Adobe launches a custom GenAI “foundry” for enterprises

Adobe unveiled Adobe AI Foundry, a build-with-us program for brand-specific generative models that sit on top of Firefly. The service will produce models for text, images, video and 3D, trained first on licensed data and then fine-tuned on a client’s IP, with usage-based pricing (not per seat). Adobe also put a scale marker on Firefly: since its 2023 debut, enterprises have generated 25+ billion assets—useful proof there’s enough demand to justify going from a one-size model to made-to-measure fittings. 

Two board-level asks dominate enterprise AI right now: predictability and permission. Predictability comes from models that understand a brand’s visual and verbal constraints; permission means licensed training sets plus a paper trail that legal, procurement, and insurers won’t choke on. If Adobe converts even a single-digit percentage of those 25B assets into foundry deals, Creative Cloud revenue shifts from static seats to metered, workload-tied ARR—and that’s where margins like to live. 

The market’s moving from “AI as a button” to “AI as a factory.” Salesforce bundles AI to CRM; Meta ties it to ads; Adobe is fusing it to content supply chains—style guides in, market-ready variants out. The bet is that brand-safe output beats raw horsepower. If Foundry lands, the switching cost becomes cultural (new workflows, new review gates), not just technical. 

If Adobe can productize bespoke without bespoke margins, it graduates from “AI inside” to AI supplier of record for Fortune-500 content ops. Watch for disclosures on customer count, model-build lead times, and the ratio of usage to seats.