What happened
OpenAI shut down its Sora video product, citing $15 million daily operational costs against $2.1 million in revenue, leading to a $1 billion Disney investment evaporating. OpenAI's non-binding DRAM letters influenced Micron's pivot; subsequent project cancellations and failed terms coincided with Micron's stock decline. Google released Gemma 4, an open-weight model achieving 85.2% MMLU Pro, runnable on phones, with significant downloads. Apple, avoiding direct frontier model costs and infrastructure spending, uses personal context, while Anthropic prioritises ecosystem lock-in.
Why it matters
High operational costs and infrastructure miscalculations introduce significant financial risks for companies prioritising frontier model development. This follows Apple's collection of $900 million in AI-related App Store commissions, demonstrating a different monetisation strategy. Platform engineers and procurement teams must evaluate the total cost of ownership, considering both model performance and the long-term financial viability of AI providers, as the market increasingly rewards integrated solutions over standalone model prowess.
