OpenAI's historic week has redefined the AI arms race (www.cnbc.com)

OpenAI this week announced a string of blockbuster moves that reposition it from model-maker to aspiring hyperscaler: an up-to-$100 billion, multi-year Nvidia investment (delivered in ~$10B tranches), an expanded Oracle/SoftBank "Stargate" buildout that could scale to $400 billion, and a formal integration with Databricks to surface OpenAI’s next-gen models (GPT-5) directly inside enterprise data tooling. CEO Sam Altman said OpenAI plans to spend “trillions” on data centers to meet surging demand; internal forecasts cited in reports project $125 billion revenue by 2029. The Databricks deal signals deeper commercial adoption as enterprises embed foundation models into workflows, while Databricks itself will continue to offer multiple providers (OpenAI, Anthropic, Gemini) to customers. Technically and strategically, these announcements make compute and energy the central battleground: OpenAI aims to ramp capacity by many millions of GPUs and roughly 17 gigawatts of power (equivalent to ~17 nuclear plants), highlighting acute constraints—limited grid capacity, slow permitting, gas turbines sold out through 2028, and long lead times for nuclear. Execution risk is high: OpenAI is not yet profitable, relies on outside capital, faces complex financing choices (debt, leases, equity), and depends on partner supply chains. If successful, this buildout could cement compute-driven scaling as the primary lever for future AI progress; if it falters, the circular dependency between chip suppliers, cloud partners, and OpenAI could strain the broader model economy.