Overview

• OpenAI and Broadcom signed a multi-year pact under which OpenAI will design chips and networking gear while Broadcom develops and manufactures them for AI data centers.
• Server racks using the custom hardware are scheduled to begin deployment in the second half of 2026, with implementation expected to run through the end of 2029 toward roughly 10 GW of capacity.
• OpenAI said the Broadcom agreement has no equity or investment component, and Broadcom will deploy racks into facilities operated by OpenAI or its cloud partners, with financing details not disclosed.
• The Broadcom plan follows large moves with Nvidia (reported up to $100 billion and at least 10 GW of systems) and AMD (a 6 GW supply deal), plus a letter of intent for a Stargate Argentina data-center project of up to $25 billion and 500 MW.
• Analysts highlight a shift to maximizing intelligence per dollar and warn of constraints in power, water and permits, while the IMF sees potential market corrections as unlikely to trigger a systemic crisis and studies report early declines in entry-level hiring alongside uneven corporate adoption in Brazil.