Semiconductors & Hardware holds 1,210 providers publishing 5,488 APIs — chips, compute, consumer electronics, and the data-center hardware everything else is built on.
Almost nothing in its top band is a chip company. What it is, instead, is the layer that rents the hardware.
The functional bands
| Band | Providers |
|---|---|
| Cloud compute primitives | Google Cloud Platform (7 APIs, 81.1), Amazon EC2 (8, 72.0), EC2 Auto Scaling (65, 71.8), Amazon Lambda (2, 70.3), Microsoft Azure (1,667, 69.8) |
| Storage primitives | Amazon EBS (6, 69.4), Amazon EFS (1, 68.8) |
| Accelerated inference | NVIDIA NIM (11, 75.7), fal (12, 75.1) |
| Data platforms on hardware | Databricks (57, 72.2), IBM (55, 74.6) |
| Design and engineering software | Autodesk (69, 76.5), Bentley Systems (87, 69.8) |
| Enterprise operations | Workday (50, 70.2) |
Fourteen exemplars. Floor of 68.8.
Compute became an API and chips did not
The clean split in this table is between companies that sell you silicon and companies that sell you access to silicon. Only the second group has a described API surface.
You cannot call AMD. You can call Amazon EC2 (72.0) and receive a machine. You cannot call TSMC. You can call NVIDIA NIM (75.7) and receive an inference. The physical supply chain remains a procurement relationship measured in quarters; the consumption of what it produces became a self-serve API measured in seconds.
NVIDIA NIM at 75.7 is the most interesting entry for exactly this reason. It is the closest thing in the table to a chip vendor with a public API — a company whose hardware is the constraint on an entire industry, shipping a described, callable inference surface. That is a chipmaker responding to the fact that most of its customers will never touch one of its cards directly.
Azure’s 1,667 APIs
Microsoft Azure publishes 1,667 APIs and scores 69.8 — by far the largest surface anywhere in the catalog, and the lowest-scoring exemplar in this table.
Set beside Google Cloud Platform at 81.1 with 7 APIs, it is the sharpest illustration available of how little surface area contributes to score. A 238-to-1 difference in API count, and the smaller one scores eleven points higher.
That is not an argument that Azure should publish less. It is an argument that a surface of 1,667 definitions is very hard to keep uniformly well described, governed, and discoverable — and that the score is measuring exactly that difficulty rather than the engineering behind it.
What has shifted recently
- Inference joined the hardware vertical. NIM and fal (75.1) are here because GPU access is now a called service, and the API is the product.
- The primitives keep winning. Lambda scores 70.3 on 2 APIs, EFS 68.8 on 1. A single perfectly-described operation outscores most large platforms.
- Design software sits in hardware. Autodesk (76.5) and Bentley Systems (69.8) are in this cohort because the things they design are physical, and both publish substantial, well-scored surfaces — better than most of the companies that manufacture from their output.
Where to start
- apis.io/industries/semiconductors-hardware/ — the full 1,210
- apis.io/providers/nvidia-nim/ — 11 APIs, 75.7, the chipmaker with a callable surface
- apis.io/providers/google-cloud-platform/ — 7 APIs, 81.1, the top score in the vertical
Takeaway
1,210 providers, 5,488 APIs, and a leaderboard made of companies that rent hardware rather than make it. The manufacturers remain a procurement conversation; the consumption layer is a self-serve API — and Azure’s 1,667 definitions score below Google Cloud’s seven.
Browse it at apis.io/industries/semiconductors-hardware/.