Brazil's data center connection queue is 26.2 gigawatts. The transmission that has actually been costed unlocks four.
EPE's own numbers: requests jumped from 19.8 GW to 26.2 GW in the sixty days after Redata was published. The only package studied, costed and recommended is R$ 1.6 billion in São Paulo, and it opens about 4 GW.
Brazil spent the last year winning the argument that it should host AI infrastructure. A heavily renewable matrix, a tax regime for compute hardware published in September 2025, and a queue of interested developers to show for it. What it has not published is a transmission answer sized to the demand it attracted. The gap is now measurable, because the planning agency published both halves.
What’s happening
- Redata (MP 1318/2025) was published on 17 September 2025, cutting the tax burden on imports and purchases of IT equipment such as GPUs, switches and racks [EPE, Notícias].
- Roughly sixty days later, an additional 6.4 GW of data center projects had filed minimum global cost studies with the Ministry of Mines and Energy seeking a basic network connection. The total in process moved from 19.8 GW in September 2025 to 26.2 GW in November 2025 [EPE, Notícias]. That is a 32 percent increase in two months, desk arithmetic on EPE’s own figures.
- The 26.2 GW number reached Brazilian trade press on 3 September, framed as a test of the grid rather than a forecast of load [Cenário Energia, 3 September 2026].
- Concluded EPE studies recommend about R$ 1.6 billion in new transmission investment for São Paulo, unlocking roughly 4 GW of connection margin. Further São Paulo studies are still running, with new works expected to compose the 2026 auctions and add about 5 GW more statewide [EPE, Notícias].
- Rio de Janeiro is scheduled for a large load insertion study in 2026 aimed at roughly 4 GW. Rio Grande do Sul has a single project at MME with demand of up to 5 GW. The Northeast has a study underway to add 4 GW of large load connection capacity [EPE, Notícias].
Four gigawatts is the whole story
Of the 26.2 GW standing at MME, one tranche has been studied, priced and recommended: São Paulo, R$ 1.6 billion, about 4 GW. That is 15 percent of the queue, desk arithmetic.
Add everything EPE lists as expected or under study and the total reaches roughly 17 GW, about 65 percent of the queue, also desk arithmetic. The components are 4 GW recommended in São Paulo, 5 GW expected from further São Paulo works, 4 GW from the Rio de Janeiro study and 4 GW from the Northeast study. Rio Grande do Sul’s 5 GW is excluded because it is a project’s demand, not released margin. The Northeast 4 GW is the Pecém and Parnaíba package below, not a separate tranche.
Every gigawatt in that 17 except the first four is a study, not a work order.
Brazil angle
EPE’s Estudo Prospectivo para Inserção de Cargas Eletrointensivas na Região Nordeste, released in June 2026, proposes R$ 5.68 billion in transmission works to connect up to 4 GW at Pecém in Ceará and Parnaíba in Piauí. It is staged deliberately: R$ 1.09 billion for 2032, then R$ 4.59 billion released only as load confirms [ePowerBay, 22 June 2026].
The filed demand behind it is 31.8 GW across the Northeast, of which 23.5 GW is hydrogen industry and 8.3 GW is data centers. Ceará and Piauí alone hold 22.1 GW across 21 potential projects. Ten access requests in the Pecém and Parnaíba regions had already been denied through January 2025 for want of remaining capacity [ePowerBay, 22 June 2026]. The same account calls 22.1 GW “58 percent” of the regional total, which does not reconcile with its own 31.8 GW figure, so this desk carries the gigawatts and drops the percentage.
Data centers in the Northeast are not mainly competing with each other for margin. They are competing with electrolysers, and they are outweighed by nearly three to one.
The physical order behind the package is a new Pecém IV substation at 500 kV and 1,848 km of 500 kV line crossing Rio Grande do Norte, Ceará and Piauí, available from 2032, ahead of the structural Bipolo Nordeste II in 2034 [ePowerBay, 22 June 2026]. That is aluminium conductor, steel towers and substation transformers on a decade-long delivery schedule. EPE is explicitly weighing faster substitutes: dynamic line rating, FACTS devices such as SSSC, batteries, and reconductoring existing lines with HTLS thermal-resistant conductors [EPE, Notícias]. Each of those trades steel and copper for power electronics and lithium. The queue is a materials order either way, and the choice of instrument decides which material.
