Power Crisis & Politics: AI’s Infrastructure Reality Check

At Datacloud Congress in Cannes, enterprises from across the world came together to discuss the future of data centres and AI’s role in it.

AI Magazine spoke to Annabel Helm, Managing Director of Datacloud about orchestrating one of Europe’s largest data centre conferences, where the conversation has shifted dramatically from last year’s AI speculation to this year’s harsh infrastructure realities.

“I think it’s less nebulous – it’s gone from being conceptual to actually here,” she says.

“The numbers tell the story: a 40% jump in attendance from last year, with C-suite representation up 49%.

“Last year AI was this huge topic because there was OpenAI ChatGPT, but now all of this demand is actually there. The impact it’s having on our infrastructure is exponential.”

The data centre priority shift

AI’s impact is changing data centres and entire business models along with it.

Companies that spent decades focused on securing prime real estate and fibre connections are scrambling to solve a more fundamental problem.

“If you think back to five years ago, even two years ago, the number one requirement if you’re a data centre operator was land, then maybe land and fibre, then land, fibre and water.

“Now it’s literally power. And that’s a really interesting change.”

The power crunch has created some unlikely conference attendees, such as West Texas Gas, a company that two years ago would never have considered a data centre event, now has representatives walking the show floor.

Energy companies are suddenly finding themselves courted by an industry desperate for megawatts.

The edge of uncertainty

Behind the networking sessions and product demonstrations, a more complex technical debate is unfolding.

As AI companies move from training massive models to deploying them for real-world applications, basic questions about infrastructure remain unanswered.

“As we transition from training to inference, what does that mean in terms of edge deployment? What is the edge? Where is the edge? Where is it going to live? Is it in the hyperscale or is it on the device?” Annabel asks.

“These aren’t academic questions – they’re determining where billions in infrastructure investment will flow.”

The uncertainty reflects just how fast the industry is changing.

“If you compare how much active data centre capacity there is this time this year to last year, it’s only going to grow exponentially. So I think it’s that opportunity mixed with caution.”

That caution is warranted, as political leaders were learning what a data centre is a few years ago, they are now announcing national AI strategies.

“When I first started working in telecoms, in no possible world would I have imagined the word data centre to come out of any world leader’s mouths – and then in January, in the space of a week, we had Biden, Trump, Macron and Starmer all talking about this.”

Sovereignty in the age of learning machines

The political attention brings new complications.

Governments worldwide are grappling with what Annabel points out as “sovereign AI” – the challenge of maintaining control over AI systems that learn from data within national borders.

“The challenge is if you put data into a model, that model is set to learn and you can’t unteach something that’s already been taught like a human brain. You can’t just go forget that – and that’s why we need to be really cautious,” she explains.

This technical reality is driving policy discussions that go far beyond traditional data protection laws.

“If you’re looking at GDPR as being the starting point of governments being able to legislate to protect people’s data, sovereign AI is the next step.”

The implications are personal as well as political: “I tell my friends and family, do not put your birthday into ChatGPT, don’t put things that you don’t want other people to know about you.”

The human element

Despite orchestrating an event dedicated to AI infrastructure, Annabel maintains a grounded view of the technology’s current limitations.

Her own conference, with its carefully curated agenda and high-profile speakers, relies more on traditional methods than algorithmic assistance.

“We use AI a bit for research purposes, but what people don’t see behind every agenda is at least a hundred hours of direct conversations – you’re spending an hour with industry leaders figuring things out. Trying to weave that in is not quite where AI is.”

It’s a perspective that cuts through the hype surrounding many AI applications.

“It’s a tool and it’s a tool that we need to define and then use to help us,” she says.