Meta’s GPU Rental Plan Threatens CoreWeave and Nebius Growth

Published by James Harris on July 2, 2026July 2, 2026

Meta’s stock jumped 8.81% on July 1 to close at an all-time high of $796.25, while the Nasdaq Composite fell 0.66% on the same day. The divergence is the story: investors are now pricing Meta not just as an advertising business but as a potential compute vendor.

The pivot matters because Meta has spent years building AI infrastructure exclusively for internal use, and the scale of that buildout has become a liability narrative. Meta expects capital expenditures of up to $145 billion in 2026, part of a collective $725 billion that Meta, Microsoft, Alphabet, and Amazon are expected to deploy on AI infrastructure this year. Investors have been asking, with increasing impatience, what the return on that looks like. Renting excess GPU capacity to outside developers answers that question with something concrete, even if the product has no formal launch date yet. The model being considered reportedly resembles either AWS Bedrock’s API-access approach or CoreWeave’s raw GPU cluster rental, which puts Meta squarely in a market that already has established pricing and enterprise relationships.

The companies most exposed are not the hyperscalers. Gil Luria of D.A. Davidson was direct: CoreWeave and Nebius rely heavily on Meta for growth, and if Meta retains that capacity internally or sells it directly, those firms lose a customer and gain a competitor simultaneously.

This dynamic has a parallel worth keeping in mind. Bitcoin miners who pivoted toward AI data center workloads did so precisely because hyperscaler capacity was constrained and neoclouds were filling the gap. Meta entering that gap compresses the opportunity that made those pivots attractive in the first place. The same logic applies to the broader AI infrastructure stack: when a major consumer of high-bandwidth memory and GPU clusters starts offering excess capacity externally, it changes the demand signal for everyone downstream. Oracle’s recent moves suggest the same pressure is reshaping how signed but undelivered infrastructure contracts get interpreted, with remaining performance obligations now carrying more weight than headline capex figures.

At the May 27 shareholder meeting, Zuckerberg described the inbound interest plainly: companies approach Meta nearly every week asking for API access or compute they would pay a premium for. His framing at the time was that Meta expected to use the capacity itself, but that selling excess compute was on the table if the company had overbuilt. The stock’s reaction on July 1 suggests investors now believe the overbuild scenario is the base case, not the hedge.