Nvidia currently dominates approximately 90% of the AI computing market but smaller companies are hoping to challenge that market hold in 2025.
Inference is the computation needed to produce the response to a user's query or request, whereas the computing required to teach a model the knowledge to answer is called 'training'.
Startups like Groq, SambaNova, and Positron are focusing on inference optimization in order to compete with Nvidia's dominance in the market.
New post-training strategies for AI in 2025 may benefit all providers of inference chips according to CEO of Nvidia, Jensen Huang.
Groq has resolved to execute, execute, execute to meet the demands of customer interest.
SambaNova Systems aims to shift the industry rapidly toward real-time applications and plans to ensure its technology scales efficiently and sustainably as inference workloads continue to grow.
Positron wants to deploy significantly more inference-specific compute in order to grow the adoption rate of 'chain of thoughts' and other inference-additional compute.
Groq's CEO, Mark Heaps, hopes this year people will recognize 'novel advancements against incumbent technologies'
SambaNova Systems CEO, Rodrigo Liang, would like to see the market move toward technologies that prioritize efficiency and sustainability to ensure AI can scale globally without overwhelming infrastructure.
Positron CEO, Thomas Sohmers, wishes to encourage the use of new AI tools to help normal, everyday people.