SambaNova Systems CEO Rodrigo Liang said the next phase of artificial intelligence competition will center on inference costs and compute shortages.
This shift in focus signals a pivot in the AI industry's priorities. While early competition focused on the massive compute power required to train large models, Liang suggests that the ability to run those models efficiently for users — known as inference — will determine market winners.
Speaking in a Bloomberg Open Interest interview on Friday, Liang said this perspective is a contrast to the strategy of Cerebras Systems. The move comes as the industry evaluates how to scale AI infrastructure profitably [1].
"The next AI war won’t be about training models, but rather it will be about inference costs, compute shortages, and who can scale AI infrastructure profitably," Liang said [1].
Liang's critique follows the high-profile public market entry of Cerebras Systems. The company recently launched an IPO roadshow with a price target of $115-$125 per share [2]. By highlighting inference over training, Liang suggests that the financial viability of AI companies will depend more on operational efficiency than on the initial creation of models.
The competition between these two firms reflects a broader tension in the tech sector. Companies are racing to solve the compute shortage that has hampered the deployment of generative AI tools. Liang said the winner will not be the company with the largest model, but the one that can deliver those results at the lowest cost to the end user [1].
“"The next AI war won’t be about training models"”
The debate between SambaNova and Cerebras highlights a critical transition in the AI economy. As the industry moves from the 'build' phase to the 'deploy' phase, the financial burden shifts from one-time training costs to recurring inference expenses. If Liang's assessment is correct, the valuation of AI hardware companies will increasingly depend on their energy efficiency and cost-per-query rather than raw processing power.
