Suppose you read that the world may need about 156 GW of AI-related data-center capacity by 2030.
What does 156 GW actually look like?
Is that 156 giant 1 GW campuses?
1,560 facilities of 100 MW?
Or something else entirely?
A useful way to picture the forecast is about 1,560 blocks of 100 MW AI capacity. But those blocks are a measuring unit—not a prediction that 1,560 identical buildings will be constructed.
By the end of this article, you will be able to convert a gigawatt forecast into 100 MW blocks, tell a capacity forecast from an annual-energy forecast, and spot when a scenario is being mistaken for a literal construction count.
Start with the Forecast, Not the Building Count
McKinsey’s 2025 continued-momentum scenario estimated global AI-related data-center capacity demand at 44 GW in 2025 and 156 GW in 2030.[1]
A later 2026 McKinsey analysis describes the same scale as about 155 GW of AI-related demand by 2030.[2]
The one-gigawatt difference is not the important story.
The useful reading is:
under this McKinsey scenario, AI-related capacity is on the order of 155–156 GW by 2030.
That is the forecast we are translating.
The Contexta Forecast Decoder
Before turning any 2030 forecast into a headline, check five things.
| Check | Question |
|---|---|
| Scope | AI workloads only, or all data centers? |
| Metric | GW of capacity/power demand, or TWh of annual electricity? |
| Boundary | IT capacity, facility demand, or another defined system boundary? |
| Scenario | Central, continued-momentum, high-growth, constrained? |
| Date | When was the forecast published or updated? |
This five-step check explains why two respected forecasts can show very different-looking numbers without actually contradicting each other.
The Direct Conversion: 156 GW = 1,560 Blocks of 100 MW
One 100 MW block is 0.1 GW.
156 GW ÷ 0.1 GW = 1,560
So the original headline answer still works:
McKinsey’s 156 GW 2030 AI-capacity scenario equals about 1,560 units of 100 MW capacity.
At 44 GW in 2025, the same normalization gives:
44 GW ÷ 0.1 GW = 440 blocks
Under that scenario, the AI-capacity pool grows from about 440 to 1,560 100 MW blocks.
This chart converts one McKinsey capacity scenario into a common 100 MW unit. It does not count announced buildings.
The Contexta Equivalent Block Rule
The conversion is simple enough to remember:
100 MW equivalents = Forecast GW × 10
Examples:
- 10 GW → 100 blocks of 100 MW
- 50 GW → 500 blocks
- 100 GW → 1,000 blocks
- 156 GW → 1,560 blocks
The rule is useful because it turns a large GW forecast into the same unit used in the previous articles.
But it is only a normalization tool.
Why 1,560 Does Not Mean 1,560 Data Centers
Data centers are not standard-size boxes.
A campus can be smaller than 100 MW.
Another can be 250 MW, 500 MW, or eventually gigawatt-scale.
The same 156 GW pool can therefore be represented in many ways:
| Illustrative average capacity per campus | 156 GW divided by that size |
|---|---|
| 50 MW | 3,120 campus-equivalents |
| 100 MW | 1,560 |
| 250 MW | 624 |
| 500 MW | 312 |
| 1 GW | 156 |
None of these rows is a construction forecast.
They show the same capacity pool divided by different assumed campus sizes.
The Contexta Building-Count Trap
A forecast becomes misleading when this sentence:
“AI-related capacity could reach 156 GW.”
quietly turns into:
“The world will build 1,560 new 100 MW data centers.”
The first is a scenario about a capacity pool.
The second is a specific claim about facility count and size distribution.
You cannot get the second claim from the first without additional assumptions.
Capacity equivalents are a ruler. They are not a construction schedule.
Why a Newer 2026 Forecast Can Look Much Bigger
In June 2026, Gartner projected 290 GW of worldwide data-center power demand by 2030 and more than 1,200 TWh of annual electricity consumption.[3]
That number is much larger than McKinsey’s 155–156 GW AI-related figure.
But the Forecast Decoder shows why we should not call that a contradiction.
- McKinsey figure: AI-related data-center capacity demand
- Gartner figure: worldwide data-center power demand
The scopes and model definitions are different.
Gartner’s 2026 release also says AI-optimized servers are increasingly important, but its 290 GW figure is not labeled as “AI-only capacity.”
So it should not replace the 156 GW numerator in the 1,560-block calculation.
Capacity and Annual Electricity Are Two Different Lenses
Capacity uses units such as MW and GW.
Annual electricity uses MWh, GWh and TWh.
The IEA’s September 2026 outlook projects total global data-center electricity use rising from about 485 TWh in 2025 to about 950 TWh in 2030.[4]
The IEA also says electricity consumption from AI-focused data centers could roughly triple over that period.
If we divide the IEA’s all-data-center 950 TWh forecast by the illustrative 0.968 TWh/year used by our earlier 100 MW model, we get:
950 TWh ÷ 0.968 TWh ≈ 981 model-energy equivalents
But this number needs a warning label.
981 is not a forecast of 981 data centers. The numerator covers all data-center electricity in the IEA outlook, while the denominator is one illustrative operating model with its own load-factor and PUE assumptions.
It is useful only as an energy-scale translation.
Read this as an energy-scale translation of the IEA outlook, not as a site-count forecast.
Do Not Average 981 and 1,560
The original article correctly warned that the two numbers measure different things.
