Datacloud trends to watch in 2025: What's next in the digital landscape

From grappling with power constraints and navigating the AI-driven surge in demand to embracing modularity and photonic breakthroughs, 2025 promises to redefine the boundaries of digital infrastructure.

Capacity in tandem with the Datacloud events series explores the key trends set to shape the industry, outlining insights into the challenges and innovations just on the horizon.

An unsurprising pick for the first area to watch in 2025. Access to power is top of mind for data centre builders in almost every market, with the issue more acute in some than in others.

In the most recent Digital Infra Leaders’ report, 30 C-level executives and investors in the data centre and connectivity space assessed the issue as the most pressing in their in-trays, outranking challenges such as cost of finance and availability of fibre.

Nuclear power has come into play as a possible solution with Google, Amazon, Meta and Microsoft all launching various schemes to make use of new and existing reactors. This could see other large players follow suit — but the lead time for nuclear power facilities can be in the decades, and small modular reactors in a lot of cases do not meet power needs on their own.

Natural gas is an option, albeit it one that threatens sustainability compliance. Also, the recent trend of building data centres nearer to power sources as the main location factor is likely to continue, especially in the US, as this map of planned facilities from McKinsey illustrates. This applies even more so for AI-dedicated facilities that are less reliant on low latency.

AI meaning higher compute requirements is another obvious one, but the devil is in the detail — will AI demand grow exponentially, lineally, or something in between?

To date there has been no real consensus on exactly how much extra data centre capacity will be needed. To take a few examples, Digital Bridge’s CEO Marc Ganzi forecast in February 2024 that 80% of all data centre demand will soon be accounted for by AI, and Goldman Sachs estimates a 160% demand growth.

However, not every company agrees. Speaking at this year’s Capacity Europe, Telefonica’s Gustavo Carvalho said he expects AI-related demand to rise around 30% year on year, which is more or less the same growth rate he is witnessing on the rest of his company’s network infrastructure. McKinsey agrees, forecasting demand growth at this level of around 30% per year.

While there is unanimous agreement that AI demand will require either more data centres or more efficient processing within the existing ones (most likely both), it is worth keeping a close eye on exactly how compute, rack deals, and processing levels track the predictions. There is still a lot of uncertainty in the forecasts – meaning the risk of either underbuild or overbuild is high.

A more predictable development in 2025, however, will be increased densification within data centres themselves.

Driven by breakthroughs in chip capacity and efficiency, each square metre of data centre space will likely work a lot harder in 2025 than it has in previous years. The Uptime Institute’s 2024 data centre survey showed a plurality of respondents across all facility classes reporting rack power increases, with 7-9kW racks becoming more popular year on year.

However, this increase in densification is somewhat unbalanced, with hyperscaler facilities growing rack performance at a faster rate than other providers. JLL estimates that average hyperscale rack density will hit close to 50kW per rack by 2027 compared to 17kW for non-hyperscale facilities, and Uptime Institute also reports hyperscaler facilities as the driving force behind an expected improvement in PUE for 2025.

2024 was Nvidia’s year: Demand outweighs supply for its Blackwell line of next-gen hardware, stock prices soaring high, and a short stint as the world’s most valuable company — not bad for a firm that started by designing graphics cards for gaming.

Nvidia’s dominance is shown from intense demand for its computing hardware, with the likes of Oracle, SoftBank, Meta,, Google, AWS, and Elon Musk’s xAI among those desperate to get a hold of its new line of chips.

Despite overcoming a production delay that CEO Jensen Huang said was entirely Nvidia’s fault, demand for Blackwell and its latest line, the H200 is high.

But not everyone can get their hands on Nvidia hardware if demand outweighs supply.

AMD with its Instinct GPU line has sought to push itself as the leading rival to Nvidia, unveiling its MI325X and MI350X this year, touting performance capabilities it contends counter that of its biggest rival.

While it was a year to forget for Intel, the struggling chipmaker is looking to offer a more cost-effective hardware solution to fill the Nvidia-sized gap for two-thirds of the cost with its Gaudi accelerators.

Nvidia will look to retain its crown as the in-demand semiconductor player but don’t discount the myriad of rivals already snapping at its heels.

First, there was generative AI. Then foundational or to some, frontier AI. Then multimodal AI. Now, it’s the turn of agentic AI to be the next buzzword.

Agentic AI in simple terms is the concept of an AI system capable of autonomously taking actions and solving problems based on context and objectives.

