Startup Blaize Inc., which supplies artificial intelligence chips for edge devices and data centers, today disclosed that it has raised $106 million in fresh funding.

The capital was provided by a consortium that included Mercedes Benz, Franklin Templeton and a half-dozen other backers. The round comes less than six months after Blaize announced plans to go public through a merger with a special purpose acquisition company, or SPAC. The chipmaker said at the time that the deal was expected to raise at least $71 million.

Most AI models are implemented as a collection of code snippets called artificial neurons. Each such code snippet performs a small portion of the task assigned to the AI model in which it runs. Blaize has developed a system-on-chip, the Blaize 1600, that it says can run neural networks with better power efficiency than rival products.

At the heart of the chip’s design is a computer science concept known as graph processing. A graph is a data structure that contains multiple pieces of information dubbed nodes. Related pieces of information are linked together by connections, or edges, that help explain the relationship between them.

Practically all AI models can be represented as a graph. A model’s artificial neurons can be encoded into the nodes, or information snippets, that make up a graph. Meanwhile, the manner in which those neurons work together to crunch data may be represented using the connections that link together the nodes.

Blaize developed its Blaze 1600 chip from the ground up to process graphs. According to the company, the processor’s specialized design allows it to run AI models using less energy than graphics cards or field-programmable gate arrays. That makes it well-suited for powering connected devices, which often have battery life constraints. 

The Blaze 1600 features 16 cores optimized for graph processing. It also includes optimizations designed to speed up the task of processing intermediate results, additional information that AI models generate while crunching input data. Blaize says the chip can manage up to 16 trillion computing operations per second.

“Our unique, fully programmable approach makes us ready for the unknown,” said Blaize Chief Executive Officer Dinakar Munagala. “This is ideal in the fast-changing AI applications landscape, de-risking and reducing cost for our customers, scaling from the edge to the data center, with one uniform and complete hardware and software solution.”

The company sells its silicon in multiple form factors. For the edge computing market, Blaize offers compute modules that combine the Blaize 1600 with supporting components such as memory. Those modules have a compact design that makes it relatively simple to integrate them into connected devices.

The company also targets its chip at data centers. It offers a PCIe accelerator, the Xplorer X1600P-Q, with a single Blaize 1600 that customers can attach to existing servers. For organizations with more advanced requirements, the chipmaker offers a server of its own that combines 24 of its processors with two Intel Corp. central processing units.

Blaize will use the capital from its newly announced funding round to finance product development and commercialization initiatives. According to the company, the effort will place an emphasis on expanding its presence in the automotive, computer vision, AI inference and generative AI segments.

Image: Blaize