AI inference companies: 15 verified, updated November 13, 2026
Presspool has verified 15 AI inference companies. 10 of them announced a round in the last six months, most recently Subconscious ($5.1 million, September 22, 2026). Latest rounds: the 14 stated exactly in dollars total $5.24 billion; 1 more in dollars has no exact figure and is left out of totals and rankings.
A subset of Presspool's AI infrastructure map: companies whose fit on that map places them in AI inference. Companies selling the layers other AI products run on: GPU clouds and neoclouds, inference and model-serving platforms, training data and web data for AI, evaluation, observability and testing for AI, vector and AI-native databases, and infrastructure for AI agents. Each company is already on the parent map, where it was taken from the company's own website and confirmed by an independent source, both linked on the entry. No company pays to be listed.
This map does not have its own weekly question set or answer sample. The questions and answer counts for these companies are on the AI infrastructure map.
| Company | What it does | Latest announced financing |
|---|---|---|
| Fireworks AI | Fireworks is a platform where companies train, customize and serve open models, with an inference engine tuned for throughput and latency. | $1.505 billion, Series D (July 16, 2026) |
| Baseten | Baseten is an AI inference company that post-trains and serves customers' models at production scale, with a focus on latency, reliability and cost. | $1.5 billion, Series F (June 22, 2026) |
| Together AI | Together AI runs an AI-native cloud: a full-stack platform for inference, fine-tuning and GPU clusters built around open models. | $800 million, Series C (July 1, 2026) |
| Modal | Modal is a cloud built for AI, giving developers primitives for workloads such as low-latency elastic inference. | $355 million, Series C (May 21, 2026) |
| Groq | Groq runs an AI inference cloud, a neocloud serving open models on its own data-center capacity, after licensing its chip technology to Nvidia. | $350 million, Series A (August 17, 2026) |
| Inferact | Inferact was founded by the creators and core maintainers of vLLM to grow it as an AI inference engine and make inference cheaper and faster. | $150 million, Seed (January 22, 2026) |
| fal | fal is a generative media platform that runs 1,000+ image, video, audio and 3D models through APIs, serverless deployment and GPU compute. | $140 million, Series D (December 9, 2025) |
| OpenRouter | OpenRouter provides one OpenAI-compatible API that routes requests across hundreds of AI models from many providers, with fallbacks, cost tracking and enterprise controls. | $113 million, Series B (May 26, 2026) |
| DeepInfra | DeepInfra runs an inference cloud that serves 150+ open-source models through OpenAI-compatible APIs on GPU infrastructure it owns and operates. | $107 million, Series B (May 4, 2026) |
| RadixArk | RadixArk, founded by the creators of the SGLang inference engine, builds open AI infrastructure for inference and reinforcement learning, including managed infrastructure for teams. | $100 million, Seed (May 5, 2026) |
| Ollama | Ollama is an API for open models that runs inference locally or in the cloud. | $65 million, Series B (July 9, 2026) |
| Parasail | Parasail is an inference cloud that serves 40+ open and frontier models through one OpenAI-compatible API and tunes deployments for speed, quality and cost. | $32 million, Series A (April 15, 2026) |
| Tensormesh | Tensormesh sells a KV-cache data management platform for AI inference that reuses computed context to cut GPU recompute, latency and cost. | $20 million, Seed extension (May 27, 2026) |
| Subconscious | Subconscious is an inference provider for long-running AI agents, built for 200K+ token sessions. | $5.1 million, Pre-seed and seed (September 22, 2026) |
| SiliconFlow | SiliconFlow runs an AI inference cloud that serves and fine-tunes more than 200 language and multimodal models through APIs. | more than 2 billion yuan ($294 million), Series B (June 17, 2026) |
AI inference startups that raised funding recently
Rounds announced in the six months to November 12, 2026, newest first. Each links the company's record and its sources.
- Subconscious: $5.1 million, Pre-seed and seed, announced September 22, 2026, led by MassVentures.
- Groq: $350 million, Series A, announced August 17, 2026, led by Disruptive.
- Fireworks AI: $1.505 billion, Series D, announced July 16, 2026, led by Atreides Management, Index Ventures and TCV.
- Ollama: $65 million, Series B, announced July 9, 2026, led by Theory Ventures.
- Together AI: $800 million, Series C, announced July 1, 2026, led by Aramco Ventures.
- Baseten: $1.5 billion, Series F, announced June 22, 2026, led by Altimeter Capital, Conviction, Spark Capital, Sands Capital and Wellington Management.
- SiliconFlow: more than 2 billion yuan ($294 million), Series B, announced June 17, 2026.
- Tensormesh: $20 million, Seed extension, announced May 27, 2026.
- OpenRouter: $113 million, Series B, announced May 26, 2026, led by CapitalG.
- Modal: $355 million, Series C, announced May 21, 2026, led by General Catalyst and Redpoint.
What counts as AI inference
This page lists companies already on the AI infrastructure map. A company is listed here when the fit recorded on that map names AI inference before any other layer. It is not a separate check. Out of scope: companies on the AI infrastructure map whose fit names a different layer first, and anything the AI infrastructure map excludes. Companies whose product is an end-user AI application, and companies whose most recent announced round is more than 12 months old. Definitions are versioned; this is version 1. Corrections need a public source: see Contact.
Changes
- November 13, 2026: Opened as a subset of the AI infrastructure map. Members are companies already on that map whose fit names this layer first. No separate company check was made for this page.