Goldman Sachs Tech Conference Signals AI Investment Shift From Compute Buildout to Compute Usage

At the Goldman Sachs Communacopia + TMT Conference held in September 2026, executives from major technology companies shared their latest views on the AI industry. The clearest signal from the conference was that the focus of AI investment is shifting from building compute infrastructure to using it, while compute supply remains severely constrained.

Nvidia CEO Jensen Huang said 70% growth is a supply ceiling, not a demand ceiling, and maintained the company's expectation of roughly 70% year-over-year growth in calendar 2027. He said the core bottleneck limiting AI growth has shifted from whether demand exists to whether the physical world can keep up, citing widespread constraints in packaging, wafers, memory, and downstream data center power and site selection.

SpaceX CFO Bret Johnsen disclosed a more aggressive compute expansion plan: raising AI compute from 2 gigawatts on the ground to 5-10 gigawatts by 2027, and deploying AI satellites to bring orbital compute costs down to parity with ground-based levels.

Broadcom CEO Hock Tan broke down the economics of compute investment, saying 1 gigawatt of Frontier compute corresponds to $30 billion in annual recurring revenue, with $10 billion as cost and the remaining $20 billion captured by the model and application layers, while another $10 billion is contested across the cloud, chip, memory, and power supply chain.

Microsoft CFO Amy Hood and other executives also shared views on agentic AI and cybersecurity. Jensen Huang named cybersecurity as AI's next major application scenario, arguing that security agents require continuous 7x24 inference, so customers would pay for ongoing monitoring rather than intermittent usage. Overall, the conference showed the AI industry moving from a compute expansion narrative toward discussions of token monetization and value distribution.