What the AI stack actually costs in 2026

The honest totals

Price bands computed from the v2026.Q3 dossiers, 2026-08-06; every underlying tier was read from the vendor's own pricing page.

Across the 58 elements: 12 are free, 30 run under $30/month, 13 run $30–150, and only 3 sit above $150. A working floor is $60–90/month; the full solo compound $150–400; and the average American already pays for 4 premium AI tools at ~$66/month — usually the wrong four, bought in the wrong order.

The billing shift eating 2026

The structural story isn't price levels — it's price models. Usage-based and credit billing spread from coding agents to everything this year: GitHub Copilot's June move produced 10–50x bill spikes for agentic power users, Kiro and Replit run on credits, JetBrains meters AI separately. At the same time free tiers retreated — Google cut Gemini CLI's individual access, Alibaba ended Qwen's hosted free tier, Amp waitlisted its free plan. The practical consequence: a stack you priced in January costs something different in August, which is why every elems price line carries a verification date.

Reading the bands

Every element carries a band — Free, $ (under $30), $$ ($30–150), $$$ ($150+) — per seat at typical solo/small-team plans, and the dataset ships them as price_band_usd_per_month so nothing has to be inferred from a glyph. The $$$ elements (3 of 58) are almost all trust infrastructure: compliance automation and premium sales tooling you buy when a contract, not a mood, demands it.

The only budgeting rule that survives contact

Fill seats on triggers, not forecasts. 52% of software licenses go unused; the cancel-and-restart rate on AI subscriptions is 53%. The stack builder marks which elements your stage actually needs now and which to defer — and the audit exists for the day the bleeding is already underway.