Magic

Magic is pre-commercial: it sells no product today and has announced no pricing, customers, or revenue. The company operates as an equity-funded research lab — roughly $465M raised, including a $320M round in August 2024 — spending on a small research team and on compute through a Google Cloud partnership to build AI supercomputers. The plan, per its positioning, is to monetize frontier software-engineering automation: models built on the Long-Term Memory architecture that reason over entire codebases, which would most naturally be sold to enterprises as usage-priced API access or a per-seat engineering product. Nothing of the sort has been announced. Until something ships, the honest description is that Magic converts investor capital into research progress on ultra-long-context models, with no interim revenue bridge. The business model is a deferred bet: capture value later from automating the most expensive labor category in software, with the product form, buyer, and pricing all still undeclared four years in.