Photo: panumas nikhomkhai / Pexels
Meta Enters the GPU Cloud: Meta Compute Targets AWS and Azure With Hosted Models and Raw Compute
Chris Harper
2 min read
Jul 5, 2026 · 20:04 UTC
TL;DR: Meta announced a GPU cloud service on July 1 that sells raw compute and hosted AI model APIs, competing with Bedrock, Vertex, and Azure; Cursor acquired Continue and its export deadline is July 15.
On July 1, Meta confirmed Meta Compute, a new cloud business that will offer developers two things: raw GPU capacity (bare-metal and virtualized, powered by Nvidia H100s and Meta's custom MTIA chip) and hosted AI model APIs including the closed-weight Muse Spark. The company has committed $182.9B to AI infrastructure as of Q1 2026 — Meta Compute monetizes the excess. Meta stock jumped 8.8% on the announcement.
Why it matters: A fourth major cloud option for AI inference means more pricing pressure and another route for hosted-model spend. If you're currently routing inference through AWS Bedrock, Google Vertex, or Azure AI Foundry, Meta Compute's pricing and Llama-family access are worth tracking when it opens in July.
Separately: Cursor acquired Continue (the open-source coding assistant with 34K GitHub stars) on June 18, in an acqui-hire that shuts down the product and hands the codebase to the community. If your team runs Continue in any IDE, the export deadline is July 15.
Sources: Meta Compute — Bloomberg, Meta stock reaction — CNBC, Cursor acquires Continue — The New Stack