Latin American Data Centers: Which Capacity Is Commercially Real?

What the headline counts

Announced MW is not equivalent to capacity that has crossed delivery gates. Across Brazil, Chile and Mexico that is true of deliverability, with evidence depth varying by market and Chile’s the thinnest. A single conversion ratio from announced to deliverable MW is absent from published figures. What matters for a decision is which MW can compete for tenants within the period a decision spans. Headline pipeline size is the figure most readily to hand, and it answers a different question.

Two pending milestones keep the distinction live. Brazil’s REDATA fiscal regime passed the Senate on 1 September 2026 and was signed into law a fortnight later, though its implementing regulation is still to come. Microsoft’s Querétaro energisation was reported, from a 2023 filing, as not before Q2 2027.

Which MW count?

Wykeham’s read is that a count starts with MW basis, then the delivery gates each asset has crossed, then demand state and economics. IT MW, the power available to customer equipment, is the basis comparable to leasable capacity; MWmed (the average electrical load over a period) and demand MW convert to it only with a load factor. Demand state is counted separately: one Chilean financing records 23 MW contracted and 30 MW reserved, and both precede live capacity. Grid interconnection means the power grid, while interconnection alone means digital connectivity. We test whether grid access is binding, and Exhibit 1 sets out the five tests that follow from it.

Test number	Test	Question	Good signal	Bad signal
T1	MW basis	What does each figure measure: IT MW, MWmed, demand or requested MW?	Each figure keeps its unit, stage, inclusion test and date	Different bases summed or compared directly
T2	Power access and energisation timing	Is usable power secured, by what reported route, and when?	An evidenced route covering the MW counted; a stated energisation date	Power secured for one phase while the whole campus is counted
T3	Permit	Are the required approvals presently executable?	No live litigation, reassessment or pending supplementary approval	Execution barred or reassessment ordered
T4	Demand state	Is demand live, contracted, reserved or tracker "Committed"?	Each state kept separate, with named counterparties	Reservations or unnamed anchors counted as take-up
T5	Achieved economics and competitive position	Is there achieved pricing or independent commercial-outcome evidence?	Achieved $/kW/month by comparable tier; independent outcomes	Asking prices, concentration alone, or an operator's own claim

Source: Wykeham analysis of the definitions used by DC Byte, EPE and the operators cited in this article.

Notes: MW bases are not converted into one another. Delivery gate and demand state stay separate; contracted or reserved demand, and DC Byte's "Committed", are not live absorption.

Limits: a diagnostic screen, not a conversion model, score, ranking or causal sequence.

Three markets, one screen

Each market runs against the Exhibit 1 screen on uneven evidence depth, and the screen assumes no common constraint across the three.

Brazil: the queue filter

Brazil’s queue divides under a financial-guarantee test. Once those guarantees applied, 39 of 94 pending processes were formalised with the system operator (ONS) as 43 access requests, 38 of them data centres totalling about 7,040 MW. EPE put 2030 data centre load at 3,457 MWmed in April (up from 2,157) with signed contracts covering 86.3% of load on average but 65% in 2030. An August revision raised the system forecast again, citing data centres.

Power is a candidate constraint on DC Byte’s interpretation, and whether physical headroom, access rules or both explain it is unresolved. Permit status and market-level demand are both still to be settled. However, DC Byte places 76% of live capacity in São Paulo and 85% of IT capacity with three operators, which is concentration, and achieved pricing remains unavailable. No Brazil operator case with a known outcome is reported.

For sizing, formalised access may deserve more weight than requests alone.

Chile: one permit gate

Chile’s evidence is the thinnest, and it rests on one dataset.

A single supply figure would be easier to work with. Three readings stand unreconciled and all are carried: DC Byte gives 600 MW of current supply and 461 MW of total supply, while a property-market report projects about 341.6 MW installed in the Metropolitan Region in 2026. At Cerrillos a court-ordered reassessment has, as reported, held delivery at zero, so approval alone did not make that asset deliverable. Still, that is one asset, short of a market finding. Curauma’s 5 MW was reported fully booked to one hyperscaler at its 2023 launch, which is asset-level evidence and keeps market absorption open. Scala’s portfolio page, last updated in May 2025, still lists 2024-dated capacity as yet to operate, cause undisclosed.

Buyers may put permit executability and passed service dates into the count.

Mexico: power and timing

Power access is reported to have gated energisation in each Querétaro case reviewed.

ODATA reports its capacity in layers: 300 MW announced, 200 MW energised upstream, 72 MW IT for building one, about 6 MW in service at launch, each of which measures a different thing. National bases differ too: DC Byte reports 1,251 MW, MEXDC 279 MW operating plus 205 MW under construction, and BNEF 244.9 MW live. CloudHQ reports 200 of up to 920 MW secured, while CFE reported 1,535 MW of 2025 state demand against 723 MW of in-state capacity met partly by out-of-state generation, and the state energy agency reports 540 MW requested from CENACE. Permit status is undetermined, and demand and position depend on ODATA’s claim of three unnamed anchors. Still, no second dataset tracks the market.

Wykeham’s read is that for the large sponsors observed, most clearly ODATA and CloudHQ, grid timing shifted schedules and required enabling investment rather than stopping the projects. Whether that transfers to the mid-market is untested.

Comparing the three markets

The same five tests applied across the three markets show different evidence depth and different candidate gates. Exhibit 2 sets out what is settled in each market, and it implies no ranking of market attractiveness.