One date discipline note. The 26.2 GW is a November 2025 stock that reached headlines on 3 September 2026, not a September 2026 reading. The Redata surge inside it is nine months old.
US angle
The United States has the same disease at larger scale and is treating it procedurally rather than physically. In ERCOT, 198 GW of large load applied for interconnection in the first quarter of 2026 alone, with 86 GW under review, roughly equal to ERCOT’s current peak load. PJM faces a capacity shortfall that could reach 15 GW by 2030 [Ascend Analytics, 5 May 2026]. ERCOT’s single-quarter intake is more than seven times Brazil’s entire standing queue, desk arithmetic.
The American response has been to invent new classes of service so load can enter before wires exist. PJM created non-capacity-backed load and interim network integration transmission service, both of which accept curtailment ahead of traditional load. SPP offers CHILLS, seven-year non-firm service as a bridge, HILLGA, a generator study completed in under 90 days, and PALS for loads that can drop to zero within five minutes [Ascend Analytics, 5 May 2026].
Nothing in EPE’s São Paulo recommendations or its Northeast study offers a bridge of that kind. Both are built around wires arriving first.
China angle
State Grid plans to invest CNY 4 trillion across the 15th Five Year Plan covering 2026 to 2030, a 40 percent increase on the previous plan, and to commission 15 new ultra high voltage lines by 2030, lifting cross-provincial transfer capacity by 35 percent [Enerdata, 4 March 2026, citing State Grid via China’s Central Broadcasting Network, 28 February 2026].
The contrast is structural, and no currency conversion is needed to see it. China commits transmission on a five year cycle, at national scale, ahead of the load. Brazil commits it study by study, state by state, after the load files. Brazil’s own documents name the reason for that caution: build corridors for projects that never arrive and consumers pay for idle infrastructure [Cenário Energia, 3 September 2026; ePowerBay, 22 June 2026].
What it means
Brazil’s binding constraint on AI infrastructure is no longer tax or generation. The hardware tax regime exists. The matrix is heavily renewable. The constraint is connection margin, released in four gigawatt increments against a queue more than six times that size.
The 26.2 GW is intent, not contracts, and EPE says so plainly: economic viability, transmission availability, telecommunications and the new connection guarantee rules all sit between a filing and a build. But the asymmetry runs both ways. EPE will not build ahead of confirmed load, and developers will not confirm load without a connection date. The Política Nacional de Acesso ao Sistema de Transmissão and its access seasons are the intended tiebreaker, and they have not been tested against a queue this size.
One stability finding deserves pricing. EPE’s simulations found no operating violations connecting up to 4 GW at Pecém or Parnaíba across 24 contingencies, but sudden loss of load blocks above 2 GW produced sustained overvoltages at substations in both regions and loss of synchronism at the UTE Companhia Siderúrgica do Pecém [ePowerBay, 22 June 2026]. A gigawatt-class campus in the Northeast is a system event, not just a customer.
The desk indexes read TAI at 107.5, down 0.83 on the week, with the power sub-index TAI-P at 111.3 against materials at 106.2 [Tantalum indexes, updated 5 September 2026]. Power has been carrying the composite. Queues like this one are why.
What to watch
- The 2026 São Paulo transmission auctions. EPE expects further works to compose the next auctions and add about 5 GW of margin statewide. If they do not appear in a 2026 auction lot, the São Paulo answer stays at 4 GW into 2027.
- The Rio de Janeiro large load study. Scheduled for 2026, targeting roughly 4 GW. Watch for a publication date before year end, since a slip pushes any resulting works into the 2027 auction cycle.
- The next EPE queue disclosure. The 26.2 GW is a November 2025 reading. The first Pnast access season result is the test of how much of the 6.4 GW Redata surge was a project and how much was a placeholder.
