The Power Bottleneck Is Pushing Data Centers to Brazil

The real constraint on data center companies right now isn't capital or customer demand. It's power.

In the United States, nearly half of the AI data centers planned for 2026 are facing delays or cancellation because transformers, switchgear, and grid interconnections aren't available. Interconnection wait times average five years. About 2,300 gigawatts of generation and storage capacity are stuck in U.S. interconnection queues — more than the entire installed generating capacity of the American power grid. Of the projects that submitted interconnection requests between 2000 and 2019, only 13% reached commercial operation by the end of 2024. Seventy-seven percent were withdrawn.

Communities have filed lawsuits against data center operators for noise, vibration, water consumption, and electricity cost pass-throughs. The legal cases are early-stage, but they add another layer of siting friction to a pipeline already strangled by grid constraints.

The data center companies that are best positioned to build aren't waiting. They're looking at countries that are actively clearing the path — and Brazil is moving first.

Brazil's Senate is voting this week on REDATA, a special tax regime that suspends four major federal levies — import duty, IPI, and PIS/Cofins — on data center equipment for five years, converting to a full exemption once conditions are met. Companies must reserve 10% of processing, storage, and server capacity for domestic use and commit to 100% renewable energy. The tax authority estimates roughly $1 billion in forgone revenue for 2026, falling to roughly $200 million annually after that. The government's stated ambition: $350 billion in data center investments over the next decade.

This matters because the same companies building hyperscale data centers in the United States — Equinix, Digital Realty — have already positioned themselves in Brazil before the vote. Equinix committed $234 million to new facilities in São Paulo and Rio de Janeiro this year. Digital Realty has announced expansion there as well. They're not betting on the policy. They're building ahead of it.

The reason this contrast between the U.S. and Brazil exists isn't political ideology. It's a straightforward difference in who's asking whom for permission.

In the United States, data centers face a pincer. On one side: grid utilities overwhelmed by demand and unable to clear interconnection queues faster than developers can file. On the other: local communities that see massive facilities consuming water, straining power lines, and raising their own electricity bills — while creating fewer than a dozen permanent jobs.

Brazil has the opposite problem. It has abundant renewable energy — hydropower, wind, solar — and direct subsea cable connections to global markets through systems like EllaLink and Firmina. Brazil accounts for 48% of Latin America's installed data center capacity with roughly 1.6 gigawatts in operation, but energy demand is projected to reach 13.7 gigawatts by 2035. The gap between current capacity and projected demand is enormous. The government's calculation is simple: if they remove the tax barrier on equipment imports, the money will flow.

The REDATA program has already stumbled once. It was originally issued as a provisional measure in September 2025 but lapsed in February 2026. It was reintroduced as an ordinary bill, passed the Chamber of Deputies on February 25, and is now in the Senate during a concentrated voting window that ends September 4 — literally days from now, before the October presidential election. The timing isn't accidental. President Lula needs a visible economic achievement heading into the election.

For investors, the question is whether this actually reaches the financial statements of the companies that dominate the data center trade.

Equinix (EQIX) trades at roughly $1,047 with a $103 billion market cap, up 37% year-to-date. The trailing P/E is 67 and EV/EBITDA is 28 — premium valuations for a company spending $6.5 billion on capital expenditures over the trailing twelve months against $3.9 billion in operating cash flow, producing negative free cash flow of $2.6 billion. Revenue grew 9.6% year-over-year, but the Q2 2026 report missed both EPS and revenue estimates. Debt sits at $26.7 billion.

Digital Realty (DLR) at $186 carries a $69 billion market cap, up 20% YTD. Its revenue growth of 17.4% nearly doubles Equinix's, and it beat Q2 EPS significantly — $1.21 versus a $0.71 forecast — though revenue came in below estimates at $1.9 billion. DLR's EV/EBITDA of 29 is roughly in line with Equinix's.

Both companies are spending aggressively to build capacity that customers have committed to, but the pipeline problem in the U.S. means a growing share of their capex is going overseas. That's not a risk — it's the whole point. The data center business model depends on securing power, building, and filling space. If the power is in Brazil, the building goes in Brazil.

The investment question is whether the valuation already reflects this geographic arbitrage. A 67x P/E on Equinix and a 29x EV/EBITDA on both operators price in sustained growth, margin expansion, and flawless execution. The Brazil market — valued at $4 billion today and projected to grow at roughly 9.5% annually — is a meaningful incremental opportunity, but it's not the kind of market that transforms these companies on its own. The U.S. remains the dominant revenue base.

What Brazil does change is optionality. If the grid bottleneck persists for years and domestic siting becomes harder, having a viable alternative market with government incentives and abundant cheap renewable energy gives these companies room to operate that wouldn't exist otherwise. It's not a valuation driver today. It's an insurance policy against a constraint that could otherwise throttle growth.

The real test for investors isn't whether REDATA passes — the Senate vote is all but assured given the political timing — but whether the grid bottleneck in the United States eases before the next data center cycle turns down. Nearly 2,000 gigawatts stuck in queues doesn't move overnight. Transformer lead times of two to three years don't compress on command. If domestic construction slows further, the companies with international pipelines — Brazil, Mexico, Southeast Asia — will capture a larger share of hyperscaler capex simply because they can deliver where others can't.

The data center stocks have run hard this year. The question going forward is whether the companies that can actually build earn their premium, or whether the constraint becomes so binding that it reshapes the competitive landscape in ways the current multiples don't reflect.