Who:
- Databricks: The $43B data lakehouse leader whose Unity Gateway routes AI traffic for 10,000+ enterprises
What Happened:
- Open-weight AI models jumped from 7.3% to 30.9% of enterprise token traffic between July-November 2023
- Chinese models (Zhipu, DeepSeek, Moonshot) dominate open-weight usage on US infrastructure despite security concerns
- 72% of enterprises now use multiple model providers, up from 36% in July
Why It Matters:
- Cost pressure is fracturing the 'one model to rule them all' myth: enterprises route bulk workloads (classification, extraction) to open-weight
- Vendor lock-in is dead: 'Built on GPT/Claude' claims now hurt more than help with enterprise buyers
- Gross margins will separate winners: products still routing all calls to frontier models face 10-100x COGS disadvantage
ARM Impact:
- **AI Sprinkler (Stage 3 (AI Sprinkler))**: Teams now benchmark open-weight vs proprietary models for each workflow
- **ARM (Stage 4 (Autonomous Revenue Master))**: Autonomous routing engines will emerge to optimize cost/performance across 7+ models per customer
- **Tab Hopper (Stage 1 (Tab Hopper))**: Beware vendors still pitching single-model integration as a feature
What to Watch:
- How OpenAI/Anthropic respond: expect 'good-better-best' pricing tiers by Q2 2024
- Security backlash: will US regulators restrict Chinese model weights running on cloud infra?
- The Kimi effect: Moonshot's chatbot now drives 40% of open-weight tokens after one release