What Is OpenAI Racing to Achieve Under Compute Constraints?

OpenAI co‑founder and president Greg Brockman recently gave an interview covering product‑strategy consolidation, the new base model “Spud”, automated AI researchers, and the controversy over a hundred‑billion‑compute investment.

He rejects the view that OpenAI is retreating to B2B; instead the company is making a harsh priority ranking forced by a hard compute budget. The internal top‑priority list contains only two items: a personal assistant and an AI work‑agent that can solve difficult problems for users. Current compute is insufficient to support both simultaneously.

The video‑generation product Sora has not been shut down but its compute allocation was shifted toward the inference‑model branch of the technology tree, reflecting the compute‑first choice.

Brockman describes the core product direction as building a unified “AI layer” that merges Chat, Codex and browser‑automation into a single entry point, replacing the current fragmented tool ecosystem. He illustrates this with a personal example: he often forgets how to set macOS hot‑corners, and now Codex can configure them directly, showing the machine adapting to the user.

Codex is planned to evolve from a tool for engineers into a universal operation portal; third‑party developers would only need lightweight plugins to integrate.

Regarding the next‑generation base model, Brockman says Spud condenses roughly two years of research into a new pre‑training foundation. It is not an incremental upgrade but a qualitative leap in instruction understanding, contextual grasp and handling open‑ended problems. He expects significant breakthroughs in scientific domains such as physics.

He cites the industry term “big‑model smell” to describe the phenomenon where, once a model crosses a capability threshold, users perceive the AI as bending to their intent and experience far fewer misinterpretations.

Brockman predicts that the new model will raise AI task‑coverage from about 20 % to roughly 80 %, turning AI from an auxiliary aid into a core element around which work processes must be redesigned. [2-1]