AI Unicorns: SambaNova’s Chip Competes With Nvidia

Photo Credit: Tony Melony from Pixabay


According to a recent report, the global artificial intelligence (AI) chip market size was estimated at $16.86 billion in 2022 and is expected to grow at 30% CAGR to reach $227.48 billion by 2032. While the market leader in the industry remains NVIDIA, there are other players like SambaNova that are playing in both the hardware and the software segment of the rapidly growing AI industry.


SambaNova’s Offerings

Palo Alto-based SambaNova was founded in 2017 by Stanford professors Kunle Olukotun and Chris Re along with Sun Microsystems and Oracle alumnus Rodrigo Liang. The trio wanted to develop a next-generation computing platform that could power machine learning and data analytics. They believed that the best way to power the next generation of AI applications was to create a new hardware and software platform instead of tweaking the already existing systems.

The company has built an enterprise-ready AI platform to help organizations gain value from AI and deep learning. Its flagship offering, SambaNova Suite, claims to help organizations realize the value of AI 22x faster. It is a complete AI solution for enterprises and governments that delivers performance and accuracy, along with security and data privacy, model ownership, simplified management, and the flexibility to scale.

The suite includes the SambaNova DataScale system, the SambaStudio software, and the SambaNova Composition of Experts (CoE) model architecture. SambaNova DataScale is among the world’s fastest hardware platform for AI with the ability to train and run hundreds of models on a single, energy efficient node. The SambaStudio offering provides organizations with access to software that can help them manage their dedicated infrastructure to easily scale, deploy, and manage access from a single app. The CoE is a model architecture that integrates multiple models to deliver more efficient and more accurate performance than would be possible from a single model.

SambaNova offers organizations the ability to tap into SambaNova’s AI system, through its Dataflow-as-a-Service, an on-demand, subscription-based solution. The offering allows enterprises to focus on the applications that run on it, instead of focusing on maintaining those systems.

SambaNova powers its offerings through its own chip – the SN40L Reconfigurable Dataflow Unit (RDU). The fourth generation of its chip has been purpose-built for AI workloads to deliver efficient dataflow architecture and memory footprint across all model sizes.


SambaNova’s Financials

SambaNova is privately held so far and does not disclose financials. It earns revenues through the subscription and usage fees it charges for access to its solutions.

It has raised $1.1 billion so far from six rounds of funding from SoftBank, Walden International, GV, Intel Capital, Red Line Capital Management, Atlantic Bridge, Celesta, Micron, Samsung Catalyst Fund, SK Telecom, Temasek, and Black Rock. Its last round of funding was held in April 2021 when it raised $676 million at a valuation of $5.1 billion.

SambaNova competes directly with companies like H2O.AI that offer a similar product. Recently, Microsoft also announced the launch of Git Hubs Models which is expected to allow developers to test and experiment with their AI models for free. The company also competes with companies like NVIDIA and Graphcore for its chip offering.


More By This Author:

AI Unicorns: Hugging Face’s Successful Pivot Away From Teenage Chatbot Engine
AI Unicorns: Mistral AI Brings Stiff Competition To Other LLMs
AI Unicorns: Icertis Gearing Up For An IPO

Disclosure: All investors should make their own assessments based on their own research, informed interpretations, and risk appetite. This article expresses my own opinions based on my own research ...

more
How did you like this article? Let us know so we can better customize your reading experience.

Comments

Leave a comment to automatically be entered into our contest to win a free Echo Show.
Or Sign in with