The rising infrastructure costs of artificial intelligence are forcing corporations to rethink their reliance on third-party API providers. Jack Dorsey’s Block has announced an AI model called Dessa, stepping away from standard vendor ecosystems to test the economics of proprietary foundational tech.
Developing a custom foundational model requires heavy upfront capital, yet companies are actively seeking ways to cut ongoing operational dependence on external vendors. Block trained this model from scratch on specific datasets, fundamentally altering its compute cost structure.
Dessa is a “true speech-to-speech model, and everything it knows came from human speech,”
This approach forces a direct financial comparison between owning proprietary infrastructure assets and paying recurring rental fees for external computing power and commercial interfaces.
Architecture Versus Market Standards
Traditional approaches to training language models rely on massive web-scraped text corpora. In contrast, Dessa is trained from scratch on nothing but human conversation, bypassing standard internet repositories entirely.
These engineering choices manifest directly in the team’s demonstration materials. As noted by Dessa lead Ragavan Thurairatnam, dialogues with the system exhibit unconventional behavior, including moments where the model initially mistakes the speaker for somebody named “Ronaldo.”
Such architectural choices raise immediate questions about the behavioral predictability of autonomous systems in core business scenarios, where the polished, frictionless responses delivered by OpenAI and Anthropic have become the established enterprise norm.
Infrastructure Constraints and Security
The practical deployment of the project introduces strict hardware constraints for local execution. The software is engineered so that it can run locally on a MacBook Pro M4 Max “or better,” shifting hardware overhead directly to the end deployment environment.
The development team frames its project goals around specific internal priorities. According to the creators, the team pursues three core motivations: “curiosity (aka fun), better understanding between humans and AI, and safety,” highlighting a research-driven rather than strictly commercial roadmap.
To validate these parameters under controlled conditions, the Dessa team states it will be granting “a small group of AI safety researchers” early access to the model, keeping the circle of external participants strictly limited during the testing phase.
This marks a clear departure from universal templates in favor of specialized, experimental infrastructure without guaranteed short-term payback in the fintech sector.