Quick answer: DC Market Insights helps hyperscalers, cloud providers and AI and GPU-cloud platforms decide where to add capacity, how to secure the power for it and whether to build, lease or partner in each region. The work is for capacity planning, real estate, energy and strategy teams who need an outside view of regions, power timelines, rules and competition before committing capital at hyperscale.
The capital at stake has no precedent in the sector. Alphabet spent USD 91.4 billion on capital expenditure in 2025 and expects USD 175 to 185 billion in 2026, according to its fourth-quarter 2025 results. Amazon expects to invest about USD 200 billion in 2026 (Amazon Q4 2025 results), and Meta expects USD 115 to 135 billion (Meta Q4 2025 results).
Money is not the constraint. Power and time are. The International Energy Agency expects data center electricity use to rise from 415 TWh in 2024 to around 945 TWh by 2030, and warns that around 20% of planned data center projects could be at risk of delay unless grid risks are addressed. For a platform with GPUs on order, a campus that energises a year late is a year of stranded hardware and lost revenue.
DC Market Insights is the data center practice of Credence Research, a research and consulting firm founded in 2015 with 200+ analysts and consultants and 450+ consulting projects a year.
What do hyperscalers and cloud providers need from a data center adviser?
They need an outside view of where capacity can be delivered, how fast and on what terms, from a team that already tracks the regions, the grids and the competitors. Their internal teams know their own demand better than anyone. What they ask us for is the evidence on everything outside their walls.
That usually comes down to five questions:
- Where next? Which countries and metros can deliver the megawatts the roadmap needs, on the dates it needs them.
- How will it be powered? Grid timelines, on-site generation, PPAs and the rules that govern each.
- Build, lease or partner? Which route is faster and cheaper in each region, and which colocation operators can deliver at scale.
- Can the market support AI density? Whether local supply chains, operators and rules can handle liquid-cooled, high-density halls.
- What do sovereignty rules require? Where data residency, public sector cloud rules and national AI programmes shape where capacity must sit.
How large is hyperscale data center spending today?
It is measured in tens to hundreds of billions of US dollars a year per company: the three largest 2026 outlooks published by Alphabet, Amazon and Meta add up to between USD 490 and 520 billion, before Microsoft is counted.
| Company | Latest full-year capex | Next-year outlook | Notes |
|---|---|---|---|
| Amazon | USD 131.8 billion (2025) | About USD 200 billion (2026) | Purchases of property and equipment (gross) |
| Microsoft | USD 115.9 billion (FY2026, to June 2026) | Not stated in the release | Additions to property and equipment |
| Alphabet | USD 91.4 billion (2025) | USD 175 to 185 billion (2026) | Capital expenditures |
| Meta | USD 72.22 billion (2025) | USD 115 to 135 billion (2026) | Includes principal payments on finance leases |
Sources: company results releases for Q4 2025 (Amazon, Alphabet, Meta) and Q4 FY2026 (Microsoft, 29 July 2026). Definitions differ by company, so the figures are not strictly comparable.
The revenue behind this spend is growing fast. AWS net sales reached USD 128.7 billion in 2025, up 20%. Google Cloud revenue was USD 17.7 billion in the fourth quarter of 2025, up 48%. Microsoft reported Microsoft Cloud revenue of USD 59.3 billion in the quarter to June 2026, up 27%, with Azure and other cloud services up 43%. Across the whole market, the IEA puts global investment in data centers at around half a trillion US dollars in 2024, nearly double the 2022 level.
At this scale, small errors in region choice become large. A wrong assumption about grid timing in one region can hold back billions of dollars of servers. That is why our work for this client type starts with power, not demand.
What do we do for hyperscalers and cloud providers?
We run five kinds of work for this client type, each linked to one of our consulting services. Most engagements combine two or three.
Region selection and market entry
We compare countries and metros on the factors that decide whether a new region can go live on time: deliverable grid capacity, connection rules, land, fiber and subsea routes, latency to user populations, tax and incentives, and the operators already present. Each candidate is scored on the same scale so the shortlist is a fair comparison. This is our market entry and power strategy service, and it often leads into site selection for specific campuses.
Power availability and sourcing
We map when each candidate region can deliver power, and how. Grid queues have grown fastest where hyperscalers want to be. ERCOT in Texas had received more than 233 GW of large-load interconnection requests by late 2025, with data centers behind more than 70%, according to Utility Dive. In Great Britain, data centers made up over half of the more than 90 GW of demand in responses to the National Energy System Operator’s demand call for input, and 81% of all responding projects said they were open to phased, ramped or non-firm connections.
We assess grid connection, on-site generation and storage, corporate PPAs and alternative supply for each region, and date each source with a confidence grade. Where rules require on-site power, we show what it adds to cost and timeline.
Self-build versus lease
We compare building your own campus with leasing from colocation operators, region by region. The comparison covers time to capacity, the operators who can deliver at your scale and density, wholesale pricing and lease terms, and the power rights each route brings. In tight markets the answer can change quickly: CBRE reported primary-market vacancy of 6.6% in Q1 2025, down 2.1 percentage points in a year. Our competitive intelligence work benchmarks the operators who could serve you.
