L5 · Training & Inference
Training & Inference — Companies & Commitments
Where compute gets operated: clouds, serving platforms, and the systems they run on. Companies are grouped by what they build, each with the latest verified signal from the engine. Methodology →
GPU Cloud & Neoclouds
6| Company | Commitment | Latest signal |
|---|---|---|
| Applied DigitalAPLD | AI/HPC data-center hosting | Applied Materials' fiscal 2024 operating margin reached 28.9%, up ~200bps YoY. |
| CoreWeaveNASDAQ:CRWV | GPU cloud (NVIDIA NCP) | Meta signed a cloud compute agreement with CoreWeave worth up to $14.2 billion through 2031 around September 30, 2025. |
| Nebius GroupNBIS | European GPU cloud | — |
| Crusoe | AI data centers & GPU cloud | — |
| Lambda | GPU cloud & clusters | — |
| Vultr | Global GPU cloud | Vultr offers NVIDIA HGX H100 instances starting at ~$2.30/GPU-hour on-demand. |
Inference & Serving
5| Company | Commitment | Latest signal |
|---|---|---|
| CloudflareNET | Workers AI edge inference | — |
| Anyscale | Ray-based compute platform | — |
| Fireworks AI | Fast inference serving | — |
| Replicate | Model hosting & inference API | — |
| Together AI | Inference & fine-tuning platform | — |
ML & Data Platforms
1| Company | Commitment | Latest signal |
|---|---|---|
| Databricks | Lakehouse + MLflow + Mosaic AI training/serving | Databricks announced it agreed to acquire serverless Postgres company Neon for approximately $1 billion on 2025-05-14. |
Servers & Systems
3| Company | Commitment | Latest signal |
|---|---|---|
| CelesticaCLS | Hyperscale hardware manufacturing | Celestica has reported capital expenditures of approximately $170.0M for an AI infrastructure project (disclosed 2026-02-17). |
| HPEHPE | AI servers & systems | HPE's Slingshot product line is currently at version 11 with 400Gbps line rate as of 2026. |
| Super Micro ComputerSMCI | GPU servers & rack systems | Super Micro anticipates total capex for fiscal year 2026 in the range of ~$155.0M (signal dated May 11, 2026). |
Hyperscaler Compute
1| Company | Commitment | Latest signal |
|---|---|---|
| Meta PlatformsMETA | Hyperscale AI compute & infrastructure | Arista Networks confirmed Meta as a top-4 cloud titan customer for AI Ethernet switching fabric (51.2T-class) on 2026-01-30. |
Training Data & Labeling
6| Company | Commitment | Latest signal |
|---|---|---|
| Gretel | Synthetic data | NVIDIA closed its acquisition of Gretel in March 2025 for a reported ~$320M |
| Labelbox | Data labeling platform | Labelbox raised a $110M Series E in January 2025 at flat $1B valuation post-Alignerr launch |
| MOSTLY AI | Synthetic data | MOSTLY AI's synthetic-data platform reportedly serves >70% of European tier-1 banks |
| Parallel Domain | Synthetic data for perception | — |
| Scale AI | Data labeling & RLHF | Tenstorrent launched its 'AI at Scale with Industry-Leading Performance' platform for scalable AI deployment in April 2026. |
| Surge AI | RLHF & data labeling | Surge AI reportedly hit ~$1B in 2024 revenue bootstrapped with no outside funding before its 2025 raise |
Other companies in this layer
5How this page is built
The roster is sourced from the engine's canonical company registry (primary_layer = L5, V15.1 taxonomy) and grouped by product sub-segment. The “latest signal” is the most relevant verified fact for each company from the layer's topics — sourced, English, and machine-verified. This is a landscape of who builds what, not a subjective ranking. For the full roster and an activity index, see /indices.
Subscribe to aiinframap Weekly
Track this layer's hot products — one email Friday. Engineers only.
Subscribe by layer →