Originally Published Research by Richard EwingCanonical Article on Beehiiv ↗
SaaS Unit Economics

Generative AI Margin Squeeze & SaaS COGS

Richard Ewing··6 min read
⚡ 5-Second Executive Takeaway

Traditional SaaS software enjoyed 80%+ gross margins because server costs were low. Generative AI shifts hosting into variable COGS, dropping margins to 50% unless you offload routine tasks to local SLMs. Exogram restores 80%+ margins automatically.

The Generative AI margin squeeze occurs as API token costs transition software hosting from fixed infrastructure expenses into variable cost of goods sold (COGS). Un-governed LLM API calls compress SaaS gross margins from 85% to under 50%. Exogram provides real-time model routing and semantic caching to restore enterprise software profitability.

Why Did SaaS Gross Margins Drop from 85% to 50%?

Historically, software companies paid fixed AWS or Azure bills regardless of whether a user clicked 10 times or 10,000 times. With Generative AI features, every click triggers LLM token generation, turning infrastructure into a direct variable cost.

How Exogram Re-Expands Gross Margins

Exogram optimizes COGS by intercepting incoming prompt requests. High-complexity reasoning is routed to frontier models, while basic tasks are offloaded to local small language models (SLMs) or zero-cost code functions.

Margin Restoration Playbook:

  • Variable COGS Control: Cap token spend per user session automatically.
  • SLM Offloading: Reduce API costs by up to 80% on routine text operations.
  • Semantic Caching: Serve identical queries instantly without paying model providers.