

Demand for artificial intelligence (AI), government intervention and data centre connectivity have been three key areas of development for the data centre industry in 2024 – in this article we explore the key developments of each and launch our 2025 predictions for each of the three key areas.
Trend #1: GenAI will be a key driver of enterprise data centre demand
Generative AI (GenAI) is here to stay. While 2023 was the year that ChatGPT brought genAI to curious consumers, 2024 was the year of enterprise testing and integration, with 1.3 million developers now having integrated OpenAI’s developer tools into software solutions in the US alone. In addition, it’s no longer just ChatGPT-4 leading the way, with 2024 bringing various proprietary and open source competitor models to market from providers predominantly based in the US, including OpenAI’s GPT-4o, Anthropic’s Claude Sonnet 3.5, Google’s Gemini 1.5, Meta’s Llama 3.2, x.ai’s Grok 2, and Amazon’s Nova. Europe is lagging behind significantly in genAI commercialisation with just one feasible competitor – France’s Mistral.
While consumers and SMBs can realise significant benefits from out-of-the-box genAI usage and basic integrations with pre-trained models, larger enterprises and service providers are evaluating the best way to serve bespoke use cases across specific languages, verticals and use cases. Companies need to be careful not over or under investing in this technology, considering the fast moving landscape. Bloomberg was an early mover on customised model training, for instance, announcing in 2023 that its BloombergGPT LLM was trained using 1.3 million hours of GPU time. However, this has since been proven to be a mis-step, with subsequent iterations of ChatGPT proving more effective in terms of accuracy.
The optimal enterprise adoption pathway for genAI for now remains unclear, and the technology still lacks a single killer use case which can itself deliver a positive return on investment. However, the holistic benefits are clear to see even before we have commercial agentic AI in live deployments. Companies across the data centre value chain are betting big on mass adoption of genAI and evaluating downstream impacts on their specific offerings, for example in ever more dense racks and cooling innovation. Nvidia gave a clear indication of this direction of travel by releasing a 120kW rack back in March 2024, specialised for model training and inferencing.
In 2025, market leading data centre operators will put an increasing emphasis on customer engagement and demand forecasting to de-risk the potentially disruptive impact of genAI adoption on infrastructure consumption patterns, working with customers to better understand consumer and enterprise genAI adoption, and the subsequent impacts on infrastructure consumption. They will seek to answer questions such as:
• In what circumstances will enterprises and service providers require denser racks for model training (100kW+) and inferencing (50kW+)? How does this vary by factors such as company size and vertical, and what are the downstream impacts on other data centre capabilities such as cooling?
• Will SaaS solutions shift further towards public cloud hosting to leverage predominantly hyperscaler-driven market leading LLMs?
• How do AI-related drivers of enterprise infrastructure consumption interact with parallel market trends such as cloud repatriation and sustainability?
• What is the impact of enterprise AI adoption on WAN and LAN traffic patterns?
Determining their role in the AI value chain, articulating this to customers and demonstrating their commitment to future AI enablement will be crucial for data centres of all sizes in 2025.
Trend #2: Regulatory outlook differs by region and type of project
Greenfield data centre projects faced a range of regulatory challenges and other limiting factors in 2024, with power and land shortages continuing to impact new builds in particular in the largest European data centre markets. Examples of how this has impacted data centre construction decisions across FLAP-D markets in 2024 include:
• Power continues to be a limiting factor in Dublin, with Google the latest company to be denied planning permission in August 2024 due to power supply constraints in the Dublin area. This is with the against a backdrop of EirGrid’s de facto moratorium on new data centres until 2028.
• Space was seen as the limiting factor for Amsterdam, who are still feeling the effects of its 2019-20 moratorium across two districts of the Dutch capital. The Dutch Datacenter Association (DDA) published its State of Dutch Data Centers 2024 report in June, noting that in recent years “growth in the Dutch colocation market had slowed significantly”, demonstrating the lasting reputational impacts which can arise from stringent data centre regulation.
The outlook has been brighter in the UK, with the change in government leading to a re-categorisation of data centres as critical national infrastructure, the benefits of which are already starting to be seen in the reversal of a decision to reject an application by Affinius Capital to build a 700,000 sqft site in Buckinghamshire.
In a similar vein to the UK, several other countries across the globe are actively designing and implementing policy frameworks to encourage data centre construction, in an attempt to compete for national and regional customers, who themselves are rapidly following more mature digital economies through their digital transformation. Nowhere is this more true than in the Middle East, where countries are seeking to gain ground on the regional leader, Dubai:
• Saudi Arabia adopted new regulations in January 2024 which seek to encourage adoption of data centre services by enterprises, with certifications based on security and service quality designed to reassure local enterprises and other prospective data centre customers that they can trust local and global providers alike. Hyperscalers are seeking to build quickly to take advantage of projected medium-term demand, with Google Cloud launching a region in Dammam in 2023, and AWS and Microsoft Azure seeking to follow in 2026 in Riyadh and Dammam respectively.
• Qatar is seeking to strengthen its data centre industry through investment in Ooredoo’s data centre business. Ooredoo, Qatar’s incumbent communication service provider (CSP), announced an additional $552 million of funding for its data centre business, with funding for this, the largest transaction ever in Qatar’s tech sector, led by the Qatari National Bank. With Microsoft Azure (2022) and Google Cloud (2023) having launched in Doha in recent years, this investment signals a desire to provide national competitors to the hyperscalers. Ooredoo also operates data centres in markets in the region in which hyperscalers are not present, namely Kuwait, Oman, Iraq and Tunisia, and this investment positions them well to be a key player in future digital transformation in these economies.
