The cost gap, and what closes it
Cost premium on a 100 MW project against the North American benchmark. ReDATA alone does not close it. Parity needs the state ICMS as well, which Congress cannot waive.
Brazil has the cleanest hyperscale-scale grid in the world and one of the world's tightest water-efficiency thresholds written into tax law. Both could anchor a decade of buildout. Neither has yet produced a financed pipeline at the scale of the announcements.
The structural advantages are real. The pipeline is not.
Brazil enters 2026 with the cleanest hyperscale-scale grid of any major market, approx 85% renewable, anchored in hydroelectric capacity that no peer can match. Microsoft and Amazon have committed a combined USD 4.5 billion in cloud-and-AI capex. The Provisional Measure that became ReDATA put one of the world's tightest WUE thresholds into law. The country has, on paper, the foundation for the largest renewables-anchored AI infrastructure market in the southern hemisphere.
And yet the regime has taken a year to arrive: the provisional measure lapsed in February 2026, the replacement bill only cleared the Senate on 1 September, and the implementing regulation that sets the thresholds has still not been written. Fitch and other observers have flagged the delay. Projects announced since early 2025 remain on hold pending clarity on the thresholds. The headline 13.2 GW MME grid-connection queue, repeated across global press as Brazil's 2030 number, is, on Entelligencia's modelling, only 40 to 55 per cent financed to completion. The rest is conditional on incentives, offtake, or capital that has not yet closed.
The thesis for the chapter is this: Brazil's binding constraint is not power, water, capital, or land. It is regulatory clarity, plus a transmission build-out from the renewable-rich Northeast to the demand-rich Southeast that has not been committed at the scale required. Until both move, the gap between Brazil's announced and delivered capacity will continue to widen, and the operators who actually deliver, verified below, will hold disproportionate market share.
Each spoke is one precondition, scored 0 to 10 for like-for-like reading – indicative editorial weighting, not a published index. Select any spoke or row to read the factor.
Ten files from the operator side, and the longest contribution this report has received. Luciano Fialho scores Brazil 80 out of 100 toward having a window, then argues the constraint is neither power nor capital but two things written in law: the tax stack on imported hardware, and a grid access process that returns connection dates measured in years. Along the way: why Scala filed for 600 MW in 2022, why capital underwrites precedent rather than potential, what ReData must get right and the two things it does not solve, why Brazil can win the tariff and lose the deal, and the two risks he says the market is not pricing. Open any file for the read.
The desk has built out three of his points as charts, with the latest published figures alongside. His reading holds across all three. Where a newer release has landed since he submitted, the newer number is the one shown.
Cost premium on a 100 MW project against the North American benchmark. ReDATA alone does not close it. Parity needs the state ICMS as well, which Congress cannot waive.
Brazil ran at 86.8 per cent renewable electricity in 2025 and curtailed about a fifth of potential wind and solar output, a loss put at R$6.5bn. That is generation already built and not used, which is the load case in one line.
Against R$25bn of total investment, of which R$5bn is the operator’s physical plant and R$20bn the customer’s IT equipment. GDP impact including induced effects, R$2.86bn.
A cost lead’s read on Latin America, contributed to The Next Hotspot
This chapter has argued that Brazil’s constraint is regulatory clarity, and that only 40 to 55 per cent of announced 2030 capacity is financed to completion on Entelligencia’s modelling. Here is that gap read from the cost desk instead. Jose Castro Vila prices builds across the region. Schedule remains a real risk on any Latin American build; his point is that cost uncertainty is typically the greater one, and he puts it at 70 out of 100. His three Brazil-specific findings sit below. The full mechanism runs in Capital.

“A regional benchmark rarely works.”
He separates the three LATAM markets by the kind of difficulty rather than the degree. Brazil is commercial and regulatory complexity. Mexico is growth-driven capacity constraint. Chile is a thin supply chain and contractor market. Similar projects carry materially different cost structures depending on local labour, tax, procurement and market conditions, and he names this as the biggest surprise for newcomers.
Read against this chapter, it is corroboration from an unexpected direction. The regulatory read here was built from the decree and the transmission auctions. He reaches the same conclusion from a cost model, and had not read the chapter.
Typical utility interconnection queue in Mexico, pushing developers onto diesel that adds up to 25 per cent to capital cost. Brazil holds about 39 per cent of regional investment on regulatory complexity rather than capacity.
