The startup Blaize Inc. revealed today that it has raised $106 million in new funding. Blaize Inc. provides artificial intelligence chips for edge devices and data centers.
A group comprising Franklin Templeton, Mercedes Benz, and six additional supporters contributed the funding. The round was disclosed fewer than six months ago, when Blaize intended to merge with a special purpose acquisition company, or SPAC, in order to go public. At the time, the chipmaker stated that the transaction was anticipated to generate at least $71 million.
Artificial neurons, a collection of code fragments, are used to implement the majority of AI models. Each of these snippets of code completes a certain part of the task given to the AI model it is running. Blaize claims that its Blaize 1600 system-on-a-chip can operate neural networks more power-efficiently than competing offerings.
Graph processing, a notion from computer science, is at the core of the chip’s design. A graph is a type of data structure made up of several nodes, or distinct pieces of information. Connections, or edges, join related pieces of information and aid in illuminating their relationship.
Graphs can be used to depict almost any AI model. Artificial neurons in a model can be encoded into the nodes, or bits of information, that comprise a graph. In the meantime, the connections that bind the nodes together may be used to depict how those neurons cooperate to process information.
Blaize started from scratch while creating the Blaze 1600 chip, which processes graphs. The business claims that the CPU can run AI models with less energy consumption than graphics cards or field-programmable gate arrays because of its unique design. This makes it ideal for charging linked gadgets, which frequently have limited battery life.
Opted for graph processing, the Blaze 1600 boasts 16 cores. The aforementioned comprises of optimizations aimed at expediting the processing of intermediate outputs, which are supplementary data produced by AI models during the process of grinding through input data. Up to 16 trillion computation operations can be handled by the chip each second, according to Blaize.
“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 silicon is sold by the manufacturer in several form factors. Blaize provides compute modules that bundle the Blaize 1600 with auxiliary parts like memory for the edge computing industry. Because of their small size, the modules are reasonably easy to integrate into other linked devices.
The chip is also aimed towards data centers by the firm. Customers can link the Xplorer X1600P-Q, a PCIe accelerator with a single Blaize 1600, to their current servers. The chipmaker offers a server of its own that combines two Intel Corp. central processing units and 24 of its processors for enterprises with more sophisticated needs.
Blaize plans to fund product development and commercialization initiatives with the proceeds of its recently disclosed fundraising round. The company states that increasing its footprint in the automotive, computer vision, AI inference, and generative AI sectors will be a key focus of the endeavor.