU manufacturing is ramping up. It is no longer about silicon but rather available, scalable energy, which for data centres marks a shift in competition from chips to grids. 

 

Here today 

There was much discussion at GTC2026 about the data centre sector’s capacity readiness for the forthcoming wave of different AI technologies. 

According to The Economist, Nvidia expects to sell $1trn of AI related GPU, hardware, networking and software over the next few years. As attested to by Jensen, the need for global investment in AI and energy infrastructure is to the tune of many hundreds of billions of dollars. 

However, data centre development is becoming more complex and increasingly difficult. Data centre insiders and market watchers will recognise that the AI opportunity has led to regular announcements of large scale projects. Yet viewed through an experienced investment, development, build and operate lens, there is evidence of some froth in the market with speculative ‘booked not billing’ projects making the news. 

Might we arrive at a point where the world of AI GPUs, CPUs, networking and software meets the reality of data centre development, where some large scale projects announced to great fanfare instead turn out to be vapourware?

GTC2026 showed that the time to question existing data centre operators on proof of ability to execute for inference AI demand has already arrived. This is where Kao Data, which has ‘been there and done it’ in terms of AI design engineering and operation, has proven ability to deliver infrastructure at the speed required for new AI workloads.

Already built to suit  

At GTC2026 Nvidia’s move beyond being a maker of GPUs to being a ‘foundational AI’ company was revealed through its push into industrial software partnerships, AI for robotics and deepening strategic partnerships with autonomous vehicle makers.

Building and training large language models led to the first wave of hyperscalers and AI companies such as OpenAI, Anthropic, XAI, Microsoft, Google and Meta committing to huge GW+ AI campus investments. Kao Data is proud to have players in the LLM sector as customers. 

For Kao Data and its customers, Nvidia’s strategy shows the focus on preparing for the next AI hyperscale wave of inference workloads has ideally positioned our data centres as locations for scaling of distributed, latency sensitive workloads. 

Kao Data is also already active in the rapidly expanding AI inference ecosystem through serving companies in agentic, production, applied, reasoning AI for verticals such as financial services, fintech, life sciences, manufacturing, pharma and energy. 

The announcements from Nvidia GTC2026 served to reinforce that operators with existing AI engineered infrastructure that is ready to accommodate scale out inference workloads have a distinct advantage. We’re in a good place, and ready for the next wave of AI deployment.