“The key risk is not finding a contractor, but finding the right contractor with proven capability to deliver at the required level of performance.”
On his account Brazil has sufficient contractor and subcontractor depth to support continued data-centre growth. What varies enormously is standard and quality, delivery maturity, safety culture, risk-management capability and mission-critical experience across that market.
This complicates the chapter rather than confirming it. The delivery pessimism above is drawn from financing and grid timing. He is saying the labour and trade base is not what breaks Brazilian projects, which narrows where the 45 to 60 per cent that does not complete actually fails.
“A contingency tailored to the project’s actual risk profile provides a more realistic basis for decision-making.”
Asked what a realistic contingency looks like against an optimistic one, his answer moves the question. It matters less what the percentage is than how it was derived. A percentage contingency drawn from historical benchmarks and applied one-size-fits-all is useful early in a project, and a weaker basis for a decision, because it describes developments that already happened rather than the one being priced.
What he argues for instead is a deterministic contingency built from an integrated risk assessment, with each risk identified, evaluated and quantified against probability and impact, producing a number that belongs to that development rather than to an average.
It matters for this chapter because it names where the optimism actually enters. Only 40 to 55 per cent of announced Brazilian capacity is financed to completion on this desk’s modelling. On his account that gap does not open because anyone estimated badly. It opens because a percentage carried forward from elsewhere was never a statement about this project.
Effective Brazilian import burden on data-centre equipment once II, IPI, PIS/COFINS and ICMS cascade. The ReDATA exemption lapsed in February 2026 when Congress did not convert MP 1.318/2025. The successor bill, PL 278/2026, passed the Senate on 1 September 2026 and awaits presidential sanction.
Contributed via the Entelligencia briefing survey, 29 July 2026, extended in follow-up and revised by the contributor on review, 30 July 2026. Nine questions were put; all nine were answered. Attribution is as the contributor elected on submission: named, with firm and title. Both findings are graded Contributed rather than Verified: no published dataset on Latin American contingency methodology or contractor delivery maturity exists to test them against, which is the reason they are worth printing. His wider argument on procurement, imported equipment and change control runs in Capital.
Clean generation, transmission and the regulatory gate that throttles delivery. Open the pipeline to model where Brazil's queue holds, and where it breaks.
Brazil's pipeline is not constrained by power, land, or capital. It is constrained by the ReDATA rulebook. The bill cleared the Senate on 1 September 2026; the regulation setting the thresholds has not been written. Until it is, the announced-vs-delivered gap continues to widen.
Brazil added around 13.2 GW of solar in 2025 to reach a cumulative 67.1 GW, but the pace is slowing. Regulatory uncertainty, grid-connection barriers for distributed generation and a sharp tariff hike have all bitten: the import duty on solar modules was lifted to 25% just as China cut its export rebate to 9%, raising utility-scale project costs by about 8%. Developers have pulled back, with Trinity Renováveis shelving major projects and ForGreen cutting planned investment by up to 35%.
The harder constraint is the wires. Transmission congestion forced 15 to 20 per cent of solar and wind output to be curtailed in 2025, more than BRL 1.7 billion in uncompensated losses for generators. The regulatory answer is storage: ANEEL has closed a second round of consultations on utility-scale batteries, proposing lower transmission fees and a simpler approval path ahead of a capacity auction, the mechanism meant to soak up curtailed power and firm the grid for the hyperscale loads now queueing for connection.
Brazil is not short of generation. It is short of the step between authorised and connected, and read that way a high power score is a statement about announcements rather than about electrons. This contributed brief takes the queue apart: why the paper is the constraint, why clearing it would free more capacity than a new transmission line, what sovereignty actually means when a hyperscaler operates the building, and why three cities competing is not yet a strategy. He puts the odds that policy fails to clear it at 30 out of 100, which makes him one of the more optimistic contributors in this chapter. Open any file for the read.
The seat above puts Brazil’s constraint in the interconnection queue and scores the delivery risk at 30. This one puts it in the statute book and scores it at 50, the exact midpoint. Six files on why legal certainty rather than power is the blocker, why the announcement of ReDATA may have done more damage than its delay, what the regime actually offered once you read it against Ex-Tarifário and the 2027 taxes, and where Brazil genuinely beats the Gulf and Indonesia. Open any file for the read.
