• Blaize claims that its Blaize 1600 system-on-a-chip can operate neural networks more power-efficiently than competing offerings.
  • Blaize provides compute modules that bundle the Blaize 1600 with auxiliary parts like memory for the edge computing industry.

AI chip manufacturing startup Blaize Inc. raises USD 106 million in the latest funding. The company manufactures AI chips for data centers and edge devices.

A group constituting Mercedes Benz, Franklin Templeton, and six other supporters backed the funding. The round was disclosed fewer than six months ago when Blaize intended to merge with a special purpose acquisition company, or SPAC, to go public. Back then, the chip developer claimed that the deal was expected to generate a minimum sum worth USD 71 million.

Artificial neurons, comprised of code segments, serve as the backbone for the majority of AI models, with each segment fulfilling a specific function within the task assigned to the AI model. Blaize asserts that its Blaize 1600 system-on-a-chip can execute neural networks with greater power efficiency compared to competitor’s solutions.

A computer science principle, graph processing forms the base of the chip’s design. A graph is a data structure comprising several nodes or distinct pieces of information. Connections, also known as edges, link correlated information fragments, shedding light on their interrelation.

Graphs offer a versatile means of visualizing nearly any AI model. Within a model, artificial neurons can be represented as the nodes or informational units of a graph. Concurrently, the connections that link these nodes illustrate the collaborative processing of information by these neurons.

Blaize embarked on a ground-up endeavor in developing the Blaze 1600 chip, tailored for graph processing. The company asserts that the CPU’s distinctive design enables it to execute AI models with lower energy consumption compared to graphics cards or field-programmable gate arrays. This makes it ideal for charging linked gadgets with limited battery life.

Sixteen cores on the Blaze 1600 are tuned for graph processing. It also contains enhancements made to expedite the processing of intermediate outcomes or extra data produced by AI models during the crunching of input data. According to Blaize, the chip is capable of performing up to 16 trillion calculations in a second.

Blaize Chief Executive Officer Dinakar Munagala said, “Our unique, fully programmable approach makes us ready for the unknown. 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 manufacturer sells the silicon in several form factors. Blaize offers compute modules that combine the Blaize 1600 chip with additional components such as memory, catering to the edge computing sector. Due to their compact dimensions, these modules are relatively straightforward to incorporate into interconnected devices.

The company also targets data centers with the chip. Customers have the option to connect the Xplorer X1600P-Q, a PCIe accelerator featuring a single Blaize 1600, to their existing servers. The chipmaker offers a server that combines two Intel Corp. central processing units and 24 processors for enterprises with more sophisticated needs.

The funds from Blaize’s recently disclosed investment round will be used to support product development and commercialization efforts. The business states that the initiative will prioritize growing its market share in generative AI, computer vision, automotive, and AI inference.