AI density and supply chain readiness
AI training and inference push rack densities far beyond what older halls were designed for. We assess whether a region’s operators, contractors and equipment supply chains can deliver liquid-cooled, high-density capacity on time, and what lead times look like for the power and cooling equipment that matters most. The IEA notes that wait times for transformers and cables have doubled in the past three years. Our upstream models for power, cooling and racks, including the OCP Rack Market, give the baseline.
Sovereignty, data residency and policy
We track the rules that decide where data and compute must sit: data residency and localisation requirements, public sector cloud rules, and national and regional AI programmes. In Europe, the Commission launched InvestAI in February 2025 to mobilise EUR 200 billion for AI, including a EUR 20 billion fund for AI gigafactories. Programmes like this create demand for in-region capacity and new partners. Our policy advisory team works with the governments on the other side of these programmes.
Why does power decide where hyperscale capacity goes?
Because a region with land, fiber and demand but no firm power date cannot host a campus on a timetable a GPU order can wait for. In many core markets, the grid now sets the pace.
Two markets show how fast the terms can change:
- Ireland. Data centers used 22% of all metered electricity in Ireland in 2024, up from 5% in 2015, according to the Central Statistics Office. The regulator’s December 2025 connection policy, as summarised by DLA Piper, requires new data center applicants to provide on-site or nearby dispatchable generation matching 100% of their maximum import capacity on a de-rated basis, and renewable electricity generated in Ireland equal to at least 80% of annual demand, with a six-year glide path.
- PJM (eastern United States). PJM now forecasts summer peak demand to grow 3.6% a year over the next ten years, with data centers the main driver, against 0.3% a year in its 2021 forecast. It now asks for firmer commitments before large loads are counted in its near-term forecast.
| Power question | Why it matters to a hyperscaler | What we deliver |
|---|---|---|
| When will the grid connect? | Sets the date revenue can start | Dated power curve with confidence grade |
| Firm or non-firm? | Decides which workloads the site can host | Comparison of connection options by region |
| On-site generation required? | Adds cost, permits and fuel risk | Cost and timeline of compliant on-site power |
| Can the power be called clean? | Affects carbon commitments | Review of PPA and certificate rules by market |
| Who else is in the queue? | Competing load can push your date back | Queue and pipeline map by grid zone |
The IEA expects renewables to meet half of the global growth in data center electricity demand to 2030, with natural gas expanding by 175 TWh. For most platforms that means a mixed supply strategy, and one that differs by region.
Who is this work for?
It is for the teams inside cloud and AI platforms that have to commit capacity before they can be sure of power, and for the people who advise and finance them.
- Global hyperscalers planning new regions, expansions and availability zones.
- Regional and sovereign cloud providers competing with global platforms on residency and local presence.
- AI and GPU-cloud platforms that need high-density capacity fast, often through leases and partnerships rather than self-build.
- Colocation operators bidding for hyperscale leases, who need to see the market as their customer sees it. See our page for operators.
- Investors and lenders backing hyperscale-leased campuses. See our page for investors.
- Equipment vendors selling power, cooling and rack systems into hyperscale builds. See our page for vendors.
How does an engagement with a hyperscaler work?
Every engagement runs in five steps, and each step produces something your team can use before the next one starts.
- Scope the decision. A call to agree the regions in play, the capacity and density you need, your target dates, the routes you will consider and the cases to model. You receive a written scope and a quote.
- Screen the regions. Our analysts score candidate markets on power, demand, supply, price, latency, rules and operator presence, starting from our published models.
- Test power and rules in depth. For the shortlist, we review grid operator and regulator publications, queue data, planning records and policy, and speak to utilities, operators, brokers and equipment suppliers. Every key assumption is checked against at least one independent source.
- Build the plan. We build the region ranking, the power plan and the build versus lease comparison, with base, upside and downside cases and every figure traced to its input.
- Present and stand behind it. A findings session with your capacity, energy and strategy teams, and with leadership if needed. Follow-up questions are answered after delivery.
What you receive
| Deliverable | Format | Used for |
|---|---|---|
| Region screening memo | Short PDF | Agreeing the shortlist early |
| Region and power strategy report | PDF with charts and maps | Capacity planning and leadership review |
| Region scorecard and power model | Excel with open formulas | Re-weighting criteria and testing cases |
| Build versus lease comparison | Excel and PDF | Choosing the route in each region |
| Operator and pipeline data pack | Excel | Supplier selection and lease negotiation |
| Findings session | Video call or on site | Questions from your team |
DC Exclusive: what does our model library add for a hyperscaler?
It gives every candidate region a tested baseline from the first day: 519 published market models, each with history from 2020, a 2025 base year and a forecast to 2035, in global, regional and country editions. Your team spends its time on the decision, not on rebuilding the market from nothing.
