Google Cloud grew 82% to Azure's 43%. Owning the model & the chip funds the aggression; renting both, & carrying half a backlog in one borrower, funds the hedge.
The harness, not the model, is now the faster-moving variable in AI cost & capability. It owns the prefix, & the prefix is 86-98% of the bill.
OpenAI's year-old GPT-OSS commands 36% of Opus 4.8's OpenRouter volume. The reason is segmentation: buyers split across cost, speed & accuracy, & mixture-of-experts just moved the local boundary.
Google Cloud grew 82% to $24.8b in Q2 2026, began recognizing TPU system sales, & posted a $514b backlog as its growth rate converges toward NVIDIA's.
Per-engineer productivity from AI coding runs from -19% to 8x. The distribution has three tiers, & the gap between them is the operating layer, not the model.
Open-weight models keep catching closed frontier systems on Chatbot Arena Elo, 2023-2026, but have not materially led.
Slate Auto's $24,950 bare-bones pickup & Thinking Machines' 975B Inkling model share one business model: ship a vanilla, open base & monetize the customization layer.
The system around the model, not the model itself, is where the next generation of AI advantage will be won. What Grok Build's codebase-upload incident reveals about the harness.
AI is like a colander. Customers pour in, & many pour out. Retention lands between a mobile game & a social network, while the price of intelligence keeps falling.
How AI reshaped the software stack, venture stage definitions, and the firms that win over three years.
