Fireside Chat, Interview, Conference Presentation
Vipul Ved Prakash, Together AI ｜ RAISE Summit 2025
- Together AI anticipates concluding its nearly three-year operational history by its third anniversary while positioning itself as a leader in maximizing GPU performance through hardware and software co-optimization.
- The industry outlook predicts a paradigm shift from SaaS to "Agentech," driving increased demand for reasoning capabilities and computational horsepower as generative AI becomes the world's digital mechanism.
- Market requirements for computing power are expected to surge as technology matures, characterized by rapid hardware progression from gaming GPUs to massive supercomputers with significantly accelerated interconnectivity between servers and chips.
- Future inference capabilities are projected to utilize reasoning and test-time compute to reduce hallucinations, alongside efficient model architectures that activate only a subset of parameters to enable more real-time inference passes.
- Hardware designs are expected to operate at very high thermal loads, necessitating robust default tolerance layers, while system configurations will dynamically prioritize inference workloads during peak loads and switch to training and reinforcement learning during idle periods.
- A platform featuring over 200 open-source models aims to deliver control over model behavior and optimized price performance, addressing enterprise needs for orchestration, observability, and granular data control without open-ended financial commitments.
- Technical advancements are predicted to enable the transfer of the entire internet's data in seconds, while industry action continues to be driven by fine-tuning, custom models, and open-source initiatives.