Brazil’s edge is power: a clean, grid-ready supply that has pulled the densest data-centre cluster in Latin America into the São Paulo southeast, with gigawatt-scale AI cities announced in Rio and the south and a subsea gateway in the northeast. Each glowing node is a real site, colour-coded by what it is: verified live or in build, or an announced megaproject. Open one for the detail, filter by region, and toggle the cables and the renewable grid. The outline follows real national boundary data and every site sits on its true coordinates.
Independently confirmed. Construction or operation visible on-site or in regulatory filings. Cross-referenced against grid-connection queues, power-purchase filings, and published reporting.
Press release, no independent confirmation. Project publicly announced but financing-to-completion, permitting, or offtake not yet verified.
Named source disputes the public figure. A senior operator, regulator, or financier has indicated the announced figure is materially overstated, conditional, or unfinanced.
| Project | Location | Announced | Verified live or in build | Status | Binding constraint | |
|---|---|---|---|---|---|---|
|
Ascenty SP4 expansion
Digital Realty / Brookfield
|
Hortolândia, SP
Campinas corridor
|
68MW |
68MW |
Verified | Grid interconnect timing |
|
|
What we verified
Construction visible on-site in publicly available imagery. Power allocation cross-referenced against ONS interconnection record. PPA in place with CPFL for hydro-anchored supply. Tenant migration underway from SP3. What would change the tag
A material delay on the second 34 MW phase, currently scheduled for Q4 2026, would move that phase to Announced. The verified figure here covers only the first phase. Sources
ONS interconnection queue · CPFL PPA filings · Reuters · Valor Econômico. |
||||||
|
ODATA SP02
Aligned Data Centers
|
Santana de Parnaíba, SP
São Paulo metro
|
36MW |
36MW |
Verified | Water permitting (resolved) |
|
|
What we verified
Live tenant operation. Closed-loop cooling design indicated in ODATA / Aligned public materials. Targeted to operate below the ReDATA WUE threshold of 0.05 L/kWh. What would change the tag
Operational. A change in tag would require evidence of off-spec performance or unannounced shutdown. Sources
ODATA / Aligned press materials · São Paulo state water authority filings · Data Center Dynamics · Reuters. |
||||||
|
Elea AI campus, Tamboré
Elea Data Centers
|
Tamboré, SP
Alphaville corridor
|
24MW |
24MW |
Verified | Resolved · liquid-cooled |
|
|
What we verified
Brazil's first purpose-built AI data centre on liquid cooling. CDU-based deployment delivered by Vertiv. PUE band of 1.15-1.20 referenced in Elea and Vertiv joint materials. What would change the tag
The flagship Elea facility is live and operating. Future phases are tracked separately and currently sit under Announced. Sources
Elea + Vertiv joint announcements · Frost & Sullivan 2025 · Data Center Dynamics · Valor Econômico. |
||||||
|
Partner case study
The CDU rollout that anchored Brazil's first AI build
Hundreds of coolant distribution units · 100 kW per rack · Frost & Sullivan 2025 LATAM Company of the Year
1.15-1.20
Measured PUE
2024 →
Live, scaling
Read the case →
|
||||||
|
Microsoft Brazil cloud + AI commitment
|
Multi-site
SP / RJ, not yet named
|
$2.7bnCapex |
–Not disclosed |
Announced | ReDATA implementing decree |
|
|
Why this is Announced, not Verified
The headline capex figure has been publicly committed but no MW figure has been published by Microsoft for the Brazil deployment, and the deployment timing is partially conditional on ReDATA's implementing decree. The original ReDATA provisional measure expired in February 2026 and was replaced by PL 278/2026, which cleared the Senate on 1 September 2026 and awaits presidential sanction, with the Ministry of Finance putting the incentive’s draw at up to R$2 trillion of private investment over the decade and R$5.2bn earmarked in the 2026 budget alone. The lapse and the scramble to replace it are why this stays Announced: the incentive is now law, but the regime has already been interrupted once. What would move it to Verified
A site announcement with named operator partner, grid-interconnection filing, and the ReDATA implementing decree being issued with WUE and clean-power thresholds intact. Sources
Microsoft press release · Fitch Ratings commentary · ReDATA public consultation documents · Reuters · Bloomberg. |
