How much will my model cost? Napkin Maths 101
Hosted by Abi Aryan
What you'll learn
Quickly estimate model costs
Most AI engineers in 2026 still have ZERO clue how much VRAM their model actually needs.
Concurrency handling is never taught
Your RAG demo works on your MacBook. You think you’re ready for production. Then your 8×H100 box OOMs at 3am with 8users
Model math doesn't translate to Agentic Systems
5 numbers to look at when evaluating costs for real agentic workflows. Single-turn v/s multi-turn runs. How they differ?
Why this topic matters
Most AI engineers jump straight into coding without thinking about cost. Most cost calculators only teach token maths but not RAM or compute costs or network costs. In this session, we will look into all that.
Because when your LLM scales from one call to thousands per day, those numbers add up fast. In this lightening talk, I break down model costs using simple “napkin math” anyone can follow.
You'll learn from
Abi Aryan
Lead Research Engineer @ Abide
Abi Aryan is the founder and lead research engineer at Abide AI, a deep tech company developing neurosymbolic models for reasoning in agents. With a decade of experience as an ML engineer building production-scale AI systems, she is also the author of two books:
- LLMOps (O'Reilly Publications)
- GPU Engineering for AI Systems (upcoming title from Packt Publishing, releasing Autumn 2026)
Go deeper with a course
AI Systems Design & Inference Engineering

Abi Aryan
Founder and Research Engineering Lead @ Abide AI
Keep exploring

.jpg&w=1536&q=75)


