Inference hosts

The Inference hosts is 84 on 2026-10-24 (n=84 host). Definition: The number of distinct hosts with at least one live inference SKU (a healthy, default-tier endpoint with a price) in the ledger.

Computed from the SKU state at the end of each UTC day (latest value: 2026-10-24; today's value is recomputed until the day ends). The constituents of the latest value are listed below; each shows when it was verified and links to its source.

Change over this window: +3.7% from 2026-10-19 to 2026-10-24 (level).

Dashed lines with a diamond mark days on which the set of constituents (n) changed; a jump on such a day can be coverage, not a price move.

Download this chart: SVG · PNG · CSV · JSON

Details of the latest value

Min coverage

1

Community rows excluded

177

Live skus

1,042

Basis

a host is counted once however many endpoints or sources list it; spelling and case are normalised, so two different companies with near-identical names would merge and one company under two brand names would count twice

{
 "first_party_hosts": [
  "BaseTen",
  "Cerebras",
  "Groq",
  "SambaNova",
  "Together"
 ],
 "hosts": {
  "AionLabs": 6,
  "AkashML": 7,
  "Alibaba": 32,
  "Amazon Bedrock": 38,
  "Ambient": 1,
  "Anthropic": 26,
  "Arcee AI": 1,
  "AtlasCloud": 23,
  "Azure": 56,
  "Baidu": 12,
  "BaseTen": 26,
  "Cerebras": 4,
  "Chutes": 5,
  "Claude Platform on AWS": 12,
  "Cloudflare": 19,
  "Cohere": 11,
  "CoreWeave": 22,
  "Crusoe": 9,
  "Darkbloom": 9,
  "Decart": 4,
  "DeepInfra": 61,
  "DeepSeek": 2,
  "DekaLLM": 10,
  "DigitalOcean": 19,
  "Featherless": 1,
  "Fireworks": 5,
  "Friendli": 6,
  "GMICloud": 26,
  "Google": 32,
  "Google AI Studio": 9,
  "Groq": 11,
  "Inception": 3,
  "Inceptron": 6,
  "InferenceNet": 8,
  "Io Net": 7,
  "Ionstream": 3,
  "Liquid": 1,
  "Makora": 7,
  "Mancer 2": 9,
  "Mara": 3,
  "Meta": 6,
  "Minimax": 6,
  "Mistral": 28,
  "Modal": 7,
  "ModelRun": 4,
  "Moonshot AI": 2,
  "Morph": 8,
  "Near AI": 3,
  "Nebius": 13,
  "Nex AGI": 1,
  "NextBit": 6,
  "Novita": 46,
  "Nvidia": 3,
  "OpenAI": 61,
  "OpenInference": 3,
  "Parasail": 40,
  "ParetoAlpha": 1,
  "Perceptron": 2,
  "Perplexity": 5,
  "Phala": 20,
  "Poolside": 4,
  "PrimeIntellect": 6,
  "Recursive": 1,
  "Reka": 10,
  "Relace": 13,
  "Sail Research": 9,
  "Sakana AI": 4,
  "SambaNova": 9,
  "ScaleDown": 2,
  "Seed": 6,
  "SiliconFlow": 40,
  "StepFun": 2,
  "StreamLake": 18,
  "Tencent": 5,
  "Tenstorrent": 2,
  "Thinking Machines": 2,
  "Together": 26,
  "Unbiased": 2,
  "Upstage": 3,
  "Venice": 34,
  "Wafer": 11,
  "Xiaomi": 3,
  "Z.AI": 12,
  "xAI": 1
 },
 "sources": [
  "baseten-pricing",
  "cerebras-models",
  "groq-models",
  "openrouter-endpoints",
  "sambanova-models",
  "together-pricing"
 ]
}

Definition

The number of distinct hosts with at least one live inference SKU (a healthy, default-tier endpoint with a price) in the ledger.

Inclusion rule

category inference; a SKU is live when its source is verified (not stale), the endpoint status is healthy, the service tier is the default and it has a price (free endpoints count as live); the host of an OpenRouter endpoint is its upstream provider, the host of a first-party host SKU (groq, together, sambanova, cerebras, baseten) is its upstream_provider or provider; community rows (the Hugging Face router, hf-router) are not hosts; host names are compared lower case without spaces or punctuation; each host counts once however many endpoints or sources list it, and carries its first-party name. A stale first-party source drops out; a stale OpenRouter source voids the series.

When it cannot be computed: Not computed: that day shows no number and states the reason; the last full value stays shown with its date. A source whose last confirmed snapshot is older than 3x its cadence (6h sources 18 h, 12h sources 36 h, 1d sources 72 h) drops out; below minimum coverage the series is not computed and says 'not verified: <source> stale since <date>'.

Minimum coverage: 1 host. Below that the series is not computed and says so instead of showing a number.

Sources feeding it

openrouter-endpoints, groq-models, together-pricing, sambanova-models, cerebras-models, baseten-pricing

Deliberately excluded

• unhealthy or non-default-tier endpoints
• endpoints without a price
• sources that are stale
• hf-inference-providers (community cross-check)

What can break it

• one company listed under two names counts twice, two companies with near-identical names merge
• hosts that list but do not serve
• first-party labs hosting their own models (OpenAI, Google, Anthropic) count as hosts
• a new inference source joining raises the count (a dated basis change)

Daily values, last 90 days

Days with a value since 2026-10-19 (days without one are left out of the data files): CSV · JSON.

API: JSON · CSV

InfraRabbit (2026). Inference hosts [data as of 2026-10-24]. https://infrarabbit.com/index/inference-host-count (accessed 2026-10-24). List prices, not transacted prices. CC BY 4.0.

@misc{infrarabbit_index_inference_host_count_2026,
  author       = {{InfraRabbit}},
  title        = {Inference hosts},
  year         = {2026},
  howpublished = {\url{https://infrarabbit.com/index/inference-host-count}},
  note         = {Data as of 2026-10-24; accessed 2026-10-24; list prices, not transacted prices; CC BY 4.0}
}