SambaNova unveils SN50 chip, nets $350M and Intel collaboration

SambaNova Systems on Thursday unveiled the SN50, an inference accelerator the company says is tuned for emerging "agentic" AI workloads, and announced more than $350 million in fresh funding along with a strategic collaboration with Intel. The package marks a push by the startup to claim a larger role in the market for hardware that runs AI systems which plan, act and chain multiple steps autonomously.

The company described the SN50 as optimized for workloads where AI models interact with tools, external data streams and decision-making loops rather than solely producing single-turn responses. Those agentic patterns amplify demands on latency, memory persistence and mixed-precision compute during inference, shifting some architecture priorities away from training throughput toward sustained, low-latency execution.

SambaNova has positioned the SN50 as a purpose-built accelerator for those trade-offs, aiming to give enterprises an alternative to general-purpose GPUs and to fill performance gaps at the inference layer. The new funding, which the company said totals more than $350 million, will underwrite productization, software development and deployment of the SN50 in customer environments. SambaNova also flagged a strategic collaboration with Intel to deliver combined solutions.

The move intensifies competition in a field long dominated by GPU suppliers. Inference hardware is increasingly the front line of enterprise AI deployment because production systems run far more inference cycles than training runs. For businesses building autonomous agents that coordinate applications, retrieve live data and execute multi-step workflows, specialized inference accelerators promise lower operational cost and improved responsiveness.

Technical details released by SambaNova were limited, and the company did not provide benchmark comparisons in its announcement. The SN50 announcement included claims of architectural tuning for agentic patterns, but independent testing will be necessary to quantify gains in throughput, latency and energy efficiency relative to available GPUs and other AI accelerators.

Beyond performance and cost, the SN50 and the trend toward agentic systems raise governance, safety and transparency questions. Systems that act across services and databases increase the surface area for errors, unintended actions and misuse. More capable inference hardware could accelerate adoption of autonomous agents in enterprise workflows such as customer service automation, finance and industrial control, amplifying both productivity gains and potential risks from malfunction or adversarial exploitation.

SambaNova's collaboration with Intel suggests integration and go-to-market coordination that could speed adoption by customers already invested in Intel server ecosystems. For buyers, the partnership may ease procurement and support hurdles, but it also reflects an industry-wide effort to package chips, software stacks and deployment services as bundled solutions for enterprise IT teams.

The SN50 launch and the accompanying financing show investors and suppliers are betting that inference, not just training, will be a defining battleground for AI infrastructure over the next several years. How quickly enterprises adopt agentic architectures will depend on measurable improvements in cost and reliability, and on governance frameworks that keep autonomous behavior aligned with legal and ethical constraints.