Wire

Meta said it had spare compute. Then it broke ground on a $9 billion data center in Canada (META, CRWV, NBIS)

Meta said it had spare compute. Then it broke ground on a $9 billion data center in Canada (META, CRWV, NBIS)

Key points

  • Meta (META) broke ground July 8 on a C$13 billion (about US$9.2 billion), 1-gigawatt data center in Alberta, its first in Canada and its largest anywhere outside the US.
  • One week earlier, a report that Meta might sell its "excess" compute sank CoreWeave (CRWV) about 14% and Nebius (NBIS) 17% in a single day.
  • The neoclouds snapped back Wednesday, a rally already underway before Meta's news broke late in the session: NBIS closed up about 11%, CRWV and IREN up roughly 8% each, while META slipped 2%.
  • The new site needs so much electricity that Meta is underwriting a brand-new gas plant. That is a demand signal for the whole AI-infrastructure group, power included.

A week ago, an unconfirmed report that Meta Platforms (META) might end up with more AI computing power than it needs erased billions of dollars from the neoclouds that rent compute to Meta. Meta would not confirm the report. On Wednesday, the same Meta broke ground on a C$13 billion data center in Alberta and signed up for an entire new power plant to feed it. A company that genuinely had spare capacity would not need to build more of it.

What Meta is building in Canada

The new site sits in Sturgeon County, Alberta, just north of Edmonton. It is Meta's first data center in Canada, its 33rd overall, and it will be the company's largest anywhere outside the United States. The numbers are hard to picture. The campus will span 2.9 million square feet and draw about 1 gigawatt of power, roughly what 800,000 homes use. Total investment is C$13 billion, about US$9.2 billion, according to BNN Bloomberg.

The power side is the detail that matters most. Alberta's grid does not have a spare gigawatt lying around, so Meta signed a long-term deal with Pembina Pipeline, Morgan Stanley Infrastructure Partners, and Kineticor Asset Management to take the output of a new $4.6 billion natural gas plant, the Greenlight Electricity Centre, planned for the same county with startup targeted for the second half of 2030. Meta says it wants its own data center running sooner than that. When a company needs electricity badly enough to fund a power plant from scratch, and still expects to outrun its construction schedule, that tells you more about its real compute appetite than anything said on an earnings call.

The week that made this awkward

Rewind to July 1. Bloomberg reported, citing unnamed people familiar with the matter, that Meta was building a cloud business to sell its excess AI computing capacity to outside developers, weighing whether to offer hosted access to its own AI models or sell raw computing power outright. Meta declined to comment on the report. That mattered less than it should have. META stock jumped almost 9 percent anyway. The companies on the other side of the trade had the worst day of their public lives. CoreWeave fell 13.9 percent. Nebius dropped 17 percent and lost about $12 billion in market value in one session. Chip names from Nvidia (NVDA) to Super Micro (SMCI) sold off with them. We covered the damage that day in our July 1 piece on the neocloud selloff.

The report was not pulled from thin air. Five weeks earlier, at Meta's annual shareholder meeting on May 27, Zuckerberg had already said a cloud business was "definitely on the table," telling investors that outside companies regularly ask to buy compute from Meta and that if Meta ever feels it has overbuilt, "then that is an option that we have." So the idea was real and on the record. The specific plan Bloomberg described, an internal effort reportedly called Meta Compute with actual product decisions being debated, was the part Meta would not confirm.

The fear was simple. Meta is one of the biggest customers these companies have. CoreWeave (CRWV) has disclosed roughly $35 billion in Meta commitments, a $14.2 billion contract from 2025 plus a commitment of about $21 billion signed this April. Nebius (NBIS) landed a Meta agreement in March worth up to $27 billion. If Meta has so much compute it can sell the surplus, the thinking went, maybe it stops renting. And maybe it becomes a competitor instead.

The week in between was not quiet, though the causes shifted. Chip and memory stocks kept sliding, just not because of Meta anymore. A separate scare hit overnight from Korea on July 2: the Kospi fell almost 8 percent and Micron gave back a strong morning open to close down about 5 percent. Then on July 7, Samsung posted a genuine earnings beat and its stock, along with Micron, SanDisk, and Western Digital, sold off anyway. The VanEck Semiconductor ETF (SMH) fell roughly 6 percent for the week. It is worth being precise about what that was and was not: chips and memory had a rough stretch, but the AI trade broadly did not. Big-cap winners like Microsoft, Apple, Amazon, and Alphabet closed higher the very day the Meta report hit, and IBM climbed about 12 percent over that same week while chips fell. Wednesday's bounce in Nvidia and Super Micro reads less like a one-day Meta reflex and more like chips finally catching a bid after several days of separate bad news.

So which is it, too much compute or not enough?

Here is the resolution, and it is not really a contradiction. Meta is not overbuilt by accident. Overbuilding is the plan. Zuckerberg unveiled a top-level initiative called Meta Compute in January, targeting tens of gigawatts of AI infrastructure this decade and eventually hundreds. The company raised its 2026 capital spending guidance to a range of $125 billion to $145 billion, up from a prior range of $115 billion to $135 billion, and up from about $72 billion spent in 2025. Selling whatever capacity Meta's own AI does not use was always part of that plan, per Zuckerberg's own May comments. It just took a leaked report, not a press release, to put a number and a name on it.

