Case study · GitHub signal → priced round

Together AI, open training stack to a $106M Salesforce Series A

Together AI's RedPajama dataset and training-stack openness led the $106M Salesforce Series A in March 2024.

At a glance

Company: Together AI
Sector: AI / training infrastructure
Primary repo: github.com/togethercomputer/RedPajama-Data
Trigger window: first half 2023, accelerating into 2024
Stars at trigger: ~4K stars at trigger window
Announced raise: $106M Series A (Salesforce) (2024-03-13)
Lead investor: Salesforce Ventures
Time-to-money read: OSS dataset release + foundation model partnerships led the Series A

Together AI's strategy is to be the open-stack counterweight to closed model labs. The RedPajama dataset release in 2023 was a category-defining moment, the largest curated open training set at the time.

Partnership announcements through 2023 (foundation-model deals, hardware partnerships) were the leading signal. Each one weak alone; together strong.

Salesforce led the $106M Series A in March 2024. The partnership signal had been compounding for at least four quarters.

Signals that would have flagged this pre-raise

OSS dataset release:RedPajama dataset in 2023
Partnership density:Multiple foundation-model + hardware partnerships
Conference signal:Active presence at ML conferences

How the timeline read

The engineering acceleration was observable during first half 2023, accelerating into 2024, while any fundraising paperwork was still private. The announced event, $106M Series A (Salesforce) on 2024-03-13, landed after the signal window: OSS dataset release + foundation model partnerships led the Series A. That ordering is the entire thesis of this series: by the time a round appears in a funding digest, the repositories were already telling the story at ~4k stars at trigger window.

In practice, that is what a weekly monitoring cadence buys you. VC Deal Flow Signal re-scores this sector (AI / training infrastructure) every week across its tracked repository, so a window like this one surfaces as a rising trend rather than a single spike you had to be lucky to catch. The pre-raise signals listed above are the rows that moved while the press stayed quiet.

Repositories

github.com/togethercomputer/RedPajama-Data

Frequently asked questions

Is dataset release a generalizable signal?

For training-infra companies, yes. Releasing a credible open dataset is a strong public commitment that draws partnership interest.

Find the next one

VC Deal Flow Signal tracks engineering acceleration weekly across twenty sectors, the same signal shapes that preceded the raise above.

Get the weekly signal report →

Related case studies

Read them side by side: these related cases span AI / foundation model, AI / model hub, each with its own trigger window and raise event, and the same pre-raise signal shape held in every one. A pattern that repeats across different sectors and different windows is what separates a repeatable signal from an after-the-fact story.