THE AI STACK AT ~$8T ARR BY 2030 Truist sees the AI economy shifting where value is captured as hardware monetizes buildout, cloud monetizes compute, models monetize intelligence and applications monetize the work AI ultimately replaces or improves: CLOUD LAYER • $AMZN selling AI infrastructure through AWS while pushing its own Trainium & Inferentia silicon to lower cost of training & inference inside its cloud • $MSFT using Azure to sell GPUs, model access & AI services while distributing OpenAI workloads through one of largest enterprise cloud footprints in the world • $GOOGL selling GPUs & TPUs through Google Cloud turning years of internally developed AI infra into an external compute business • $ORCL building OCI around large GPU clusters & dedicated AI campuses while winning workloads where customers want enormous blocks of compute rather than traditional cloud services • $CRWV is essentially selling Nvidia infrastructure as a specialized cloud by using large GPU clusters & long-term contracts to compete for highest intensity AI workloads • $NBIS is building an AI-native cloud around Nvidia GPU capacity, power & data centers with business becoming more designed around selling large dedicated compute deployments to model builders & enterprises • $IREN starts with the scarce asset most AI clouds need first, energized power & is converting that infrastructure into GPU clusters plus contracted AI cloud capacity • $NSCL is building vertically integrated AI infrastructure around GPU clusters, data centers & sovereign compute deployments HARDWARE LAYER • $NVDA owns accelerator platform at center of the stack through GPUs, NVLink, networking & CUDA which is why almost every cloud & model company above ultimately depends on Nvidia capacity • $TSM manufactures leading-edge logic behind Nvidia, AMD & custom AI accelerators making advanced process capacity one of most important physical bottlenecks in AI • $AMD is building primary merchant alternative to Nvidia through Instinct accelerators & EPYC CPUs giving hyperscalers another full-scale compute platform • $MU supplies HBM & high-performance DRAM where every increase in accelerator performance requires a ton more memory capacity & bandwidth • $SKHY has become one of most important HBM suppliers in AI supply chain by supplying high-bandwidth memory packaged alongside leading accelerators • $DLR owns & develops physical data center campuses where cloud providers & enterprises deploy increasingly power-dense AI infrastructure APPLICATION LAYER • $PLTR turning foundation models into production workflows through Foundry & AIP with value coming from deploying AI against an enterprise’s actual data & operations • $SNOW & Databricks sit at enterprise data layer using Mosaic AI & Snowflake Cortex to turn proprietary company data into the models, agents & AI applications enterprises actually deploy MODEL LAYER • $META building Llama underneath Meta AI & Muse by using its own models to power consumer agents across WhatsApp, Instagram & Facebook while Muse pushes Meta further into an agent that can actually take actions for users • $GOOGL developing Gemini through DeepMind while simultaneously distributing it across Search, Workspace, Android & Google Cloud • $SPCX sits in model layer through Grok & now owns a major application layer after acquiring Cursor for $60B giving it a vertically integrated path from massive GPU infrastructure to models to an AI coding product developers actually use • OpenAI & Anthropic are pushing model layer using GPT, Claude, reasoning, agents & products like Claude Code to turn foundation models into systems that can actually complete work