In 2025, we expect to see a continuation of regulatory and financial stimulus in markets with a discrepancy between data centre capacity and digital transformation velocity, as governments seek to reduce their digital supply chain dependency on the USA and China, while encouraging and enabling local digital innovation. This will be particularly prevalent in the Middle East, where rapid digital transformation is complemented by comparably prevalent funding, space and power. India and South East Asia are also expected to be markets to watch from a regulation and construction point of view in 2025, as well as tier 2 European cities such as Milan and Copenhagen.
Trend #3: Connectivity is a key consideration as infrastructure consumption patterns change
There is widespread industry consensus that current networking equipment is not fit for purpose for future AI data centres, with even mainstream consumer media, such as the Wall Street Journal, reporting on this industry need. Companies across the ecosystem are investing in next-generation networking solutions, in a welcome change in fortunes for a telecoms sector which has had a rocky couple of years. Cisco’s chair and CEO Chuck Robbins’ first comment on its most recent earnings report was: “Cisco is off to a strong start to fiscal 2025. Our customers are investing in critical infrastructure to prepare for AI…”, demonstrating the importance of this renewed emphasis on networking for AI to one of the leading networking equipment vendors.
It’s not just traditional networking vendors such as Cisco who are investing in AI-optimised networking components, with other companies from across the ecosystem such as Meta (more to come on them later) and Nvidia making significant connectivity investments and announcements in 2024. Nvidia, most notable as a semiconductor company, now lists ‘Networking’ as one of its six core hardware product categories – the same standing as its famous GPUs. This suite is a product of its 2020 acquisition of Mellanox for $6.9 billion, at a time when Nvidia’s share price was just 5% of today’s level. Back in March, they unveiled new switches “optimised for trillion-parameter GPU computing” with ethernet and InfiniBand offerings forming “the world’s first networking platforms capable of end-to-end 800Gb/s throughput”. For more details on how AI is transforming data centre connectivity, please see our long read on the impact of AI adoption on data centre connectivity.
Changes in enterprise digital infrastructure consumption patterns, spearheaded by AI, will not just impact customer requirements of networking components, but has the potential to fundamentally shift WAN and LAN commercial models and preferred suppliers. AI is only as good as the quantity and quality of data inputs, putting pressure on bandwidth requirements for both training and inferencing. Network latency, particularly for inferencing for mission critical and/or near real time enterprise use cases, is also a key focus point. With access to training infrastructure still limited for many businesses, and the threat of another GPU shortage on the horizon, AI innovators will seek flexibility to spin up and down high reliability, high bandwidth connections to facilities with the appropriate infrastructure to support model training and inferencing. Network complexity will also increase with the projected growth in reliance on edge sites for AI use cases such as federated learning.
Another key development from 2024 was the rise of the ‘data centre-neutral carrier’, a reverse of the traditional ‘carrier-neutral data centre’. As enterprises shift towards multi-cloud and repatriate certain workloads away from public cloud in search of bespoke configurations and cost reductions, proprietary interconnect from a hyperscaler or global colocation provider (such as Equinix) becomes a less favourable solution. Increased enterprise AI adoption alongside, or integrated within, their existing software suite will lead to increased application interdependence and more diversity in what they require from data centre facilities. We expect enterprises to look towards a neutral third party provider who can guarantee best in class, data centre agnostic connectivity while enabling flexibility in workload distribution across facilities. Lightstorm are an innovator in this space, offering robotics-led automation of data centre interconnections within minutes to enable a flexible and dynamic multi-cloud and multi-facility approach – you can read more about its productised NaaS approach in our December report.
The other data centre connectivity trend which persisted in 2024 was disruption around subsea cables, most clearly demonstrated by the formation of the United Nations Submarine Cable Advisory Group, announced in November. The International Telecommunication Union (ITU) press release stated that its aim will be to “address ways to improve cable resilience by promoting best practices for governments and industry players to ensure the timely deployment and repair of submarine cables, reduce the risks of damage, and enhance the continuity of communications over the cables”. This follows several incidents of suspected sabotage in areas of geo-political tension across 2024, including to the C-Lion1 cable connecting Germany and Finland in November, as well as several cables in the Red Sea back in February, in a trend which is back in the spotlight following another suspected sabotage event near Taiwan in January 2025.
These incidents shone some light on the fragility of critical global digital infrastructure on top of the already increasing risk of natural disasters with climate change, such as the earthquake which cut off Vanuatu’s internet access for 10 days in December, as well as the critical subsea cable repair industry. Meta is one company who have been reported to consider taking this critical dependency in house, with reports emerging late in 2024 that Meta is planning to launch a $10 billion project to lay a worldwide cable network, avoiding areas of current geo-political tension.
2025 will see a revolution in enterprise connectivity requirements, as enterprises seek both a technical enabler of a flexible and AI-ready technology architecture, as well as a commercial advantage in reducing their barriers to switch data centre and cloud providers. Data centre providers should ensure they have a compelling proposition articulating how their facilities provide best-in-class connectivity as well as compute and storage offerings, while incumbent network operators from LAN all the way through to subsea cables should quickly identify opportunities to innovate and adapt to compete in an market which looks to be in for a dynamic 2025.
Are you looking for advisory services in data centres?
How must data centre connectivity adapt to AI demands
Data centre connectivity is evolving to meet AI demands. Discover how networking tech is adapting to support growing data and infrastructure needs.
This year in data centres: The global push for scale and sustainability
A recap of some of the stories that we found interesting in the data centre world in 2024.
GPU-as-a-Service: What it is, Trends and Leading Providers
As AI demand grows, GPU-as-a-Service (GPUaaS) offers scalable, cost-effective access to powerful GPUs, avoiding heavy upfront costs. Explore its trends here.