Skip to content 
The AI Infrastructure Race
The executive virtual summit for the leaders building the future of AI infrastructure.
About the summit
The AI infrastructure race is usually told as a story about capital and chips. This summit tells the real one. Across five executive panels, senior leaders from colocation, hyperscale, neo-cloud, investment and policy confront the constraints that actually decide the build-out — power, land, water, cooling and public consent — and the risks few will name on a main stage. The through-line is simple: the race won’t be won by whoever raises the most capital, but by whoever can secure what money alone can’t buy. Join 500+ decision-makers for the candid, data-led briefing on what’s really building the AI economy.Who should attend
CEOs & founders · colocation providers · hyperscalers · data-centre operators · AI infrastructure & neo-cloud operators · infrastructure investors, private equity & sovereign wealth funds · utilities · government & policy leaders · enterprise CIOs & CTOs · infrastructure developers.Event Agenda
The build-out of AI infrastructure has moved from a technology story to an economic and geopolitical one. In under three years, the data centre has gone from a cost line inside a cloud business to a category that is drawing sovereign capital, reshaping electricity markets and appearing in national industrial strategy. The scale of committed investment is now large enough that its constraints — not its ambitions — are what will determine how much actually gets built.
This opening address frames the day. It sets out the four forces the summit will examine in turn — capital, power, public consent and physical design — and makes the case that the winners of the next phase will be decided less by who can raise the most money and more by who can secure the inputs that money cannot quickly buy: grid connections, land, water and the trust of the communities that host these facilities.
What this session answers
- How large is the gap between announced AI-infrastructure investment and what can realistically be delivered this decade?
- Which constraints — capital, power, land, or public consent — are now the binding ones?
- Why has infrastructure, rather than models or chips, become the strategic bottleneck for AI?
Before the panels begin, Compute Forecast Research presents a short, data-led briefing that establishes the factual baseline for the day. Rather than opinion, this segment puts the numbers on the table: where capital is flowing, which campus announcements are the largest, how GPU capacity and data-centre M&A are trending, and what is happening in power markets and sovereign-AI programmes.
The aim is to give every attendee — operator, investor, vendor or policymaker — the same current picture, so the discussions that follow argue from shared facts rather than competing headlines.
What this session answers
- What do the latest investment, capacity and deal figures actually show?
- Which regions and segments are accelerating, and which are cooling?
- What signals separate durable demand from speculative announcement?
Governments, hyperscalers, infrastructure funds, private equity and colocation providers are committing unprecedented capital to secure AI compute. But ownership of that capacity — and the risk attached to it — is being decided now, often on assumptions that have not yet been tested by a full cycle. The economics of an AI data centre differ from those of a traditional facility: the largest cost sits in rapidly-evolving hardware, demand is concentrated among a small number of tenants, and a meaningful share of announced projects may never be built.
This panel examines who ends up owning the infrastructure behind the AI economy, and who carries the exposure if demand softens or if the assumptions behind today’s valuations prove optimistic. It looks past headline investment totals to the questions that determine returns: the real useful life of GPU assets, the difference between reported capacity and actual utilisation, and the durability of the contracts underpinning new build.
The discussion is intended for a room of investors, operators and finance leaders, and will favour specifics — numbers, contract structures and worked examples — over generalities.
What this session answers
- How much of the capital being announced is genuinely financeable, and how much is signalling?
- What useful life are investors underwriting for GPU assets, and how does a shorter life change the model?
- How exposed are operators to take-or-pay power and long-dated commitments if demand shifts?
- Why does utilisation, not installed capacity, increasingly determine whether these assets perform?
- Where is capital most likely to flow next — and which positions are most at risk of being stranded?
Access to electricity has become the primary constraint on AI infrastructure growth. In many markets the limiting factor is no longer whether power can be generated, but whether a site can be connected to the grid within a workable timeframe — and whether the equipment needed to do so can be procured at all. Interconnection queues now run for years in some regions, and lead times for transformers and high-voltage switchgear have extended well beyond historical norms.
This panel moves the power conversation past headline megawatts to the practical realities that set delivery dates: interconnection timelines, equipment supply, and the growing question of how utilities manage the sharp, variable load profile of large AI training clusters. It also examines the options operators are exploring to bridge the gap — renewable power purchase agreements, on-site and behind-the-meter generation, and longer-horizon bets on nuclear and small modular reactors — and asks, honestly, which of these are viable this decade rather than in principle.
The session is aimed at operators, utilities, power developers and the investors funding them, and will focus on what is actually deliverable rather than what has been announced. Alongside the global picture, the panel examines India specifically — how grid availability, open access and captive power are shaping the country’s gigawatt-scale AI ambitions.
What this session answers
- What is really setting delivery dates today — generation, interconnection queues, or equipment lead times?
- How are utilities responding to the volatility and scale of AI training loads?
- Which bridging options — PPAs, on-site generation, nuclear, SMRs — are realistic this decade, and which are not yet?
- How wide is the gap between announced projects and those that will actually be powered?
- Who should bear the cost of the grid upgrades the build-out requires?
In this interactive segment, the audience answers the summit’s central questions in real time, with results displayed instantly and discussed live by the panellists. It is designed to surface where a senior, cross-industry room actually stands on the issues the day has raised — and to test whether the experts on stage agree with the people building and funding this infrastructure.
The questions span the biggest bottleneck to AI growth, whether governments should treat AI infrastructure as nationally significant, which region will lead investment, whether enterprises will keep building their own capacity, and whether liquid cooling is becoming unavoidable. Every response feeds the Executive Sentiment Report published after the event.
What this session answers
- Where does a senior industry audience actually stand on the day’s key questions?
- Do the experts on stage agree with the room — and where do they diverge?
