Glossary
AIaaS
Meaning and details
AIaaS is AI as a utility. Speech-to-text, image generation, embeddings, chat completion: the model, its serving infrastructure and its scaling are the provider's problem, and your product integrates the capability throug...
AIaaS meaning
AI as a Service: AI models and capabilities delivered as metered APIs, no training or infrastructure required by the buyer.
What is AIaaS?
AIaaS is AI as a utility. Speech-to-text, image generation, embeddings, chat completion: the model, its serving infrastructure and its scaling are the provider's problem, and your product integrates the capability through an API key with per-token or per-call pricing. OpenAI, Anthropic, Google, Groq, DeepInfra and the long tail of specialized providers, voice, transcription, vision, constitute the tier, and the fall in per-token prices has been the single largest enabler of the indie AI product wave.
Why it matters to an indie builder: AIaaS means the intelligence layer of your product is a line item, not a research team. The strategic skill is cost and quality engineering: routing each task to the cheapest model that handles it, caching what repeats, batching what waits, and measuring the margin per feature, because in AI products the model bill is the cost of goods sold and it scales exactly with success. The llm-ops tooling exists to make that observable.
A concrete example: a meeting-notes product. Transcription via a speech API at $0.006 per minute, summary via a mid-tier model at roughly $0.002 per meeting, both behind a $12 monthly plan with a 20-meeting fair-use cap. Per-active-user cost lands near $0.35, gross margin 97 percent, and the numbers only close because each component is metered AIaaS, if either were self-hosted, the fixed cost would eat a solo operation alive at this scale. When volume grows tenfold, the same arithmetic points at batch discounts and smaller custom models.
Common mistakes: frontier-modeling everything, the default habit that turns a 90-percent margin into a 20-percent one, ignoring latency as a product feature (a voice agent that answers in 800 milliseconds feels magical; three seconds feels broken; provider choice moves this more than model quality), no fallback routing when a provider throttles or degrades, single-vendor lock-in with no abstraction seam, and shipping AI features whose unit cost nobody in the company can state. The aiaas blocks on IndieBoot, 46 of them, exist for side-by-side comparison on exactly these axes.