SambaNova Raises $1B, Signs JPMorganChase as a Customer

2026-07-08T07:45:00Z

PARIS, France

Established AI chip startups have seen valuations rocketing after high-profile exits from Groq and Cerebras, but SambaNova’s valuation is based on its true market value, SambaNova CEO Rodrigo Liang told EE Times.

“The new valuation is reflective of our business hockey-sticking,” Liang said. “Having 40 or 50 customers of a certain size was meaningful to our investors.”

Investors understand the limitations of other architectures, Liang added.

“Big models need memory capacity,” he said. “Even the open-source models are multi-trillion parameters now, and as far as I can see, most people are not willing to give up quality for quantized versions of the model… that’s what we’re solving for.”

SambaNova’s SN50 can be connected in large clusters of hundreds of chips, he said.

SambaNova is considering its options for a potential IPO, he said, but the fundraise has given the company ample runway.

“With the fundraise, we can go indefinitely, because we have a very low burn [rate], relatively,” he said. “We don’t build data centers; we don’t have to secure land or power; we don’t do any of those things. We don’t have leasebacks. We’re in the business of building and shipping racks. That gives us a balance sheet that can go for years.”

SambaNova expects to be profitable next year, he added.

Inference customer

SambaNova also announced JPMorganChase as a customer. The bank will deploy SambaNova’s SN40 and SN50 systems to power secure on-prem AI inference.

“The banks have been very active [using AI], but nobody is more active than JPMorganChase,” Liang said. “The [enterprise] market is getting to the point where people need inference, and they want to bring it back on-prem. They don’t have the power, and the models are too big, so there’s a nice convergence with the models people want, the fact that they want data privacy and security.”

SambaNova has a multi-year agreement in place with JPMorganChase.

“This is a breakthrough because most suppliers haven’t figured out how to get into the enterprise,” Liang said. “The enterprise [market] is going to be significant; it won’t be 100%, but it’s going to be a meaningful player in the overall market, and we’re starting to see that.”

Enterprise customers are looking to host private models and private data in their own environment, which is unlikely to be a gigawatt-scale data center, Liang said.

“They love air cooling, single racks, and open standards,” he said. “They love standard Ethernet and standard Red Hat, and they’re rolling [racks] into existing data centers that they already deployed.”

Disaggregated inference

At Computex a few weeks ago, SambaNova demonstrated its SN40 chips running decode acceleration in a system with Nvidia Blackwell GPUs used for prefill and Intel Xeon 6 CPUs running the application. This setup boosted performance 2× to 3× over a GPU-only system on a per-chip basis, according to SambaNova.

Disaggregated inference is about companies coming together to find better solutions for customers, Liang said, noting that while SambaNova can capably handle both prefill and decode on the same hardware, he is nevertheless happy that Nvidia is making it easier to interface with its GPUs in this kind of architecture.

“The reason we’ve leaned into it hard is memory costs are up,” he said. “We can offer our decode accelerator with a lower memory configuration, which saves our customers money, and then reuse all the memory that’s inside the GPUs that are already in data centers.”

Leaning into disaggregation saves customers money, improves economics, and allows SambaNova to deploy faster because customers can see the racks running side by side, Liang said.

“If they don’t need it, they can redeploy it later; it’s all dynamic, it’s all open interfaces,” he said.

Disaggregated architectures with Nvidia GPUs plus SambaNova RDUs are the preferred configuration for cloud service providers, Liang said. However, enterprises that need data privacy and security with predictable response times don’t need disaggregation and are better suited to SambaNova-only on-prem clusters.

This Nvidia-plus-SambaNova inference architecture will be deployed at scale by a new venture from the venture capital firms Vista Equity Partners and Cambium Capital, called Vector Core Compute (VC2). VC2 is a distributed cloud provider for ultra-low-latency and ultra-low-cost tokens based in the biggest metropolitan areas in the U.S. These clusters are being deployed in existing data centers in urban areas where gigawatts of power aren’t necessarily available, but these areas are very close to the businesses they serve. VC2 has placed a $3.5-billion order with SambaNova, Liang said, to be deployed over the next three years.

VC2 is a standalone entity that is now securing existing data center space, which is enabling the new company to get services up and running quickly. The venture capital firms behind the new entity have dozens of portfolio companies all moving to AI-first operations, Liang said, making a ready-made customer base.

“With their own cloud, they get to create more performance for their companies and capture value on the back-end too,” he said.

For agentic AI where the systems of record are still running on CPUs, SambaNova is working with Intel to enable efficient interconnect to Xeon CPU racks. The solution uses standard open interfaces such as vLLM, but there is a level of optimization between the Intel Xeon stack and SambaNova’s stack to give a better user experience, Liang said. The new go-to-market relationship with Intel will kick in when the SN50 becomes available in production quantities.

The eventual aim is to make today’s applications agent-ready so they can interface with

“In the end, you’ll see merging of the AI model world and the traditional application world,” Liang said.

SambaNova Abandons Intel Acquisition, Raises Funding Instead

SambaNova Teams Up With Intel on Disaggregated Inference

RELATED TOPICS: AGENTIC AI AI AI INFERENCE FUNDING ICS SEMICONDUCTORS STARTUPS VENTURE CAPITAL FUNDING

COMPANIES: SAMBANOVA SAMBANOVA SYSTEMS

Sally Ward-Foxton covers AI for EETimes.com and EETimes Europe magazine. Sally has spent the last 18 years writing about the electronics industry from London. She has written for Electronic Design, ECN, Electronic Specifier: Design, Components in Electronics, and many more news publications. She holds a Masters' degree in Electrical and Electronic Engineering from the University of Cambridge. Follow Sally on LinkedIn