Five things we learned in the data centre September events season 2025

Listen Download

In September, thousands gathered under the grey skies of London for DCD Connect and on the sun-soaked coast of Antibes for Platform Global. Optimism about AI-driven growth was on display, but conversations revealed a more complex reality. Enterprises are still struggling to turn pilots into production, while neoclouds face questions over resilience and business models. Europe’s geography of compute is shifting as operators chase power and permits, while in the UK, bold announcements risk outpacing delivery. And across Europe, sustainability is no longer optional but increasingly a key standard by which projects are judged and approved.

1. Enterprise AI adoption is the critical focal point as data centres push to monetise capacity

Throughout the events, a recurring theme was that as operators seek to monetise their compute capacity, enterprise challenges in adopting AI effectively remain a critical hurdle for longer-term aspirations for AI. This was a topic issue across September’s events in the aftermath of the recent MIT report which saw 95% of enterprise fail to deliver value from their respective AI pilots. Across multiple panels, speakers described the same recurring failure pattern: organisations begin without a clear strategy, operate with weak data governance, and bolt security on at the end. The result is an endless multiplication of proofs of concept while actual production value stalls. The scarce resource for many enterprises looking to adopt AI is not silicon or data centre capacity. It is the cross-functional delivery capacity of architects, data engineers, and MLOps specialists, together with a brokered partner ecosystem of OEMs, integrators, and consultancies that can reliably take a use case from slideware to deployed service.

One of the most practical themes to emerge was the need for disciplined operating models across enterprises if the widening gap between effective AI adoption and compute investment is to be closed. Speakers argued for three priorities for enterprises. An SVP of operations at a data centre operator encouraged enterprises to build ‘T-shaped teams’: people with broad awareness across infrastructure, security and governance, paired with deep specialists who can solve complex problems in their lane. This blend helps enterprises avoid siloed missteps and accelerates time to production.

Further advice split into two practical strands. First, learn from the pilots that work. Amid all the failed experiments, there are successes, and enterprises should codify these into reference architectures and data contracts so that each new workload doesn’t start from scratch. Building a record of “what good looks like” helps establish guardrails for future deployments and avoids repeating the same mistakes.

Second, design for flexibility. The next few years will almost certainly see shifts in the landscape of model providers. Betting everything on a single vendor risks lock-in, especially if a data centre environment only offers access to one ecosystem, such as OpenAI. Several panellists urged enterprises to think modularly – to structure systems so that model providers can be swapped without re-engineering pipelines. For infrastructure operators, this translates into packaging services in composable, interoperable ways that allow customers to remain agile as the market consolidates around winners.

Equally, a cultural reset was considered as important as tooling for enterprises looking to adopt AI. Organisations were urged not to idolise the technology but to treat it with informed scepticism. A VP at a data centre operator observed, “don’t assume that AI is a ‘trained hunting dog’ always brings back the right prey (in this case the correct information); sometimes it’s a golden retriever with a random stick”. Measurement was highlighted as a discipline in itself. Instead of simply tracking number of pilots, enterprises should measure execution outcomes: time to first use case, time to production, governance throughput, and cost to performance per workload.

The conclusion was clear. The winners among enterprise adopters will not be those who spend the most but those who build disciplined delivery organisations and orchestrated partnerships that reliably convert compute into business outcomes. For data centre operators, that means focusing less on who signs the biggest lease and more on identifying tenants – especially service providers and SaaS players – whose core business ensures they will actually utilise capacity and scale into new racks.

2. Approach neoclouds with healthy scepticism

The message from the events was clear: not all neoclouds are built to last. Announcements are running well ahead of committed capital, and typical build cycles run three to five years from signature to energisation — leaving plenty of scope for risk. Compared to hyperscalers, neoclouds carry materially higher exposure: business models are unproven, while M&A valuations look stretched.

For investors and partners, the guidance was simple: treat neoclouds as a credit test, not a category bet. The focus should be on whether they actually have solid anchor customers and the ability to deliver capacity when promised. That means looking past the hype to see if power is secured at a viable price, interconnects are in place, and the management team has a track record of building on time.

On the business model itself, speakers noted that most neoclouds will only take on long-term contracts tied to their newest chips in order to depreciate capex – which generally results in low-margins. This is due to high upfront capital costs, competitive contract driven pricing and the risk that contracted rates will lag behind market prices as the technology evolves. The neocloud providers that differentiate will be those, who as tenants in third party data centres, can layer in an on-demand pool of older silicon, monetising last-generation GPUs for bursty inference workloads at competitive price points while reserving their cutting-edge hardware for contracted base load.

The implication is stark: one CEO of a data centre operator warned the audience to “expect waves of winners and losers over the next 12–36 months”. For such operators engaging with neoclouds, the takeaway is to diversify exposure, stage commitments to milestones, and structure deals around a two-track model – base-load capacity anchored by long-term contracts, complemented by flexible pools of legacy hardware that can generate incremental yield. In a market where hype is cheap but power and land are scarce, this discipline will separate sustainable partnerships from costly dead ends.

3. Europe’s data centre map is being redrawn as operators chase power and permits

One of the most striking narratives from the event was how quickly the map of prime data centre markets is shifting. The traditional equation of ‘build close to the user to cut latency’ is being rewritten. For many AI workloads, latency is less of a binding constraint, which means operators are optimising instead for power availability, permitting velocity, and interconnect density. This change is reshaping Europe’s geography of compute.