Evidence status

Market	T1MW basis	T2Power access and energisation timing	T3Permit	T4Demand state	T5Achieved economics and competitive position
Brazil	Incompatible bases	Candidate; cause open	Not reached	Not reached at market level	Concentration only
Chile	Unreconciled	Contested at market level	Supported at Cerrillos only	Asset level only	Not reached
Mexico	Layers, not a ratio	Supported, Querétaro cases only	Not reached	Thin	ODATA leadership claim only

Country evidence

Test	Brazil	Chile	Mexico
T1 MW basis	Incompatible bases: 39 of 94 processes formalised as 43 requests, 38 of them data centres, about 7,040 MW requested; 2,157 to 3,457 MWmed load in April, load not supply; August revision raised the system forecast	Unreconciled: 600 / 461 MW; about 341.6 MW projected for 2026	Layers, not a ratio: 300 / 200 / 72 / about 6 MW. National bases differ: 1,251 MW; 279 plus 205 MW; 244.9 MW
T2 Power access and energisation timing	Candidate on DC Byte's interpretation; cause open	Contested at market level	Supported, Querétaro cases only: 200 of up to 920 MW secured; energisation not before Q2 2027, from a 2023 filing
T3 Permit	Not reached	Supported at Cerrillos only: delivery held at zero, as reported; original route later abandoned, new 2026 consultation pending	Not reached
T4 Demand state	Not reached at market level	Asset level only: Curauma 5 MW, fully booked at its 2023 launch	Thin: three unnamed anchors
T5 Achieved economics and competitive position	Concentration only: São Paulo 76% of live capacity; three operators 85% of IT capacity; pricing not reached	Not reached	ODATA's own leadership claim only; pricing not reached
Main evidentiary limit	No operator case with a known outcome	No second instrument; cause of Scala slippage undisclosed	Large sponsors only; no second instrument; mid-market transfer untested

Brazil

T1 MW basis

Incompatible bases: 39 of 94 processes formalised as 43 requests, 38 of them data centres, about 7,040 MW requested; 2,157 to 3,457 MWmed load in April, load not supply; August revision raised the system forecast

T2 Power access and energisation timing

Candidate on DC Byte's interpretation; cause open

T3 Permit

Not reached

T4 Demand state

Not reached at market level

T5 Achieved economics and competitive position

Concentration only: São Paulo 76% of live capacity; three operators 85% of IT capacity; pricing not reached

Main evidentiary limit

No operator case with a known outcome

Chile

T1 MW basis

Unreconciled: 600 / 461 MW; about 341.6 MW projected for 2026

T2 Power access and energisation timing

Contested at market level

T3 Permit

Supported at Cerrillos only: delivery held at zero, as reported; original route later abandoned, new 2026 consultation pending

T4 Demand state

Asset level only: Curauma 5 MW, fully booked at its 2023 launch

T5 Achieved economics and competitive position

Not reached

Main evidentiary limit

No second instrument; cause of Scala slippage undisclosed

Mexico

T1 MW basis

Layers, not a ratio: 300 / 200 / 72 / about 6 MW. National bases differ: 1,251 MW; 279 plus 205 MW; 244.9 MW

T2 Power access and energisation timing

Supported, Querétaro cases only: 200 of up to 920 MW secured; energisation not before Q2 2027, from a 2023 filing

T3 Permit

Not reached

T4 Demand state

Thin: three unnamed anchors

T5 Achieved economics and competitive position

ODATA's own leadership claim only; pricing not reached

Main evidentiary limit

Large sponsors only; no second instrument; mid-market transfer untested

Source: DC Byte; EPE, ONS and CCEE; a property-market report for Chile; Google Cerrillos court records and press; Scala; ODATA; Microsoft; CloudHQ; CFE and the Querétaro state energy agency; MEXDC; BNEF; A&O Shearman. Wykeham analysis.

Notes: figures keep their original definitions, units, dates and inclusion tests. "Not reached" means the admitted evidence does not establish the test, not that there is no effect.

Limits: no scoring or cross-market ranking; bases not summed or converted; Chile and Mexico lack a second repeated instrument.

How much is settled

Grid access and grid interconnection are the most recurrent candidate constraint across the three markets. Given that, the strongest available statement is a market-by-market answer, and a cross-market ranking would need more. Capital was not reported as the gate for the large sponsors observed, and Scala closed $254 million of project finance, leaving the mid-market position open.

Market-level absorption, monetisation and pricing, market-wide competitive-position effects and the constraint ranking are all unresolved. That’s a limit of the sources, not a finding about the market.

Consider discounting or pausing the count when bases resist reconciliation, a gate is unpassed, a stated service date has passed with capacity still to operate, power serves only part of the MW counted, requests are treated as pipeline or reservations are presented as live.

What each reader tests

An acquirer underwriting an asset or a pipeline can test power route, permit status, energisation and demand state before admitting future MW, treating stated dates as claims until capacity operates. Ask whether power serves the full campus or one phase.

For an owner or operator assessing its own position, control of scarce enabling resources is a question to test. Independent take-up, achieved pricing and win/loss evidence may be needed before scarcity links to commercial position.

Where the task is sizing the market, a sponsor or corporate can rebuild supply on comparable MW bases and demonstrated delivery state, asking what each figure measures and which gate it has crossed. Whether smaller sponsors secure power as large ones reportedly did is untested.

On a named asset

Public evidence can drive this screen, but a named asset needs more: take-up with live, contracted and available capacity kept apart, achieved pricing by deployment tier, and the gate the asset has crossed. Commercial due diligence gets to those on a live engagement. To discuss how these tests apply to a specific asset, please get in touch.