Descriptions:
SambaNova has closed the first tranche of its Series F funding round, raising $1 billion at an $11 billion valuation, as the AI chip startup accelerates its push to challenge NVIDIA in the AI inference infrastructure market. In this Bloomberg Tech interview, CEO Rodrigo Liang explains how SambaNova’s SM40 and SM50 chips target the decode phase of inference specifically, delivering 2–3x throughput gains when deployed alongside NVIDIA GPUs handling the prefill phase — achieving more output without increasing data center investment.
Liang highlights the company’s data-flow, SRAM-based architecture, which keeps power consumption at roughly 10 kilowatts per rack versus the 100 kilowatts typical of standard GPU racks. Unlike some SRAM-first competitors, SambaNova’s design also incorporates HBM to support trillion-parameter frontier models at full precision without quantization — a technical distinction Liang uses to differentiate from Cerebras.
The interview’s headline customer announcement is JPMorgan, which has selected SambaNova as its enterprise AI infrastructure provider, deploying on-premises racks to run regulated financial workloads entirely within the bank’s own firewalls. Liang notes SambaNova already serves dozens of production customers — cloud providers, service providers, and model builders — and expects to reach triple-digit customer count by year-end, betting that enterprise inference demand will be the next major growth wave in AI infrastructure.
📺 Source: Bloomberg Tech · Published July 08, 2026
🏷️ Format: Interview