||||||
|
Amazon Brazil cloud + AI commitment
|
Multi-site
SP-led
|
$1.8bnCapex |
–Not disclosed |
Announced | ReDATA implementing decree |
|
|
Why this is Announced, not Verified
Capex figure committed publicly. No MW disclosure, no site selection disclosure, no operator partner publicly identified for the deployment. Sits behind the same ReDATA policy gate as the Microsoft commitment. What would move it to Verified
Named site, named operator partner, grid-interconnection filing or evidence of construction. Sources
AWS Brazil announcement · Reuters · Folha de S.Paulo · Data Center Knowledge. |
||||||
|
Scala AI City
Scala Data Centers
|
Eldorado do Sul, RS
Greater Porto Alegre
|
4,750MW potential |
54MW initial |
Announced | Enabled; 5 GW grid pathway |
|
|
Why this is Announced, not Verified
Scala and the Rio Grande do Sul government disclosed an initial US$500 million / R$3 billion investment, with 54 MW of initial IT capacity in Eldorado do Sul and expansion potential to 4.75 GW across more than 1,730 acres. The local enabling law was sanctioned in December 2024, and in May 2025 the Ministry of Mines and Energy recognised a technical solution for a 5 GW grid connection. Enabled and in development, with no operating phase or named anchor customer in the public record. What would move the campus figure to Verified
A named anchor tenant, grid-interconnection energisation and an operating first phase. Separately, Scala's Tamboré buildings in Barueri, SP (SGRUTB09 to 11, about 84 MW) are under construction within a wider 900 MW-plus Scala programme. Sources
Scala and Rio Grande do Sul government announcements · Ministry of Mines and Energy · ONS queue filings · Valor Econômico. |
||||||
|
ODATA SP04
Aligned Data Centers
|
Osasco, SP
Greater São Paulo
|
48MW |
48MW |
Verified | Live, early 2025 |
|
|
What we verified
Announced March 2025 and described by ODATA as operational from early 2025. Investment of more than US$450 million, with the total campus put at R$2.6 billion in later technical material. ODATA's second verified Brazilian site after SP02, carrying the same closed-loop cooling and sustainability framing. What would change the tag
Operational. A change would require evidence of off-spec performance or an unannounced shutdown. No anchor tenant is publicly named. Sources
ODATA / Aligned press materials · Data Center Dynamics · Reuters. |
||||||
|
Ascenty AI hyperscale wave
Digital Realty / Brookfield
|
Campinas region, SP
Sumaré, Vinhedo, São Paulo
|
150MW pre-leased |
SPO05live |
Announced | Pre-leased; phased to 2027 |
|
|
Why this is Announced, not Verified
In May 2026 Ascenty said it had signed 150 MW of new AI contracts with global hyperscalers, fully pre-leased on a long-term basis, supporting four new hyperscale data centres across Sumaré and Vinhedo plus a sixth São Paulo site. SPO05 has entered operation and SPO06 is under construction for May 2027. The clearest pre-leased hyperscaler-demand signal among the named operators, though customer names are not disclosed and most of the capacity is not yet energised. What would move it to Verified
Energisation of the named Sumaré and Vinhedo phases, with grid-interconnection evidence beyond the live SPO05 building. Sources
Ascenty May 2026 announcements · Digital Realty disclosures · Data Center Dynamics. |
||||||
|
Elea Rio AI City
Elea Data Centers
|
Rio de Janeiro
Olympic Park region
|
1,500MW phase 1 |
approx 80MW in build |
Announced | Phased; 300 MW+ power secured |
|
|
Why this is Announced, not Verified
Announced May 2025. RJO1 is operating and RJO2, about 80 MW, is under construction for 2026, with RJO3 and RJO4 adding a further 120 MW. First-phase target 1.5 GW, long-term potential 3.2 GW. Elea's April 2026 transaction, I Squared Capital's majority-stake acquisition, cited more than 300 MW on land with secured power and more than 1 GW planned. Separately, Elea won a named anchor: the Petrobras supercomputing data centre, 30 MVA on a 17-year, R$2.3 billion contract. What would move it to Verified
Energisation of RJO2 and a named hyperscaler anchor for the AI City itself, distinct from the Petrobras HPC contract. Sources
Elea announcements (May 2025, Apr 2026) · Petrobras procurement · Data Center Dynamics. |
||||||
|
Takoda dual-campus
Takoda / TIVIT
|
Sumaré, SP + Barra, RJ
Multi-site
|
160MW |
36MW phase 1 |
Announced | RFS Q3 2027 |
|
|
Why this is Announced, not Verified
