SambaNova

Scientific discovery is entering a new AI era and we're proud to help make it possible. SambaNova is honored to be part of U.S. Department of Energy (DOE)'s Genesis Mission Consortium. We're excited to be part of what's next. 🦾

"As the cost of tokens continue to rise because the infrastructure is so expensive, people are looking for alternatives. So at SambaNova, we come in with a much lower-cost infrastructure, much higher performance. We drive it at 10 kilowatts, an order of magnitude lower power than the traditional GPUs." — Rodrigo Liang During his chat with Matthew Miller on Bloomberg Television, Rodrigo explained how we're able to make AI more affordable: disaggregated inference. Instead of relying on a single architecture, it combines GPUs for prefill, RDUs for high-speed decode, and CPUs for orchestration, allowing each processor to do what it does best. The result is faster inference, lower latency, and significantly lower-cost tokens for enterprises and service providers deploying AI at scale. This architecture powers the world's first commercially available disaggregated inference cloud, including VC2, helping organizations bring AI into production with better performance and better economics. Watch their conversation below ⬇️

"Where dataflow comes in is... you can actually just flush these things like a river flushes through their own environment with a very fast pace." - Rodrigo Liang One of the biggest challenges in AI today isn't the model, it's getting data to the right place at the right time. That's where dataflow architecture comes in, helping eliminate memory bottlenecks and unlock faster, more efficient inference at scale. 🦾 🎥: Forbes

Your Claude Code just got even faster thanks to SambaCloud's latest update. ⚡️ Developers get instant access to our high-performance inference via the Anthropic SDK—no code changes required. Read more in our blog ⬇️

Say bonjour to our team at RAISE Summit 🇫🇷 We'll be talking about why one chip isn't enough anymore, and why the future of AI inference is all about using the right chip for the right workload. AI agents are changing what's required from infrastructure. Long contexts, multi-step reasoning, and continuous tool use demand a different approach to inference than traditional GPU-only architectures. That's why we're building premium inference with disaggregated infrastructure: • GPUs for prefill • RDUs for high-speed decode • CPUs to orchestrate the agent loop Faster inference, higher throughput, and better economics. See you in Paris: https://lnkd.in/gv7b3jrq

Open-weight models keep getting better and Gemma 4 31B is a great example. It combines frontier-class reasoning with production-ready coding performance and native agentic capabilities, making it a strong choice for everything from coding assistants to multi-agent applications. Some highlights: • 89.2% on AIME 2026 (no tools) • 80.0% on LiveCodeBench v6 • Native function calling, structured JSON output, and system prompt support for AI agents When paired with SambaCloud, developers also get the lowest latency available, helping agents respond faster and making interactive AI experiences feel more natural. We break down what makes Gemma 4 31B stand out—and why speed matters just as much as model quality. Read the blog: https://lnkd.in/e-QMY8K5

Inference is now the defining workload in AI, and that changes the economics of the whole stack. At Deep Tech Week SF, SambaNova Chief Product and Strategy Officer Abhi I. put it plainly: Inference is the defining workload for AI now and for the future. He further emphasized that training is a cost center; inference is where you make money. As such, hardware has to be purpose built for the task at hand. It’s not a one-size-fits-all story anymore. Heterogeneity is the order of the day: a combination of chips and networking that produces the best results. And, a combination of models—frontier and open source, large and small—that are fit to the task. Disaggregated inferencing is the right architecture for the Agentic Era. GPUs handle the prefill stage, SambaNova's RDUs handle decode, and CPUs orchestrate and run the tools. That is what lets you serve premium tokens, run the largest models running fast, with the power efficiency communities demand, at a TCO that works for the enterprise. As Abhi says, “More intelligence per joule — that’s the strategy.” It’s about achieving the lowest possible power consumption and deploying into existing data centers (air cooled). Abhi was joined by Rajiv K. of Velaura AI on the power layer and Barun Kar of Upscale AI on the networking layer, in a conversation moderated by Sriram Viswanathan of Celesta Capital. Three companies, three layers of the stack, one shared view: the GPU-centric, vertically integrated stack is the bottleneck, and purpose-built, heterogeneous infrastructure is the answer for the defining workload in AI - inferencing. Thank you to Celesta Capital for hosting. Lip-Bu Tan's Fireside Chat with Michael E Marks was a conversation between two silicon valley veterans full of wise advice to startup founders, such as “Stay flexible and adjust your trajectory if customers and markets change.”  We at SambaNova appreciate the call out for the success we are having from both of them and take their advice to heart.

"You need premium because your models are bigger, your models are lower latency, your models are much higher throughput." — Rodrigo Liang The bigger opportunity is better inference. Faster tokens. Better efficiency. Lower cost. That's what makes AI agents practical at scale. Thanks Akash Pasricha The Information TITV for the great conversation on where AI infrastructure is headed. 🦾 Vista Equity Partners

June was all about faster agents, premium inference, and the next gen of AI infrastructure ⚡️ TL;DR: - The first disaggregated inference demo for AI agents is live - Gemma 4 31B is running fastest on SambaCloud - Customers are scaling AI faster with SambaNova - Meet us in Paris in July 🔗 https://lnkd.in/ebtBBCSD

Our CEO Rodrigo Liang and Vista Equity Partners' Monti Saroya joined The Information TITV today to discuss Vista and Cambium Networks' cloud initiative & why inference is becoming the next battleground in AI. The future isn't one chip doing everything...it's disaggregated inference, where GPUs, RDUs, and CPUs work together to deliver premium inference at scale. 🦾 Watch their segment below ⬇️ https://lnkd.in/ghEcapkE