just committed another $21 billion to CoreWeave (CRWV) — on top of the $14.2 billion it promised last fall. That pushes Meta’s bill with a single cloud provider to roughly $35 billion, locked in through December 2032. On the same morning, CEO Andy Jassy told shareholders the company plans to spend $200 billion on capex this year. Two data points. One conclusion: hyperscaler AI infrastructure spending has gone parabolic, and the picks-and-shovels names are the trade.
CoreWeave shares popped as much as 7% in pre-market Thursday before settling near $87.80, while Meta traded up to $625.26. The moves are modest relative to the news — and that’s the opportunity. The market is still digesting what $35 billion of locked, multi-year compute capacity from a single hyperscaler actually means for the only pure-play AI cloud operator at institutional scale.
The $21 Billion Contract Nobody Saw Coming
Here’s the thing about CoreWeave’s Meta deal: it wasn’t an expansion. It was a fresh order form signed March 31, disclosed this morning in an 8-K filing. The capacity runs through December 20, 2032, and Meta also exercised an option for additional compute tied to the 2023 master services agreement. Some of that capacity will include the first deployments of ’s Vera Rubin platform — the R100/R200 architecture Jensen Huang debuted at GTC 2026 in March.
CoreWeave CEO Michael Intrator kept it short: “This is another example that leading companies are choosing CoreWeave’s AI cloud to run their most demanding workloads.” What he didn’t say — and what matters more — is that the focus of the new contract is inference, not training. The industry is quietly pivoting from building models to running them at scale, and inference is where the durable, high-utilization revenue lives.
CoreWeave’s backlog now stands at roughly $66 billion before you even count this deal. Full-year 2025 revenue landed at $5.13 billion, up 168% year-over-year. Bank of America reinitiated coverage on March 24 with a Buy rating and a $100 target. The Yahoo-aggregated analyst consensus sits at $119.40, which implies 36% upside from the current share price. Even the bears at Bernstein — who carry a Sell with 35% downside — concede the backlog is real. They just worry hyperscalers will eventually build their own data centers and cannibalize the business.
That bear case ran into a brick wall this morning. Meta didn’t insource. Meta wrote the $21 billion check.
Amazon’s $200 Billion Confession
While CoreWeave was filing its 8-K, Jassy was dropping a bomb of his own. Amazon’s CEO used his annual shareholder letter to disclose that AWS’s AI revenue run rate has hit $15 billion in Q1 2026 — a number the company has never broken out before. He also confirmed the company plans to spend roughly $200 billion on capex this year, most of it on AI infrastructure.
The receipts Jassy laid out are the kind that make analysts update their models in real time. AWS added 3.9 gigawatts of new power capacity in 2025 and plans to double total capacity by the end of 2027. The custom chip business — Graviton, Trainium, Nitro — is now generating a $20 billion annual run rate, growing at triple-digit percentages. Trainium2 is sold out. Trainium3, which began shipping in January, is nearly fully subscribed. Trainium4, which isn’t broadly available for another 18 months, already has significant reservations.
And then there’s the OpenAI contract. Jassy confirmed a portion of the 2026 capex is already backed by an over-$100 billion commitment from OpenAI alone, with “several other customer agreements completed (and unannounced), or deep in process” — his words, pulled from the letter. On “We’re not investing approximately $200 billion in capex in 2026 on a hunch” — Jassy’s own framing of the thesis.
Amazon shares traded up to $225.20 Thursday morning, recovering part of the 7% selloff that followed February’s initial capex disclosure. Cantor Fitzgerald raised its price target to $260 from $250 yesterday. The broader analyst consensus sits at $281.27, implying nearly 25% upside.
How to Play It
The setup here isn’t subtle. JPMorgan CEO Jamie Dimon used his own annual letter to JPMorgan shareholders last week to forecast that five hyperscalers — Amazon, Microsoft, Alphabet, Meta, and — will collectively raise AI-driven capex from $450 billion in 2025 to $725 billion in 2026. That’s a 61% increase in a single year, and it all has to go somewhere. Here’s my read on the five names best positioned to catch the checks.