Read that way, the Alberta groundbreaking is not Meta walking back the excess-compute story. It is Meta manufacturing more inventory for it. Whatever its own AI models do not use, it plans to rent out, the way Amazon turned spare retail infrastructure into AWS. The point we made the day after the selloff still holds: a company that genuinely believed the AI buildout was ending would not commit another $9 billion and a gas plant to it.

Wednesday's scoreboard: a one-week reversal

One timing detail belongs up front, before the numbers below: Meta's announcement broke in Calgary around 3:40 to 3:56 p.m. ET, five to twenty minutes before the 4 p.m. close, based on the timestamps on Pembina's press release and BNN Bloomberg's report. That is too late in the day to have driven the full session. Nebius was already up about 5 percent and IREN about 4 percent when we covered the AI-infrastructure bounce earlier Wednesday, a rally already running for its own reasons before Alberta ever came up. The honest read: the sector was already rallying, and the Alberta news landed in the closing minutes as a fresh reason to keep buying, not as the cause of the closing numbers below.

Numbers aside, the reversal from one week to the next is real. The group that got hit on July 1 finished Wednesday sharply higher, and the stock doing the spending finished lower.

StockWednesday closeMove
Nebius (NBIS)$216.75+11.0%
IREN$43.00+8.0%
CoreWeave (CRWV)$90.03+7.8%
Super Micro (SMCI)$28.18+7.3%
Nvidia (NVDA)$204.14+3.7%
Meta (META)$603.03-2.0%

META slipping 2 percent while its suppliers rallied is still the mirror image of July 1, when META popped and everyone else bled. Just call it a same-week reversal, not a same-headline one.

Why this helps the whole sector

The bigger takeaway is about power, and it does not depend on the exact minute Meta's news broke. Meta could not find a spare gigawatt on the grid, so it is paying to create one. Multiply that across every hyperscaler racing to the same target and you get the demand backdrop for independent power producers like Constellation Energy (CEG), which closed up 2 percent Wednesday, plus Vistra (VST) and Talen (TLN), which finished roughly flat to slightly lower, since an Alberta-specific bump would only have had the last few minutes of the session to show up. The structural case is the same one we laid out in our piece on AI and nuclear power: compute is no longer the bottleneck. Electricity is.

For the neoclouds, the honest read has two sides. The bullish side is that Wednesday confirmed demand for AI capacity is still growing fast enough that the biggest spender on earth needs a new power plant to keep up. The cautious side is that the July 1 fear was never fully wrong. Meta building more capacity of its own does eventually mean Meta needs to rent less of it, and a future Meta cloud would compete directly with CoreWeave and Nebius for the same customers. That risk did not disappear this week. It just stopped being the only thing anyone could see.

Earnings watch

Meta (META) is expected to report second quarter results around July 29, after the market close, per our earnings calendar. The date is an estimate until Meta confirms it. Capital spending guidance will be the first number everyone checks.

Frequently asked questions

Why is Meta building a data center in Canada?

Meta broke ground July 8, 2026 on a C$13 billion (about US$9.2 billion) data center in Sturgeon County, Alberta, its first in Canada and its largest outside the US. The 1-gigawatt site will draw roughly the power of 800,000 homes, and Meta signed a long-term deal with Pembina Pipeline for a new natural gas plant to supply it.

Did Meta itself say it has excess compute, or was that just a media report?

Both, at different times. At Meta's May 27, 2026 shareholder meeting, Zuckerberg said on the record that a cloud business was "definitely on the table" if Meta ever felt it had overbuilt. The specific plan that crashed neocloud stocks on July 1, an internal effort reportedly called Meta Compute, came from a Bloomberg report citing unnamed sources, and Meta declined to comment on it directly.

Did Meta contradict itself on having excess compute?

Not really. Meta raised 2026 capital spending guidance to $145 billion and intends to rent out whatever its own AI does not use, similar to how Amazon built AWS. The Alberta build shows the excess-capacity plan is deliberate, not a sign Meta overspent by accident.

Why did CoreWeave and Nebius stocks drop on July 1, 2026?

A Bloomberg report that Meta would sell excess AI compute raised fears the company could rent less from its suppliers and eventually compete with them. CoreWeave (CRWV) fell 13.9% and Nebius (NBIS) fell 17% that day. Meta has roughly $35 billion in disclosed commitments to CoreWeave and a deal worth up to $27 billion with Nebius.

Is Meta's Alberta data center bullish for AI stocks?

It is a strong demand signal, and neocloud stocks rallied the day it was announced, with Nebius up about 11% and CoreWeave up almost 8%. But it is not a guarantee for any single stock, and the risk that Meta eventually competes with the neoclouds it currently rents from has not gone away.

Dennis Singleton
Dennis Singleton

Dennis Singleton has followed the markets closely for years and still finds them genuinely fascinating. He writes about stocks, AI, and semiconductors in plain language, cuts through the hype, and is straight about the risks as well as the upside. He does this because he wants readers to win.