The Rise of Action-Oriented AI
Artificial intelligence is entering a transformative phase as consumer-facing AI agents evolve from simply answering questions to proactively performing tasks on users’ behalf. According to Goldman Sachs Research, this evolution could fundamentally reshape everyday digital activities, including shopping, search, travel, and advertising.
This shift marks a significant move from experimental AI applications towards widespread commercial integration, with companies embedding AI deeply into both their internal operations and consumer products. Instead of merely recommending a hotel or flight, a future AI agent could autonomously search, compare, and book these services, or even manage calendars and purchases.
From Conversation to Action: A New Paradigm
Goldman Sachs Research describes this as a transition from a “conversational relationship” with AI to an “action-oriented relationship.” However, this increased autonomy comes with a critical requirement: trust. For AI agents to operate independently, they will likely need access to sensitive information such as calendars, passwords, and payment details. The speed at which agentic AI becomes mainstream will heavily depend on robust security measures and consumer confidence.
If this model gains widespread acceptance, AI agents could effectively establish themselves as a new intermediary layer between users and the broader internet, streamlining countless digital interactions.
Reshaping Online Shopping & Advertising
The advent of AI agents could profoundly alter online shopping experiences. Consumers might no longer need to navigate multiple websites, compare products, and complete transactions manually. Instead, an AI agent could manage much of this process, creating a new AI platform layer above existing technology infrastructure and consumer applications.
Goldman Sachs anticipates that many mass-market AI agents will adopt existing internet business models, offering free services supported by advertising, alongside premium subscriptions for advanced features or ad-free experiences. Advertising itself is also poised for greater automation, with AI enhancing ad creation, targeting, placement, and performance measurement, potentially boosting efficiency and reducing costs across the industry.
Massive Infrastructure Investment & Challenges
This boom in consumer AI agents is driving a parallel, massive build-out of AI infrastructure. Goldman Sachs Research projects that U.S. hyperscalers will maintain elevated capital spending through 2027, potentially deploying an estimated $1.4 trillion in capital by that year.
Despite this substantial investment, the expansion faces significant constraints. Challenges include shortages of memory chips, limitations in electricity supply, land availability, and other critical components of the supply chain. The bank specifically expects semiconductor shortages to persist in the near term, citing the approximately three years required to construct a new chip fabrication facility.
The Cost-Usage Dynamic & Safety Concerns
A crucial factor fueling AI adoption is the declining cost of AI inference, which involves processing AI models. Goldman Sachs anticipates that cheaper token processing will make increasingly complex AI applications economically viable. Earlier research indicated that token consumption could surge 24-fold between 2026 and 2030, potentially reaching 120 quadrillion tokens monthly as both consumer and enterprise adoption of agentic AI expands.
This combination of falling inference costs and rapidly rising usage is expected to accelerate AI adoption and improve the economic viability of running large-scale AI services. However, this expansion occurs amidst ongoing debates about the pace and safety of frontier AI development. While discussions at Goldman Sachs’ technology conference touched upon potentially slowing development, the bank currently does not expect these debates to significantly alter its outlook for the AI infrastructure investment cycle through 2027, noting that the ultimate impact will largely depend on how regulation is designed and enforced.