Nowhere is this clearer than in Iberia and the Mediterranean corridor, stretching from Madrid through Zaragoza and Barcelona to Marseille. Spain in particular has elevated data centres to priority project status, with local authorities fast-tracking bureaucracy in ways that measurably cut time to market. Subsea cables are reinforcing this momentum: links from the Carolinas to Spain and routes connecting Portugal to Brazil make such locations ideal connectivity hubs. Coupled with hyperscaler plans to extend availability zones beyond the traditional FLAP hubs, the region is positioning itself as a credible AI-factory corridor, or as, the CCO of a data centre operator put it: “the crossroads of Europe”.

At the same time, even the Nordics and Arctic Circle, once considered far beyond the frontier, are now “very much in play.” As one participant put it, wave three of AI build-out will be driven less by customer preference and more by structural factors, cheap, abundant power and planning certainty, with proximity to end customers a secondary concern. In this world, large-scale AI factories follow fundamentals first.

Threaded through all of these examples is a common lesson: local authority support has become a decisive factor. In Spain and elsewhere, municipalities that once fought data centres are now courting them, recognising their role in economic growth and grid modernisation. As an SVP at a major real estate company summed it up, “the top ingredient for data centre location selection – aside from power -is local authority support.” Equally important is securing support beyond government halls, as STL Partners’ research on public perception shows that community sentiment can be a decisive factor in avoiding costly data centre delays. For operators, that means winning the next wave of demand will depend as much on political and community capital and permit velocity as on technology or real estate.

4. Announcing capacity in the UK is easy, delivering capacity is harder

The UK is awash with AI investment headlines — ministerial podiums, balloons, and bold announcements from household names. Yet inside the room in London, the mood was sceptical. The consensus was that headline figures are running far ahead of committed capital and energised megawatts. Several examples illustrated the gap: much-publicised investments turned out to be aspirational rather than truly committal, while many projects were sub-50MW or routine expansions. The takeaway: watch the conversion, not the press release. Returns will be determined by delivered capacity, not by hype.

The fundamentals begin with power economics. As speakers were quick to remind the audience, UK electricity costs are estimated at around four times higher than in the US, while a significant amount of a data centre’s total cost of operation comes from power alone. Without cheaper, more predictable energy—and certainty around transmission, not just generation – grand ambitions risk being uneconomic.

This issue is compounded by planning. Permitting processes remain sluggish, often taking years to complete, and the UK also carries a ~400MW legacy estate. Unless this older capacity is upgraded or rationalised, it risks dragging down both efficiency and credibility in meeting the country’s “AI powerhouse” ambitions. One panellist, a VP at a middle-market private equity firm, memorably captured it: the sector must sort out “the bowels before the sunny AI stuff”: power, data quality, and planning discipline before chasing glossy announcements.

Against this backdrop, investors are tightening their focus. Several speakers warned that the UK risks confusing announcements with assets, and that capital will only back projects with credible pathways to revenue. As one speaker put it, “Capex is a weapon, but in the wrong hands, it can do real damage.” For the UK, that means high-profile AI pledges will count for little unless operators can de-risk projects with bankable power contracts, realistic build timelines, and management teams capable of delivery. Without that discipline, many of today’s bold promises may never translate into energised capacity – or into the returns policymakers are banking on.

5. In Europe, the sustainability impetus is as strong as ever

What came through strongly is that sustainability has moved from being a reputational lever to a hard requirement for both financing and customer procurement. Green bond eligibility is tightening, with a PUE of around 1.5 now viewed as a baseline. Several investors, including a managing director from a major British bank, indicated they are already looking for 1.4 or better, alongside stricter water usage efficiency (WUE) criteria. This matters because it shifts sustainability from the margins of project design into the centre of the investment case – projects that can’t evidence clear efficiency gains will struggle to unlock capital at scale.

On the customer side, sustainability is embedded in every RFP. European buyers in particular are now treating efficiency, renewable sourcing, and transparent reporting as entry conditions, not differentiators. In other words, sustainability is becoming the ticket to the table, not a competitive edge. As one speaker warned, the challenge will be to embrace higher standards without letting them paralyse delivery – balancing more ambitious environmental targets with the need to maintain build velocity in a tight supply-demand environment.

For operators, the implications are twofold. First, design for compliance from day one: build credible roadmaps for PUE and WUE improvements, backed by monitoring and reporting systems that will satisfy both investors and regulators. Second, don’t overlook the legacy estate. Upgrading older capacity is increasingly critical, not just for efficiency gains but also to ensure it doesn’t drag down portfolio averages and undermine credibility. In a market where ambition is everywhere but discipline is rare, those who can evidence measurable, year-on-year sustainability improvements will be the ones who secure capital, win RFPs, and deliver megawatts that actually matter.

Final thoughts

Across London and Antibes, the message was consistent: the AI build-out is real, but the road ahead is far from straightforward. The industry’s trillion-dollar trajectory will hinge not on announcements but on execution – whether enterprises can move past pilots, whether neoclouds can prove their models, whether Europe can align its power, planning and political capital with demand, and whether sustainability standards are embraced as a foundation rather than a hurdle. For data centre operators, investors and their enterprise customers, the challenge now is to combine ambition with discipline: to balance speed with rigour, and scale with credibility. hose who will succeed are the operators that back tenants able to scale, pick markets where power and permits are bankable, and meet rising sustainability bars – turning announcements into utilised megawatts

At STL Partners, we’re working with stakeholders across the ecosystem to navigate these transitions – from evaluating shifting demand patterns for corporate development to rethinking infrastructure design, and aligning commercial strategy with tomorrow’s customer requirements. For data centre operators, tenants and investors alike, the path forward will require sharper foresight, closer collaboration, and a willingness to rethink assumptions in an industry where the only constant is change.