> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gdeltcloud.com/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Compute (Epoch AI)
> AI models, ML hardware, AI data centers, chip sales, and AI-company metrics from Epoch AI's published datasets, resolved into the GDELT Cloud entity spine.
The AI Compute API surfaces **Epoch AI's published research datasets** — notable AI
models, ML hardware, AI data centers, chip-sales estimates, and AI-company financials —
loaded offline into GDELT Cloud and resolved into our unified entity registry. Because
every organization-bearing row carries a resolved `entity_id`, an Epoch record joins the
live Events / Stories / Entities graph: an entity like NVIDIA or Anthropic surfaces the
models it developed, the hardware it makes, the data centers it owns, and its cumulative
chip sales.
The AI Compute endpoints are **generally available and plan-gated** — they require the
`can_use_epoch_ai` entitlement, carried by the Corporate & Supply Chain and Global
Intelligence plans. A key without it receives a clean `403 PLAN_REQUIRED` naming the
missing entitlement, never a silently empty result. A dedicated
**[/ai](https://gdeltcloud.com/ai)** explorer browses the same data in the web app.
Epoch AI data is published under **[CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/)**
and sourced from [epoch.ai/data](https://epoch.ai/data). Attribute Epoch AI when you
redistribute these values. It is a data source, not a subprocessor.
## Endpoints
All five endpoints return a JSON `data` array (a `403 PLAN_REQUIRED` for keys without `can_use_epoch_ai`).
Every endpoint accepts `limit`; the data-centers, models, and hardware endpoints also accept
`offset` (alias `cursor`) for paging. Individual filters are noted per endpoint below.
### AI Data Centers
`GET /api/v2/epoch/data-centers` — tracked AI data centers with owner, power (MW), compute
(H100e), cost, users, and current chip types, plus a country rollup.
| Param | Description |
| ------------------------- | ---------------------------------------------------------- |
| `limit` | Max rows to return (default 25, max 200). |
| `offset` (alias `cursor`) | Row offset for paging. |
| `country` | Case-insensitive substring match on data-center country. |
| `owner` (alias `org`) | Case-insensitive substring match on the data-center owner. |
```bash theme={null}
curl -H "Authorization: Bearer $GDELT_CLOUD_API_KEY" \
"https://gdeltcloud.com/api/v2/epoch/data-centers?country=United%20States&owner=Microsoft&limit=25"
```
### AI Models
`GET /api/v2/epoch/models` — the notable-models catalog (developer organization, domain,
task, training compute, training hardware, accessibility, open-weights, frontier flag).
| Param | Description |
| ----------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `limit` | Max rows to return (default 25, max 200). |
| `offset` (alias `cursor`) | Row offset for paging. |
| `organization` (alias `org`) | Case-insensitive substring match on the developing organization. |
| `organization_categorization` | Case-insensitive substring match on the developer's organization type. Accepted tokens: `Industry`, `Academia`, `Government`, `Research collective`. A model may carry several comma-composited types, so a token matches any model with that contributor type (e.g. `Academia` = any model with an academic contributor, not academia-exclusive). |
| `domain` | Case-insensitive substring match on model domain (e.g. `Language`, `Vision`). |
| `frontier` | Set `1` or `true` to return only frontier models. |
| `min_compute` | Minimum training compute in FLOP (returns models with `training_compute_flop` at or above this value). |
| `sort` | Result ordering. Default (empty) sorts by latest publication date; `compute` sorts by highest training compute. |
```bash theme={null}
curl -H "Authorization: Bearer $GDELT_CLOUD_API_KEY" \
"https://gdeltcloud.com/api/v2/epoch/models?organization=Anthropic&frontier=1&sort=compute"
```
### AI Hardware
`GET /api/v2/epoch/hardware` — the ML-hardware catalog (accelerators/GPUs) with
manufacturer, type, release date, and price.
| Param | Description |
| ---------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `limit` | Max rows to return (default 25, max 200). |
| `offset` (alias `cursor`) | Row offset for paging. |
| `manufacturer` (alias `org`) | Case-insensitive substring match on the hardware manufacturer. |
| `type` | Case-insensitive **exact** match on the accelerator class (validated enum — an unknown value returns `400 INVALID_ENUM`). Accepted: `GPU`, `TPU`, `GPGPU`, `XPU`, `XPU-R`, `NPU`, `ASIC`, `LPU`, `Hybrid CPU`, `Other`, and the composite `DCU (GPGPU)`. |
| `foundry` | Case-insensitive substring match on the fabricator. Observed values: `TSMC`, `SMIC`, `Samsung`, `Intel` (plus dual-fab `SMIC,TSMC` parts, which `SMIC` or `TSMC` both return). |
```bash theme={null}
curl -H "Authorization: Bearer $GDELT_CLOUD_API_KEY" \
"https://gdeltcloud.com/api/v2/epoch/hardware?manufacturer=NVIDIA&type=GPU"
```
### AI Companies
`GET /api/v2/epoch/companies` — the tracked AI labs and their latest financials
(company type, product domains, staff count, annualized revenue, valuation, total equity
funding, founding date), ordered by valuation.
| Param | Description |
| ------- | -------------------------------- |
| `limit` | Max rows to return (default 50). |
```bash theme={null}
curl -H "Authorization: Bearer $GDELT_CLOUD_API_KEY" \
"https://gdeltcloud.com/api/v2/epoch/companies?limit=50"
```
### AI Chip Sales
`GET /api/v2/epoch/chip-sales` — cumulative chip-sales estimates by designer
(latest window per manufacturer: compute in H100e, unit count, and power in MW).
| Param | Description |
| ------------------- | ----------------------------------------------------------------------------------------------------------------------------------- |
| `limit` | Max rows to return (default 50). |
| `chip_manufacturer` | Case-insensitive **exact** match on the chip designer. Accepted values: `Nvidia`, `Google`, `Amazon`, `Huawei`, `AMD`, `Cambricon`. |
```bash theme={null}
curl -H "Authorization: Bearer $GDELT_CLOUD_API_KEY" \
"https://gdeltcloud.com/api/v2/epoch/chip-sales?chip_manufacturer=Huawei"
```
This preview reads the `epoch` ClickHouse schema, which is loaded offline from Epoch AI's
published CSVs (`scripts/load_epoch.py`), separately from the GDELT ingest pipeline. The
tables are small and fully enumerable; owners and manufacturers are resolved into the same
entity spine used across the rest of the API, so you can pivot from a chip designer or model
developer straight to its GDELT Cloud entity page.