- Which issues show consensus, and which remain genuinely contested?
As AI drives a rapid wave of data-centre development, local opposition has emerged as one of the industry’s most significant and least-discussed constraints. Concerns about power and water use, land, tax incentives and the actual local benefit of these facilities are increasingly shaping planning decisions, delaying projects and influencing policy. The problem is compounded by a growing number of proposed facilities that exist only on paper, which creates confusion and erodes trust among local stakeholders.
This panel examines what a credible community-engagement approach looks like in the age of AI infrastructure. It moves beyond transactional outreach to ask how developers build genuine, durable trust; what the economic benefits of a data centre actually amount to when measured rigorously; and how communities and policymakers can distinguish speculative proposals from credible long-term investment. It also considers how water availability, planning reform and data-sovereignty rules are reshaping where facilities can realistically be built.
The session is intended for operators, developers, policymakers and economic-development stakeholders, and treats community consent as a serious commercial and planning issue rather than a public-relations exercise. The session also looks at how these dynamics are playing out in India, the world’s fastest-growing data-centre market, where state-level policy, land, water and incentives increasingly determine where AI infrastructure gets built.
What this session answers
- Where has community opposition genuinely delayed or stopped projects, and why?
- What do the economic benefits of a data centre look like when measured rigorously?
- How should communities distinguish speculative proposals from credible long-term investment?
- How binding are water availability and environmental compliance becoming as constraints?
- What can operators and policymakers learn from both successful and failed projects?
The power density of AI hardware has risen to the point where cooling is no longer an engineering detail but a strategic decision that shapes facility design, capital planning and operating economics. Liquid and immersion cooling are increasingly presented as inevitable, but the reality for most operators is more complex: facilities are hybrid, retrofits are costly, and the true total cost of ownership is often obscured by vendor positioning.
This panel examines the honest economics of cooling AI infrastructure at scale. It looks at where liquid cooling genuinely pays off and where air still competes; the practical realities of retrofitting existing sites versus building greenfield; and the less-discussed questions of leak risk, insurance, coolant supply and the trade-off between energy efficiency (PUE) and water use (WUE). It also considers whether the waste heat these facilities produce can be turned from a disposal problem into a monetisable asset.
Aimed at operators, engineers and the vendors and investors around them, the session prioritises real numbers and design trade-offs over product claims.
What this session answers
- At real density, what does liquid cooling actually cost once retrofit is included — and where does air still win?
- How should operators weigh retrofit against greenfield for AI-ready facilities?
- What are the practical realities of leak risk, insurance and coolant supply at scale?
- How should the trade-off between PUE and water use (WUE) be managed?
- Can waste-heat reuse become a genuine source of value rather than a cost?
The summit closes with a live debate on the question underneath every session that precedes it: who will actually own and control the majority of AI compute capacity, and what role private capital plays in deciding it. Three models are competing — the hyperscalers building at unprecedented scale, the colocation providers positioning to capture enterprise AI demand, and the sovereign-AI programmes seeking national control of compute — and each faces a different version of the constraints the day has examined.
In a format built for genuine disagreement rather than presentation, four leaders make their case and are pressed on the weaknesses of their position. The session ends with a live audience vote on who will own the majority of AI capacity by 2035, closing the loop on the summit’s central thesis.
The debate is deliberately unresolved going in; the value is in hearing credible, opposing views tested in real time before a senior audience.
What this session answers
- Will hyperscalers keep their lead, or is it structurally temporary?
- Can colocation providers capture enterprise AI, or will that demand default to the cloud?
- Will sovereign AI reshape ownership, or run aground on the power constraint?
- What role will private capital play in deciding the winners?
- Who controls AI infrastructure by 2035 — and what does the room think?
To close the main programme, a single senior investor sits down with the editor for a candid conversation about capital. Having heard the day’s discussion of constraints, the focus turns to allocation: where experienced capital is deploying as the easy money thins, what it is quietly avoiding, and which parts of the market are most over- and under-funded.
The conversation is intended to be direct and specific — a considered read on the next phase of AI-infrastructure investment from someone deploying into it, rather than a promotional set piece.
What this session answers
- After a day on constraints, what changes about how you would deploy capital?
- What is the most over-funded and the most under-funded part of this market?
- What gets written down first if the cycle turns?
- Which infrastructure bets are most likely to survive a correction?
What You'll Take Away
- The Executive Sentiment Report — key insights from every panel, executive quotes, market-intelligence highlights and the live audience-poll results, plus AI infrastructure predictions for 2030.
- The AI Infrastructure Leadership Index — Compute Forecast’s annual scorecard ranking AI-ready regions on power readiness, policy maturity, community acceptance, investment attractiveness and cooling innovation.
JS
James Sutherland
MW
Dr. Marcus Webb
RP
Riya Patel
TK
Tom Keller
Register
Sponsorship Tiers
- Names the summit
- Opens and closes it
- Full lead list
- Market Intelligence Partner- owns the opening Market Intelligence Brief
- Capital & Ownership Partner — Panel 1: Who Will Own the Next Generation of Compute?
- Power & Energy Partner — Panel 2: Power Is the New Oil
- Policy & Communities Partner — Panel 3: Can Communities & Policy Keep Pace?
- Cooling & Density Partner — Panel 4: Cooling, Density & the Future of AI Facilities
- Closing Debate Partner — Panel 5: Hyperscalers vs Colocation vs Sovereign AI
- Leadership Index Partner (£6,000) — co-brands the annual AI Infrastructure Leadership Index
- Executive Sentiment Report Partner (£5,000) — co-brands the live-vote report (The Room Decides)
Don't Miss The AI Infrastructure Race
Thursday, September 24, 2026 · 10:00 AM ET · Free to attend
