Brazil's data center boom is stuck behind a queue, not a shortage

Brazil's grid operator fielded requests for more than 20 GW of new connections from 367 projects in its first formal Access Season, spread across 64 substations. As of its most recent award round, in October 2026, ONS had granted firm connection to three data center projects. Combined capacity: 228 MW. Every press release about Brazil's AI build-out cites gigawatts. The grid operator's own ledger, so far, counts in megawatts.

Most coverage of this pipeline goes straight to the resource. Brazil has wind and solar output the grid cannot absorb, curtailment ONS itself projects could reach up to 40 GW a year between 2027 and 2030, and a hyperscaler sitting on top of that surplus looks like the cheapest clean power deal in the hemisphere. That framing skips the step deciding whether any of it gets built on schedule. Before a data center buys a megawatt-hour of curtailed output, it needs a wire into the substation serving it, and that wire is the scarce asset here.

Where the exposure actually sits

ONS's own numbers explain why the queue is the binding constraint here. Roughly 65% of the connection requests filed with the grid come from renewable generators themselves, per analysis from The Intercept Brasil and Mongabay. The generation meant to supply cheap curtailed power to a data center is competing for the same substation slots as the load that wants to consume it. Both sides of the deal are standing in the same line.

ONS has also started flagging a risk nobody writes into a power purchase term sheet. A hyperscale campus behaves like a generator in reverse. The planning director has warned publicly about the system-stability exposure of a sudden 2 GW trip if a large data center disconnects during a frequency event, the kind of contingency utilities historically reserved for a power plant tripping offline. That concern is already reshaping how ONS evaluates new requests, and evaluation runs slower than a press release.

Press releases cite gigawatts committed. ONS has awarded 228 megawatts. That gap is the whole story, and it lives entirely inside the interconnection queue.

The free market was supposed to clear the surplus

Brazil already built the mechanism meant to let a large consumer buy curtailed renewable output directly: the ACL, its free energy market. 2025 was its best year yet, adding roughly 21,700 new consumers and pushing free-market demand to about 43% of national electricity consumption, per market reporting. In the first two months of 2026, new sign-ups contracted by roughly 32%, as rising spot prices and curtailment-driven volatility made staying on the regulated tariff the safer bet, per CNN Brasil's coverage of the slowdown.

That is the market a data center developer is told will deliver cheap, curtailed power on demand. It just posted its sharpest contraction since the current migration wave began, for reasons tied directly to the curtailment and price volatility the pitch depends on. That resource should not be underwritten at the discount everyone is still quoting from eighteen months ago.

What we think follows

Brazil's data center story currently reads as a transmission-underwriting problem wearing an energy-abundance costume. The capital commitments are real, Microsoft's multi-year, multi-billion-dollar build-out and ByteDance's Pecém project among them, but capital committed and megawatts connected are different line items. Until ONS clears meaningfully more than 228 MW, the interconnection process, rather than the Northeast's curtailed output, is what sets this pipeline's pace.

The counterargument is real and worth stating plainly. REDATA, the law built to fast-track this build-out, is now in force, and ONS's planning director has said roughly 10 GW of data center demand already sits at some stage of connection progress, with about 3 GW under signed grid-use contracts. That is a far larger number than 228 MW, and if even a third converts to firm access over the next two years, the queue argument made here weakens considerably.

We don't have assets in Brazil. Our own pipeline runs through Peru, where COES's generation queue sits roughly twenty times deeper than the transmission capacity leading out of it, a mismatch we size debt around directly across 870+ MW of projects advancing toward a 2 GW+ target by 2030: Olmos Wind, approximately 135 MW in Lambayeque with battery storage integrated into the project design, targeting Ready-to-Build in the first quarter of 2028; Frontera, approximately 135 MW, RtB 2028; Aurelion, a 300 MW bifacial platform in Moquegua, RtB 2028; and Solar II, 300 MW, RtB 2029, alongside our Fitcon Energy platform. Brazil's data center queue is the same generation-ahead-of-wire mismatch, showing up on the demand side instead of the supply side.

The open question

The real test is timing. REDATA exists to compress a connection process that currently moves in months per project and years per gigawatt, and whether it compresses fast enough to match a capex clock already running toward 2027 and 2028 is something only the next two award rounds will show.

Genuinely open: does ONS turn its staged 10 GW of in-progress connections into firm, contracted capacity inside that window, or does Brazil's AI build-out end up gated by the same queue-versus-wire mismatch slowing new generation elsewhere in the region, just arriving from the load side this time?

See the full active portfolio or read more Insights.