AI infrastructure

5 of 43 services below report degraded performance or worse. Grouped by the role each plays in a production AI app; every entry is a service Statuswall already reads.

Statuswall last refreshed just now; individual sources may be older. For the full AI & ML category see all AI services.

What a status page does not tell you

  • A vendor status page reports availability and error rates it chooses to publish. It does not measure model quality, hallucination rates or output regressions.
  • Rate limits, token or spend quotas and per-tenant throttling are account-specific and usually do not appear on a public status page.
  • A model can be deprecated, renamed or have its behaviour changed without any incident being posted.

Model APIs

First-party model providers

Inference and model hosting

Serve open or custom models

Model gateways

Route and proxy across providers

GPU cloud

Raw accelerator capacity

Vector databases

Retrieval and embeddings storage

Application databases

Common backing stores for AI apps

LLM observability and evaluation

Tracing, evals, prompt management

Speech and voice

Speech-to-text, text-to-speech, voice agents

Cloud platforms

Where most AI workloads run

Checks on your side

Verification steps when an AI provider looks unhealthy, or when your users report problems the status page does not show.

  • Compare your own error rate and latency per model and per provider (4xx vs 5xx, timeouts).
  • Check for 429 responses and quota headers on your account, separately from the vendor's status.
  • Run a small fixed evaluation set to catch output-quality changes a status page will not report.
  • If you use a gateway or fallback, confirm which provider actually served recent requests.