Gemini token math: serving 1B AI Mode queries costs Google only ~$4M
gaganghotra_ · x · 2026-11-13
Responding to conference claims that Google's AI Mode is expensive to run, the author runs the numbers: at 6 input words and 400 output words per query, serving 1 billion queries with Gemini costs only about $4M in net token costs.
He acknowledges other engineering costs exist, but argues model token cost is peanuts for Google even at multi-billion query scale — a small fraction of monthly revenue — so the high-cost narrative circulating among SEOs is overstated.
More from Infra
- Full Text Adventure Game Running on a Single Local Gemma4 26B Model — rosie254 · 2026-11-13
- RTX 5070 Ti Benchmarks: Tuned ComfyUI Flags Cut H3 Video Time by Over Half — AnalElder · 2026-11-13
- More Workloads Move From Public Cloud to Edge Data Centers for Cost Savings — DavidLinthicum · 2026-11-13
- Cloudflare launches Spend Management with hard spending limits for Workers — irvinebroque · 2026-11-13
- Vulkan vs CUDA Benchmarks for Qwen3.6-35B: Even on a 3060, 5-30% Slower on Datacenter GPUs — Serious_Scarcity7818 · 2026-11-13
- GTC attendee blasts Jensen Huang over circular financing and weak physical AI division — RockyCreamNHotSauce · 2026-11-13