Inside the AI Arms Race: LLMs, Economics, and Strategy
A strategic view of vendor pricing games and how buyers should respond.
Open cxotalk.com →
Building & Product
Podcast
Podcast
Article
Thread
Guide
Guide
Article
Article
Article
Article
Article
Article
Article
Article
Article
Article
Article
Tool
Tool
How much does it cost to run AI features, and how do I keep costs sane?
You pay per token (roughly per word) in and out, so cost scales with usage; a small feature can run on tens of dollars a month while a chat-heavy product can quietly eat 30-60% of revenue. The good news is prices for the same capability keep falling about 10x a year, and three habits (prompt caching, routing easy requests to cheap models, and batching non-urgent work) routinely cut bills 50-80%. Instrument cost per user per feature from day one so pricing decisions are based on data, not fear.
Go deeper
19 resources.
Simon Willison: Things we learned about LLMs in 2024 (Latent Space)
Willison narrates the price collapse and capability gains in one listenable year-in-review.
Open latent.space →Welcome to LLMflation: LLM inference cost is going down fast
The definitive chart: equivalent-quality inference gets 10x cheaper per year, which changes what you can afford to build.
Open a16z.com →a16z on LLMflation (thread)
The five-minute version of the cost-decline argument, useful for convincing a worried co-founder.
Open x.com →Prompt caching (Claude docs)
Cached reads cost 10% of normal input price; the single highest-leverage cost lever, from the primary source.
Open platform.claude.com →Batch API (OpenAI docs)
A flat 50% discount for anything that can wait up to 24 hours, like digests, enrichment, and backfills.
Open developers.openai.com →The State of AI Gross Margins in 2025
Grounded numbers on what AI COGS actually look like across real startups, not vibes.
Open tanayj.com →Exclusive Data: Startups Are Learning What AI Profits Look Like
Fresh survey data showing margins improving from 41% to 45% and what the improvers did differently.
Open upstartsmedia.com →LLM Cost Optimization: 5 Levers to Cut API Spend 70-85%
Routing, compaction, prompt trims, caching, batching: each lever quantified so you know what to do first.
Open morphllm.com →LLM Token Optimization: Cut Costs and Latency
Engineering-level detail on token budgets and semantic caching from an infra company that measures it.
Open redis.io →8 LLM Cost Optimization Techniques, Visually Explained
Diagram-first explanations your whole team can absorb in ten minutes.
Open pub.towardsai.net →Top 5 LLM Gateways in 2025
A gateway gives you routing, fallbacks, and spend visibility in one move; this compares the options.
Open helicone.ai →LLM API Pricing Comparison (2025): OpenAI, Gemini, Claude
A sober side-by-side of per-token prices across the big three, including budget tiers.
Open intuitionlabs.ai →Prompt Caching is a Must: From $720 to $72 Monthly
A builder's receipts: one technique, a 90% real bill reduction, with the code that did it.
Open medium.com →The Unit Economics of AI Agents: Budgeting LLM Costs with Real Numbers
Actual production numbers for agent workloads, the costliest and least predictable AI feature type.
Open medium.com →Inference Unit Economics: The True Cost Per Million Tokens
Explains why output tokens cost more than input and how billing categories really work.
Open introl.com →X-raying OpenAI's unit economics
Understand your supplier's economics to predict where API prices go next.
Open exponentialview.co →Helicone cost tracking cookbook
Practical recipes for attributing AI spend per user and per feature, which is how you find the leaks.
Open docs.helicone.ai →LLM API Providers Leaderboard
Live price, speed, and latency comparison across 500+ endpoints; check before you commit spend.
Open artificialanalysis.ai →More in AI in Your Product
- Should my startup add AI features, and which ones actually matter to users? 20 →
- How do founders ship their first AI feature (APIs, wrappers, RAG in plain terms)? 19 →
- Which AI model should I build on (GPT, Claude, Gemini, open-source), and how do I avoid locking myself in? 17 →
- How do I test whether my AI feature actually works before launch (evals, in plain terms)? 18 →
- How do I stop my AI feature from making things up in front of customers? 17 →
- Prompt engineering, RAG, or fine-tuning: which one does my product actually need? 18 →
- How do I design AI features people trust (chat vs buttons, streaming, showing the work)? 18 →
- What is my moat if I'm building on someone else's model? 19 →
- Should I put an AI agent in my product, or is a simple workflow enough? 18 →
- Is it safe to send customer data to AI APIs, and what about privacy laws like India's DPDP Act? 18 →
- How should I price AI features: per seat, per use, or per outcome? 17 →
- My AI feature is live. How do I monitor it and make it better with real usage? 18 →
Also in Starting Up
The same ground, over in Build the product, our Starting Up track.
eChai Partner Brands