The library is organised in five layers, and hyperscale work draws on all of them:
| Layer | What it covers | How it helps a hyperscaler |
|---|---|---|
| DC Core | Colocation, hyperscale, edge and modular facilities | Supply, vacancy and lease options by region |
| Upstream | Power, cooling, construction and equipment | Build cost, lead times, AI density readiness |
| Downstream | Cloud, AI compute, interconnection and hosting | Competing cloud and GPU capacity by region |
| Ecosystem | DCIM, security, maintenance and finance | Operating partners and service costs |
| Extended | Adjacent markets in their data center form | Second-order demand around a new region |
For example, our OCP Rack Market report values the market at USD 1.63 billion in 2025, rising to USD 13.34 billion by 2035, a 23.4% CAGR, which tracks the open hardware that hyperscale builds rely on. Our Iberia Data Center Market report values Spain and Portugal at USD 11.53 billion in 2025, a baseline for platforms weighing southern Europe.
Three proprietary datasets are in development and will feed future projects: a Capacity Tracker (live and pipeline MW by country, city and operator), a Colocation Price Index (USD per kW per month by market) and a Deal Tracker (M&A, funding, land and power deals).
How do we build the numbers?
We build every market figure two ways and only use it when the two agree:
- Bottom-up: installed and pipeline capacity, operator revenue, contract values and equipment shipments, counted market by market.
- Top-down: total cloud and IT spend, electricity use and construction spend, and the share that data centers take.
- Reconcile: where the two differ, we find the input that is wrong and fix it, rather than averaging the gap away.
- Power cross-check: demand in each region is tested against what the grid operator says it can connect.
- Time frame: history from 2020, a 2025 base year and a forecast to 2035.
Where your own demand data is better than ours, we use it. A project exists to get your decision right, and a better input gives you a better answer.
Who works with hyperscalers and cloud providers?
A named senior team leads every engagement, with our wider bench of analysts behind them.
- Amit Jain, Senior Consultant, ICT & Emerging Technologies. Amit heads our ICT consulting and syndicated research practice and has more than 20 years in technology market research. He leads our work on data center infrastructure, cloud and telecom, including market sizing, vendor benchmarking and commercial due diligence.
- Satyabrat Rajawat, Senior Research Analyst, Strategic Advisory. Satyabrat covers power, cooling and rack equipment, the upstream markets that decide build cost, lead times and AI density readiness. He has contributed to more than 40 syndicated market studies.
- Deepti Agrawal, Senior Editor, Research. Deepti is the final quality gate for our reports and consulting work. She checks models, growth rates and company profiles, and owns our house style.
Which DC Market Insights research supports hyperscale decisions?
Every engagement links back to the published reports for the regions and equipment involved. Useful starting points:
- OCP Rack Market for hyperscale rack and open hardware demand
- Iberia Data Center Market for demand in Spain and Portugal
- UK Data Center Thermal Management Market for cooling spend, with London and Slough holding more than 65% of the UK market
- The full data center research library, organised in five layers
- Our consulting overview and related services: market entry and power strategy, site selection, market sizing, competitive intelligence and commercial due diligence
Frequently asked questions
What does a hyperscaler engagement cost?
Fees are quoted per engagement and depend on the number of regions, the depth of power and regulatory review and the deadline; you receive a written quote after the scoping call.
We have large internal teams. What do you add?
An outside view, built from models and data we already hold. Internal teams know their own demand and designs. We bring independent evidence on regions, grids, rules, operators and competing capacity, checked against people who work in each market, so your plan is tested against more than your own assumptions.
Can you compare self-build and lease in the same region?
Yes. We compare time to capacity, the operators able to deliver at your scale and density, pricing and lease terms, and the power rights each route brings, using the same cases for both.
Do you cover AI and GPU-cloud capacity?
Yes. Our models cover AI compute and cloud as downstream markets, and power, cooling and rack equipment as upstream markets, so we can assess both where AI demand is heading and whether a region can supply high-density capacity.
How do you handle data residency and sovereignty?
We review the data residency, public sector cloud and national AI rules in each candidate market, and show where they require or reward in-country capacity. We work alongside your legal advisers on the legal reading of those rules.
Is our information kept confidential?
Yes. Client data and project findings are used only for your engagement, and confidentiality terms are agreed in writing before any data is shared.
How do we start?
Send us the regions you are considering, the capacity you need and your timeline through the form below. A senior consultant will reply within one business day to arrange a scoping call.
Sources
- Alphabet: fourth quarter and fiscal year 2025 results (Form 8-K exhibit 99.1) (4 February 2026)
- Amazon.com announces fourth quarter results (5 February 2026)
- Meta reports fourth quarter and full year 2025 results (28 January 2026)
- Microsoft: FY26 Q4 press release (29 July 2026)
- International Energy Agency: Energy and AI, executive summary (2025)
- Utility Dive: ERCOT’s large load queue jumped almost 300% last year (6 January 2026)
- NESO: Demand Call for Input, summary of responses (March 2026)
- CBRE: Global Data Center Trends 2025 press release (24 June 2025)
- European Commission: EU launches InvestAI initiative to mobilise EUR 200 billion of investment in artificial intelligence (11 February 2025)
- Central Statistics Office: Data Centres Metered Electricity Consumption 2024, key findings (10 June 2025)
- DLA Piper: New Irish large energy users connection policy (7 January 2026)
- PJM Inside Lines: PJM’s updated 20-year forecast continues to see significant long-term load growth (14 January 2026)
