Together AI, which helps developers access Nvidia server chips to train AI models, raised $106M led by Salesforce Ventures at a $1.25B valuation
Context & Ripple Effects
Together AI had previously raised a $102.5M Series A for more efficient open-source model training and fine-tuning, establishing it as an infrastructure provider rather than a model developer. A February report had already flagged the planned $100M-plus Salesforce Ventures-led round, making this financing the confirmation of a capital step already underway.
The round matters because it assigns a billion-dollar valuation to a company whose product sits between developers and scarce Nvidia-based training capacity, a position where access and service execution can be as important as model software.
First-order effects
- Together AI gains $106M of additional financing and a $1.25B valuation benchmark, strengthening its ability to serve developers seeking Nvidia server-chip access for model training.
- Salesforce Ventures becomes the lead investor, tying a major enterprise-software investor more closely to Together AI's developer-facing AI infrastructure business.
Second-order effects
- Other AI-compute access providers face greater pressure to show that their capacity, developer tooling, or model support can differentiate them from a newly better-funded rival.
- For developers, the deal reinforces the emergence of specialized intermediaries for obtaining training compute, rather than requiring every customer to directly assemble infrastructure.
Third-order effects
- If similar financings continue, AI infrastructure may consolidate around well-capitalized platforms that package compute access with the software layer needed to train and operate models.
- The pattern points to compute access becoming a finance-intensive platform market: providers will need both hardware supply relationships and repeated capital raises to compete.
The trend: This is one data point in the platformization and commercialization of AI compute, as startups raise large rounds to turn scarce accelerator capacity into developer-facing services.