Takoda, the data-centre company TIVIT spun out in 2023, markets two hyperscale AI and GPU campuses: Sumaré, São Paulo at 96 MW and Barra, Rio de Janeiro at 64 MW, 160 MW in total. Phase 1 is 18 MW per campus, with ready-for-service in Q3 2027. Takoda says land, permits, critical-equipment contracts and construction contracts are already in force, but no customer is named. What would move it to Verified
Start of construction with grid-interconnection evidence and a named anchor tenant. Sources
Takoda 2026 project pages · TIVIT corporate materials. |
||||||
|
Tecto TGRU1
Tecto / V.tal
|
Santana de Parnaíba, SP
São Paulo metro
|
200MW |
–in build |
Announced | Part of US$2bn plan to 2028 |
|
|
Why this is Announced, not Verified
Tecto, V.tal's data-centre platform, lists TGRU1 in Santana de Parnaíba as under construction for high-density AI and hyperscale loads, up to 200 MW, part of a wider US$2 billion investment plan through 2028. Tecto also markets TPOA1 in Porto Alegre, RS, 20 MW with a first phase in Q4 2026 tied to the Malbec subsea cable, and TFOR3 in Fortaleza. The platform runs on certified renewable energy; no anchor is named. What would move it to Verified
Energisation, grid-interconnection evidence and a named tenant for TGRU1. Sources
Tecto / V.tal site and announcements · Reuters · Data Center Dynamics. |
||||||
|
Northeast AI corridor (composite)
Multiple announced operators
|
Ceará / Pernambuco / RN
Cable-landing corridor
|
approx 3,200MW |
approx 40MW |
Contested | Offtake + transmission |
|
|
Why this is Contested
Aggregate announced capacity across the Northeast corridor exceeds three gigawatts. Cross-referenced against MME grid-queue filings and named financing, Entelligencia assesses a significant share of these announcements as materially overstated or unfinanced. Transmission capacity from the renewable-rich Northeast to demand centres in the Southeast has not been built at the scale required. What would change the tag
Named transmission build-out commitments by ONS / MME, plus offtake agreements for specific named operators in the corridor. Sources
MME grid-queue analysis · Reuters · Financial Times · Valor Econômico · Entelligencia cross-reference against publicly named projects. |
||||||
|
Aggregate "2030 hyperscale pipeline"
Industry body composite
|
Brazil-wide
All regions
|
approx 13,200MW |
approx 1,000MW today |
Contested | Composite |
|
|
Why this is Contested
The 13.2 GW figure is the MME grid-connection request pipeline, repeated in media as "Brazil's 2030 capacity". The Entelligencia view, cross-referenced with the named advisory council, is that 40 to 55 per cent of that pipeline is currently financed to completion. The rest is conditional on incentives, offtake or capital that has not yet closed. What would change the tag
The ReDATA implementing decree being issued with thresholds intact, plus a wave of named offtake announcements moving aggregate financed share above 70%. Sources
MME grid queue · ANEEL filings · Entelligencia aggregation · Reuters · Valor Econômico · Bloomberg. |
||||||



Three primary sources behind Brazil’s build: the tax regime, the energy signal and the grid queue, each graded claim by claim against the record. Select a tab to bring a document forward, an underlined claim to read the analysis, or filter by grade.
Contributions gathered through the report’s open call. Named voices appear with consent; the hyperscaler operator is verified and held on file at their request.








































Capital flows left to right, from investors through suppliers and operators to the hyperscalers, corporates and AI workloads that consume the stack. Hover any node to trace its position. Click for the company brief.
06 tiers · 41 entitiesDescription.
The clearest read of who is actually delivering Brazil's data-centre buildout, what they think happens next, and what the real capacity picture looks like through 2030. around 120 pages. Site visits, named interviews, the people-map of the market.
Each briefing is a quarter of a flagship: narrower scope, sharper edge, faster cadence. The delta ledger shows what moved between Verified, Announced, and Contested. Named interviews. Operator updates.
Two closed-door convenings flanking the flagship publication. Advisory council, operators, hyperscalers, capital. The dossier is what they take home; the room is what they paid to be in.
Where the build actually happens, and what is in the way.


