OpenAI's Windows Play (stratechery.com)

OpenAI has rolled out a flurry of coordinated moves—massive data‑center deals (including an Oracle buildout), a $100B investment and 10 GW chip commitment tied to Nvidia, a separate multibillion-dollar pact to buy 6 GW of AMD MI450 chips (plus warrants for up to ~160M AMD shares), partnerships with Samsung and SK hynix for memory, the Sora 2 video‑generation model and app, GPT‑5 Pro/API availability, Codex GA, AgentKit and a new Instant Checkout product. Most consequential: OpenAI now embeds third‑party apps directly inside ChatGPT (Booking, Canva, Coursera, Expedia, Figma, Spotify, Zillow now; DoorDash, Uber, Target soon), turning the chatbot into an OS‑like platform where apps run within conversational context. For the AI/ML community this signals a pivot from isolated models and hardware rivalry to platform dominance—OpenAI is aggregating users first, forcing developers to integrate into ChatGPT to reach those users, and using scale to secure diverse chip supply to reduce dependence on a single vendor. Technically, the deals lock in multi‑gigawatt inference capacity (10 GW Nvidia, 6 GW AMD) and emphasize MI450‑class accelerators, while relying on fabs like TSMC. The implications: higher centralized demand for inference infrastructure, stronger incentives for alternative silicon (AMD/TPU/Apple/Intel), shifting integration and conversion burdens onto third‑party devs, and a tighter control point for data, monetization (Instant Checkout), and standards in AI application deployment.