Meta Pops 10% on Cloud Push; CoreWeave, Nebius Sink 13%

Meta Platforms (NASDAQ: META) surged 9.68% to
$617.83 on July 1, 2026, its biggest
single-day gain in more than five months, after reports that the company plans to sell excess
capacity from its AI data-center buildout to outside enterprises. The move added roughly
$138 billion in market value in a single session, lifting Meta’s
market cap to $1.568 trillion, and it repriced the entire AI-infrastructure
leaderboard within hours.

The market’s read is straightforward: for the first time since AWS launched in 2006, a fourth
serious hyperscaler is stepping into the enterprise-cloud arena — and it is doing so from a
position of already-built, already-powered, already-cooled AI capacity. That framing is why the
reaction outside Menlo Park was so violent. CoreWeave (NASDAQ:
CRWV) fell 13.12% to $86.49 and
Nebius Group (NASDAQ: NBIS)
dropped 13.96% to $237.62. Between them, the two AI-neocloud specialists lost roughly
$17 billion in combined market value on the day.

What the reports actually say

Sell-side notes circulating this morning described a plan under which Meta would package a
slice of its AI training and inference capacity as a paid service for third-party developers and
enterprises, initially targeting frontier-model labs and large corporate machine-learning teams.
The pitch, as described in the reports, leans on three assets Meta has been building for its own
Llama and Reels-ranking workloads: proprietary silicon (MTIA), tightly coupled InfiniBand fabrics
running out of the company’s Prometheus and Hyperion campuses, and a global backbone
network already terminated inside every major carrier hotel.

Meta has not filed anything with the SEC confirming the launch, and the company declined to
comment on the reports. But the strategic logic tracks with the direction Mark Zuckerberg
telegraphed on Meta’s last two earnings calls, when he described the company’s
AI-infrastructure spend as “multi-year” and pledged to “build ahead of
demand.”

The capex math forced the pivot

Meta’s 2026 capital-expenditure envelope — explicitly aimed at AI compute —
runs into the tens of billions of dollars per quarter, on top of an already record 2025 outlay.
When you build ahead of demand at that scale, two states are possible at any given moment:
either the internal workload absorbs every rack, or you have expensive machines idling. Selling
surplus capacity to the market is the obvious pressure valve.

It is also the exact playbook Microsoft and Amazon have run for a decade. Microsoft’s
Intelligent Cloud segment, which houses Azure, reported
$34.7 billion in revenue in the March 2026 quarter, up 30% year-over-year,
with Azure itself growing 40%. AWS remains the segment leader, generating cloud revenue at a run
rate above $130 billion annualized. Alphabet’s Google Cloud crossed a $50 billion
annualized run rate last quarter. Meta walking into that market is not a peripheral experiment
— it is a bid to own the fourth seat at the hyperscaler table.

One important nuance: the popular story that AWS was born from Amazon’s excess
retail-computing capacity is mostly wrong. As
AWS’s own history
records, Andy Jassy’s original 2003 memo pitched AWS as an
“Internet operating system” of shared infrastructure primitives — storage,
compute, databases — that Amazon would build for internal engineers first and then sell.
The excess-capacity story is a founding myth. Meta’s reported move actually fits that
myth more literally than AWS ever did: purpose-built AI capacity that becomes salable the moment
internal demand can’t consume it.

Why CoreWeave and Nebius took the hit

CoreWeave and Nebius have built their equity stories on renting Nvidia GPUs at scale to
frontier-model labs and enterprise AI teams — the exact customer set Meta would target.
Both companies still carry premium multiples: CoreWeave’s $47 billion market cap and
Nebius’s $60 billion market cap embed multi-year revenue growth assumptions that
depend on landing and retaining large training contracts. Any hyperscaler entering with idle
capacity to burn and no revenue target for that capacity compresses those assumptions.

The counter-argument, which
Nebius bulls
continued to make, is that AI compute demand still runs well ahead of aggregate supply and that
purpose-built neoclouds still hold pricing power for the most contested SKUs. But even bulls
concede the price of that pricing power has shifted — from a duopoly-of-specialists to a
market with a fourth deep-pocketed generalist.

The day’s AI-infrastructure moves

What to watch next

Three questions determine whether Wednesday’s move sticks. First,
whether Meta confirms or walks back the reports on its Q2 2026 earnings call — and, if it
confirms, whether the go-to-market is a broad enterprise offering or a curated white-glove
program for a handful of frontier labs. Second, whether Nvidia signals any
change to allocation among CoreWeave, Nebius, and the hyperscalers; a shift in GPU allocation
would matter more to CRWV and NBIS than any pricing move. Third, whether AWS
and Azure discount reactively to hold enterprise workloads, a move that would shave the fat
margins that make cloud hyperscaling worth doing in the first place.

One data point the market didn’t fully price in on Wednesday: enterprise cloud sales
cycles are twelve to eighteen months long. Meta announcing a business today does not translate
into a revenue line for four quarters, at minimum. The compression in CoreWeave and Nebius
happens on the multiple; the compression in Meta’s cost of AI capital happens on the day
the market believes the story.

Sources

• Yahoo Finance — Meta Platforms (META)
• Yahoo Finance — CoreWeave (CRWV)
• Yahoo Finance — Nebius Group (NBIS)
• Microsoft FY26 Q3 Earnings — Intelligent Cloud & Azure revenue
• Amazon Web Services — founding history
• Meta Investor Relations — earnings and filings

Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.