Full-stack AI cloud betting on owned compute -- infrastructure advantage or capital-intensity trap?

Altis View

Together AI is both a beta bet on AI inference growth and an alpha bet on owned compute. The beta is straightforward: if AI usage keeps growing, production inference volumes rise, and open-weight models gain share, Together AI should benefit alongside the broader managed inference category. The alpha comes from its infrastructure strategy. By owning or controlling more GPU capacity, Together AI is positioned to offer better availability, performance, reliability, and margin control than more asset-light peers. The tradeoff is that this alpha also creates greater downside risk. Together AI is not just monetizing inference software; it is underwriting sustained demand for compute. If AI usage slows, the risk could become existential: Together AI would face not only weaker growth, but also the burden of underutilized GPU capacity and a more capital-intensive cost structure than asset-light inference peers. In that scenario, the same owned-compute strategy that drives upside in a supply-constrained market would become the source of structural downside.