It’s still an emerging area and should not be construed with so-called ‘artificial general intelligence’ or AGI. Agentic AI is rather the concept of a system performing assigned tasks on behalf of a user or another system.

 It’s a topic that’s still nascent in the AI space but it’s gaining a lot of traction: Both AMD CEO Lisa Su and Nvidia CEO Jensen Huang referenced it during speeches at events this year as among the top tech trends, while Gartner listed agentic AI as a top tech trend for 2025.

Be prepared to see agentic AI referenced in conversations around network operations such as monitoring security for detecting unusual traffic patterns, as well as personalising contact centre queries.

Modularity is quickly becoming a key focal point across the digital infrastructure landscape.

In the data centre space, modular data centres from the likes of Vertiv and Schneider Electric can provide customers with rapid deployment of additional computing resources.  

Modularity extends beyond simply the computing units to cooling solutions, like coolant distribution units, enabling operators to apply cooling solutions while making appropriate use of floor space.

The shift to modularity isn’t limited to data centres. Companies like Modular Mast Systems are revolutionising the tower industry with customisable, scalable solutions. These modular towers allow operators to scale up or down with ease, deploy systems faster, and provide landowners with confidence that installations won’t cause long-term damage.

As digital infrastructure evolves, modularity is set to be a driving force in 2025, offering adaptable, efficient solutions to meet the sector’s growing demands.

While quantum consistently appears on ‘what to watch’ lists year in year out, it’s hard to ignore its ever-increasing amount of breakthroughs.

2024 marked another year of progress for quantum, with improvements across hardware, software and error mitigation. Milestones like Europe's first Quantum data centre from IBM and Taiwan’s first full-stack superconducting quantum computer were among the notable highlights for an emerging technology area that’s slowly but surely finding its footing.

Impressive engineering feats like Google’s Willow chip helped to showcase why quantum computing will eventually revolutionise the way we process data. With Willow, Google engineers were able to crack the quantum error correction conundrum that has stumped scientists for more than 30 years. 

While quantum supremacy — the demonstration of tasks no classical computer can replicate — may soon be within reach, true quantum advantage (delivering practical, beyond-classical applications) is still on the horizon.

2025 promises continued breakthroughs in hardware, software, and error mitigation as quantum creeps closer to real-world relevance. Expect faster processing speeds, innovative research, and a steady march toward redefining computation as we know it.

Like quantum, photonics is an emerging evolution for semiconductors — and one that might just make an impact that little bit sooner.

Unlike traditional chips, which use electrons to process and transmit data, photonic hardware uses light which offers far higher speeds, lower latency, and reduced energy consumption, especially for data-intensive applications like AI and telecoms.

The photonics space is slowly emerging, but we saw some impressive projects in 2024, like one from IBM where its researchers applied fibre optics inside a data centre rack through a dedicated module that boosted bandwidth between chips as much as 80 times compared to electrical connections.

Nokia’s coherent Photonic Service Engines (PSEs), a photonics-based networking solution, secured a myriad of users in 2024, setting record speeds in deployments with customers like Viettel, OTE Group, and Aramco.

Photonic-related innovation isn’t just coming from the big names in tech, with startups like Oriole Networks, Lightmatter, and Ephos raising millions of dollars to develop next-gen chips that leverage light.

As the race to harness the power of light continues, 2025 could be the year photonic chips move from cutting-edge curiosity to a pivotal force reshaping the semiconductor landscape.

Data centres in space is one of those topics that is regularly claimed to be a few years away from reality without ever actually happening. But recent developments have brought the idea of edge data processing in space closer to reality.

HPE’s Spaceborne 2 is currently aboard the International Space Station, and its hardware is more or less what is found inside terrestrial data centres (KIOXIA 960GB RM Series SAS drives, 1,024GB NVMe drives and 30.72TB PM6 Enterprise SAS SSDs for reference) with the added benefit of running on solar power thanks to a direct current setup.

Three of the biggest challenges for data centre provisioning are power, cooling and room to locate the actual infrastructure, and in theory space-based data processing would solve all of them. Capacity looked in detail at the prospects for any development in this area in the near future — it is worth keeping an eye on.

techoraco launches Datacloud Energy & ESG Europe to tackle data centre energy challenges

What we learned at Datacloud Global Congress 2024

12 key takeaways from Capacity Europe 2024