That distinction is worth making even stronger.
| Lens | Numerator | What the equivalent means |
|---|---|---|
| Capacity lens | 156 GW AI-related capacity scenario | 1,560 blocks of 100 MW capacity |
| Energy lens | 950 TWh all-data-center annual electricity scenario | ~981 copies of one illustrative 0.968 TWh/year energy model |
Different scope.
Different metric.
Different denominator.
Therefore: different question.
Why the Forecast Can Still Move
McKinsey explicitly models multiple scenarios because future AI capacity depends on uncertain variables such as AI adoption, hardware and software efficiency, semiconductor supply, regulation and project delivery.[1]
The IEA reaches the same general conclusion from the energy side: bottlenecks can slow near-term deployment, while stronger AI adoption and infrastructure investment can create upside later.[4]
Recent public discussion shows that readers are already wrestling with exactly this uncertainty: when a forecast says 100 GW, 121 GW or 290 GW, the next question is often “Can the grid, turbines, transformers, chips and construction system actually deliver it?”
That is the right next question.
The Contexta Forecast-to-Infrastructure Ladder
A capacity forecast is not only a server forecast.
It implies a chain of physical requirements:
AI Capacity → Facility Power → Grid Connection → Generation → Transformers → Cooling → Buildings → Skilled Labor
McKinsey’s 2025 analysis estimated about $5.2 trillion of cumulative investment by 2030 to meet worldwide AI-related data-center demand under its central/continued-momentum assumptions.[1]
That total includes far more than accelerators.
Power, electrical systems, cooling, site construction and network infrastructure all have to arrive.
This is why a 156 GW capacity number is really a coordination problem hidden inside one forecast.
Run the Conversion Yourself
forecast_ai_capacity_gw = 156
block_size_mw = 100
capacity_equivalents = (
forecast_ai_capacity_gw * 1000
/ block_size_mw
)
print(capacity_equivalents)
# 1560Change 156 to another forecast.
Change 100 to 250, 500 or 1,000 MW.
The arithmetic changes instantly.
The harder work is still the Forecast Decoder: making sure the input number actually means what you think it means.
What Should You Watch Next?
- Forecast scope: AI-only or all data centers?
- Inference growth: McKinsey’s later work expects inference demand to grow particularly quickly toward 2030.[5]
- Power-secure capacity: how much announced capacity has a credible grid or onsite-power path?
- Campus size: does new capacity arrive as many 100 MW sites or fewer 500 MW–1 GW campuses?
- Efficiency: does better compute-per-watt reduce electricity needs, or does cheaper compute drive more use?
- Project realization: how much forecast demand becomes energized, operating capacity rather than a queue entry?
The Main Idea
The question in the title sounds like it asks for a building count.
It does not have one reliable building-count answer.
What we can say is more precise:
Under McKinsey’s 2025 continued-momentum scenario, 156 GW of AI-related data-center capacity in 2030 equals about 1,560 blocks of 100 MW capacity. A later McKinsey update puts the same scale at about 155 GW. These are capacity equivalents—not 1,560 literal buildings.
The next useful question is physical:
What has to be built so that this capacity can actually turn on?
Continue Reading
- How Much Power Is 100 MW? An AI Data Center Compared with Entire Cities — understand the scale of one 100 MW block.
- What Must Be Built to Power the AI Data Center Boom? — follow the physical infrastructure behind the forecast.
- AI Data Center Power Cost Calculator in Python — change the assumptions yourself.
Key Terms
- capacity equivalent: a common-size block used to express a larger capacity forecast; it is not necessarily one physical building
- capacity demand: the amount of data-center power capacity expected to be required for workloads
- annual electricity consumption: the energy used over one year, usually expressed in TWh at global scale
- AI-related capacity: data-center capacity associated with AI workloads under the source forecast’s methodology
- scenario: a modeled future based on specified assumptions rather than a guaranteed outcome
- continued-momentum scenario: McKinsey’s scenario in which current growth drivers continue broadly along the modeled trajectory
- scope: the population included in a forecast, such as AI workloads only or all data centers
- metric: the quantity being measured, such as GW of capacity or TWh of annual electricity
- inference: using a trained AI model to generate outputs for real requests
- energized capacity: capacity that has actually obtained power and can operate, rather than merely announced or queued capacity
Sources
- McKinsey — The cost of compute: A $7 trillion race to scale data centers — 156 GW AI-related capacity scenario and investment model.
- McKinsey Global Institute — Colocation data centers: The infrastructure race behind AI — 2026 restatement at about 155 GW AI-related demand by 2030.
- Gartner — Data Center Electricity Consumption to Grow 26% in 2026 — 290 GW worldwide data-center power-demand forecast and >1,200 TWh 2030 electricity outlook.
- IEA — Key Questions on Energy and AI, Executive Summary — 485 TWh in 2025 to about 950 TWh in 2030; AI-focused electricity roughly triples.
- McKinsey — The next big shifts in AI workloads and hyperscaler strategies — training and inference workload split toward 2030.
Status checked September 30, 2026. The Forecast Decoder, Equivalent Block Rule, Building-Count Trap, and Forecast-to-Infrastructure Ladder are The Contexta analytical frameworks. The 1,560 figure is a normalization of McKinsey’s 156 GW scenario into 100 MW blocks, not a forecast that 1,560 identical facilities will be built. Forecasts from McKinsey, Gartner, and the IEA use different scopes, metrics and assumptions and should not be treated as interchangeable.