# Open model inference startups see near-exponential growth in token volume

> Fireworks alone serves over 40 trillion tokens daily to custom models, and its inference business more than doubled since this summer. Open model startups are now scaling at near-exponential rates that rival closed-source giants.

- Type: AI news
- Page: https://completeaitraining.com/news/open-model-inference-startups-see-near-exponential-growth/
- Published: 2026-11-05
- Updated: 2026-11-05
- Author: Jeroen Erne
- Jobs: Education, Marketing, PR and communications, Sales
- Reading time: 2 min

Open model inference startups are processing hundreds of trillions of tokens daily, with volumes charting near-exponential growth that challenges assumptions about closed-source dominance. The data, drawn from public disclosures by Baseten, Fireworks, and Together AI, provides the clearest signal yet of the open model economy's scale - and it is no longer a rounding error next to OpenAI and Anthropic.

#### The token volume race

Fireworks announced at its first company conference this week that it serves more than 40 trillion tokens per day to customers' customized models. That figure covers just one slice of its inference business, a vertical the company said grew over 2x from this summer. The total number this fall likely sits well above 100 trillion tokens.

These startups serve as bellwethers because their fates correlate directly with open model adoption. Hyperscalers like Amazon Bedrock and Azure's model garden almost certainly hold major footprints, but they do not break down usage by model category. OpenRouter's API, which reports open model share of its token volume, offers another data point.

The growth curve, plotted on a log scale, shows near-exponential expansion before the uptick expected from GLM 5.2's step change in open agents. Google's earnings reports, which release token volumes for its entire platform and API usage, provide a grounding comparison. When overlaid, the open model inference startups' trajectory cuts a steep line against the tech giant's baseline.

#### Why precision remains elusive

Every measurement of open model adoption carries noise. Platform metrics are imprecise. Startup disclosures mix marketing with substance. Public evaluations remain imperfect. "It takes an informed eye to figure out where things are going," the research note states. These approximations represent the best available synthesis of fragmented data.

The dashboard aggregating these metrics remains freely available. The accompanying research notes, which are more raw and exploratory, sit behind a paid tier. The goal is to build more value than the platform captures.

#### Why this matters for marketing, sales, and communications professionals

The token volume numbers carry a practical message about cost and vendor strategy. When open model inference grows at this rate, the per-token cost of running AI workloads drops fast. Marketing teams building AI-powered personalization, sales organizations deploying agents for outreach, and communications staff generating content at scale all face a shifting procurement landscape. The question is no longer whether open models can handle production workloads - it is whether your team's infrastructure bets account for a market where the fastest-growing inference option may not come from the company with the best-known brand name.

---

This is the Markdown form of https://completeaitraining.com/news/open-model-inference-startups-see-near-exponential-growth/. The whole site is mapped at https://completeaitraining.com/llms.txt.