Gartner's Aug 17 prediction that AI inference costs will jump 5x+ by 2028 even as token prices keep falling
Token prices dropped over 90% in two years. Gartner just predicted your AI bill will still 5x by 2028.
Most builders price their AI features assuming falling token costs keep margins safe.
That assumption breaks the moment you add agents.
→ Cheaper tokens don't lower your bill — they raise the ceiling on what feels "affordable" to build
→ A basic chatbot call is cheap. Route it through an agentic/reasoning flow and cost jumps 5x+, before task complexity even grows
→ Every retry, tool call, and reasoning step in your pipeline is a multiplier, not a rounding error
→ "It got cheaper so I added more AI" is how a side project turns into a five-figure API bill
Cheaper tokens don't buy you a cheaper product. They buy you permission to spend more per task.
#AIEngineering #LLMOps #BuildInPublic #DataScience #StartupTips
The interesting part is that falling inference costs can change the product economics in both directions.
Once AI becomes cheap enough to use more heavily, the bigger pricing question may become what customers start expecting to be included rather than just what each call costs.
That's correct !
Exactly. It’ll be interesting to see how those economics affect what users eventually expect to be included.