- CoreWeave (CRWV) — $87.80. The purest expression of the trade. 36% upside to consensus, a $66 billion backlog before today’s Meta deal, and now a multi-year validation from the most vocal AI capex skeptic-turned-customer on the list. The stock is down from its $187 October high, which is the kind of drawdown that usually precedes a rerating when a new contract of this magnitude lands. The risk is dilution — CoreWeave has leaned on convertibles and term loans to finance its buildout — but with Meta locked in through 2032, the revenue visibility finally matches the capex.
- Meta Platforms (META) — $625.26. The buyer, not the seller, of compute — but worth owning here anyway. Meta guided 2026 capex to $115–$135 billion in its February release, and the company has now publicly committed to spending that aggressively regardless of near-term earnings sensitivity. KeyBanc trimmed its target to $760 from $855 this morning citing the cost of the Llama 4 buildout, but the stock still trades at 19x forward earnings with $81.6 billion in cash and a $1.55 trillion market cap. The Yahoo consensus target of $860.25 implies 38% upside. Q1 earnings drop April 29.
- Amazon (AMZN) — $225.20. The stock has been in penalty box since the February capex reveal, down roughly 13% from its $258.60 52-week high. Jassy’s letter today reframes the narrative from “Amazon is burning cash” to “Amazon is securing a multi-decade moat with customer commitments in hand.” Forward P/E near 30x looks rich until you remember AWS operating income is accelerating and the AI revenue run rate just ticked over $15 billion from essentially zero three years ago.
- Nvidia (NVDA) — $182.08. The shovel seller behind everything. Meta’s new CoreWeave capacity will include Vera Rubin deployments. Nvidia’s Q4 FY2026 data center revenue hit $62.3 billion, up 75% year-over-year. Full-year FY2026 revenue was $215.9 billion, and Wall Street now expects $370 billion in FY2027 on the Rubin ramp. Huang disclosed at GTC that Nvidia sees $1 trillion in cumulative Blackwell/Rubin demand through 2027. The stock trades at roughly 22x FY2027 earnings — a forward multiple below Apple’s or Microsoft’s, on a company growing revenue 4x faster. Consensus target $268.22. That’s the cheapest the AI leader has been in years.
- Nebius Group (NBIS) — $123.79. The small-cap neocloud with a $27 billion Meta contract of its own and a 310 MW AI factory under construction in Finland. Nebius is up more than 450% over the past year and still trades at a 74% discount to Northland’s $215 target. The company raised $4 billion via convertible notes in late March to fund buildout. If CoreWeave is the obvious CRWV play, Nebius is the asymmetric one — higher beta, more runway, same core thesis.
The Bear Case
Not everything is clean here. H100 lease rates have surged roughly 40% from their October 2025 lows — a sign of tightness, yes, but also a sign the cycle is late. Goldman Sachs warned in December that hyperscaler capex growth is projected to slow from 75% year-over-year in Q3 2025 to 25% by the end of 2026, even as the absolute numbers climb. Bernstein’s short thesis on CoreWeave — that hyperscalers will eventually insource — isn’t wrong on a 10-year view; it’s just wrong this quarter.
The other risk is power. Vera Rubin racks consume 600 kilowatts each. New data center power capacity takes three to five years to permit and build. If the grid can’t keep up, GPU orders get delayed, not canceled — but the timing mismatch could bleed into quarterly numbers.
None of this changes the near-term setup. When five hyperscalers are collectively writing $725 billion of checks, you don’t want to be on the side of the trade that’s cashing paychecks. You want to be on the side cashing the checks.
What to Watch
Three catalysts are worth circling:
- April 29 — Meta Q1 earnings. Capex guidance gets its first public update since the CoreWeave deal. Expect the $115–$135 billion range to shift higher.
- April 29 — Nebius Q1 earnings. The first read on how the $27 billion Meta contract is flowing through revenue. Expect guidance revisions.
- May 20 — Nvidia Q1 FY2027 earnings. The Vera Rubin ramp is now underway and Jensen’s $1 trillion backlog number gets tested against actual bookings.
Wall Street spent Q1 arguing about whether AI capex had peaked. This morning, Meta and Amazon answered at the same time. The answer was no.