2025 valuation: .3B. July 2026 valuation: .3B. Together AI multiplied 2.5x in 18 months.
This isn't an AI bubble - it's open model infrastructure becoming the new AI industry foundation.
Three Structural Drivers
Open model inference costs 10x less - running Llama 4/Mistral/Qwen is an order of magnitude cheaper than GPT-4o/Claude. Enterprise customers aren't "buying cheap" - they're doing 10x more with the same budget.
AI infrastructure ? cloud infrastructure - Traditional cloud is general compute. AI infrastructure needs GPU cluster scheduling, fast model weight loading, batch inference, multi-model hybrid deployment. Together AI optimized for these specifically.
Data sovereignty - More enterprises require models running in their own environments. Open models naturally satisfy this - closed-source can't.
Market Landscape
| Layer | Companies | Competition |
|---|---|---|
| Closed-source model | OpenAI, Anthropic | Capability + brand |
| Open model | Meta, Mistral | Performance + open source |
| Inference infra | Together AI, Fireworks | Cost + speed |
| General cloud | AWS, GCP, Azure | Scale + ecosystem |
Together AI's barrier isn't model capability - it's inference efficiency and scheduling optimization.
Risks
- GPU supply bottleneck
- Open models reaching closed-source parity - cost advantage evaporates
- Cloud giants entering inference services
Developer Impact
- API costs continue declining - Together forces closed-source to cut prices
- Open models aren't "second-best" anymore - most scenarios suffice
- Hybrid deployment becomes standard - critical: closed, daily: open, unified scheduling
Bilingual version at wdsega.github.io
Top comments (0)