SambaNova CEO Rodrigo Liang said the AI infrastructure market has entered a phase defined by inference workloads, operating efficiency, and clear return on investment. The company closed its Series F round with participation from new strategic investors during the same period. Liang made the remarks in a live interview on Bloomberg Technology from the CAIS Summit in Beverly Hills.
Context
The prior wave of AI spending centered on building large training clusters and developing frontier models that require massive compute resources. That period rewarded rapid capacity expansion and access to the largest available systems. Enterprises now face pressure to demonstrate concrete value from those investments, which has shifted attention toward the costs and constraints of running models in production.
Liang noted that organizations including JPMorgan have chosen to deploy AI infrastructure behind their own firewalls rather than depending entirely on public cloud providers. He also stated that frontier models and private AI systems are not mutually exclusive. This combination allows companies to keep sensitive workloads on premises while still reaching the largest public systems when required.
Details
Liang identified efficiency improvements as the main factor driving current deployment decisions. The Series F round supplies additional capital and brings strategic partners into SambaNova at a moment when customers seek lower cost per inference query. Private deployments give enterprises direct control over data while still using high-performance hardware optimized for their specific models.
The CEO linked the funding close directly to customer demand for infrastructure that produces measurable returns. He cited JPMorgan as an example of an enterprise that moved AI workloads on premises to satisfy security and compliance needs. Liang described the ability to run optimized private models alongside access to frontier systems as a practical way to match model size and cost to individual use cases. The interview occurred live on Bloomberg Technology.
No financial terms or valuation figures were released in the appearance. The discussion stayed focused on the operational shift from training-scale builds to inference economics and data-residency requirements.
Why it matters
Technical teams evaluating AI hardware and software now face different purchase criteria than they did two years ago. Early projects accepted high operating costs in exchange for quick deployment and access to the largest models. The next set of decisions will turn on cost per token, data residency rules, and the ability to run workloads without routing every query through a third-party cloud.
SambaNova’s Series F close with strategic investors indicates that at least some capital providers see durable demand for systems built around these constraints. For engineering organizations already operating large clusters, the immediate task is to measure and improve inference efficiency while preserving required performance levels. Enterprises that treat private infrastructure as a credible alternative to frontier APIs gain leverage in vendor discussions and tighter control over data movement.
This change does not eliminate the need for the largest models. It adds a second track where cost, compliance, and ownership determine which workloads stay on premises and which continue to use public systems. The result is a market in which efficiency metrics move from secondary considerations to primary drivers of hardware and software selection.
---
Sources:
No comments yet