AI industry's priorities: the race is increasingly moving from simply building larger and more capable models toward making AI cheaper, faster and more efficient. Rudina Seseri, founder and managing partner of Glasswing Ventures, argues that the compute-heavy and data-intensive nature of modern AI creates a major economic challenge. As models become increasingly capable, companies need to focus on how much intelligence they can deliver for each dollar of compute rather than assuming that scaling models indefinitely will remain the best strategy.
A major part of that efficiency equation is data quality. The article argues that carefully curated, relevant data can be more valuable than simply increasing the volume of training data. Companies are therefore exploring approaches such as smaller specialized models, better data preparation, inference optimization and custom AI chips. The goal is to extract more useful performance from existing resources rather than continually increasing the size and computational requirements of models.
This has implications for the AI business model itself. As foundation models become more accessible and inference costs fall, competitive advantage may increasingly move toward how companies integrate AI into products and workflows. Seseri points to a transition from traditional SaaS toward AI-embedded and outcome-oriented software, where customers pay for the value an intelligent system creates rather than simply for access to a piece of software. This also makes efficient inference and high-quality enterprise data strategic assets.
The broader message is that the next phase of AI competition may be an efficiency race rather than a pure intelligence race. Companies that can combine high-quality proprietary data, efficient models, specialized hardware and well-designed applications could outperform companies that simply spend more on ever-larger models. In that environment, data becomes a competitive moat, while model capability increasingly becomes one component of a much larger system. The question shifts from “Who has the biggest model?” to “Who can turn the best data and compute into the most useful intelligence at the lowest sustainable cost?”
