The problem
Frontier AI models are costly and generic; companies want cheaper models trained on their own data that they own
What they do
Cloud that trains, serves and keeps improving custom open-weight AI models for enterprises
Traction Run rate ~$12.8M (Nov 2025, The Information) to $100M (Nov 2026, company); team ~4x
Hiring trend
- Open roles
- 15
- Ops roles
- 3
- Headcount
- ~70
Team ~17 (Jan 2026) to ~70 (Nov 2026); 6 of 15 open roles posted since Oct 2026
Hiring in
- San Francisco
Funding
- Last round
- $350M
- Stage
- Series C
- Date
- Nov 2026
Led by Elad Gil
- Hiring+1.515 open roles on the live job board (15–19 roles = +1.5)
- Funding recency+2Series C · $350M, Nov 2026 (this month): 6 months or less = +1.5, $100M+ round +0.5
- Traction+1.5Strong growth: Run rate ~$12.8M (Nov 2025, The Information) to $100M (Nov 2026, company); team ~4x
- Red flags0None found in the last 24 months
- Total5.01.5 + 2 + 1.5 − 0 = 5.0 → Sprint (4.75 or more)
Distance (10K) is a separate read on how long the lead can last. It doesn't change the score.
Distance
10K · edge: hard to leave
Fast growth with big-name customers and 40% using both training and inference, but it sells into a crowded open-model cloud field (Together, Fireworks, Baseten) and frontier labs offer their own fine-tuning; customer concentration is undisclosed.
Sources: Applied Compute: $100M run rate, $350M raise, 40% of customers use training and inference (Nov 2026) ↗ · Forbes: trains custom open-source models and runs them on its own cloud; customers DoorDash, Nvidia, Harvey (Sep 2026) ↗
Last checked Nov 4, 2026