Blaize GSP Edge AI Platforms and AI Studio

Blaize is a Nasdaq-listed (BZAI) edge AI computing company based in El Dorado Hills, California, co-founded by Dinakar Munagala, Satyaki Koneru and Ke Yin, with offices in the US, UK, India, Japan and the UAE. It builds programmable, energy-efficient processors and software so that cities, retailers, factories and automakers can run multimodal AI inference near cameras and sensors instead of relying only on GPU data centres.

Its Graph Streaming Processor (GSP) architecture powers the Blaize 1600 SoC with 16 GSP cores and 16 TOPS, delivered as the Pathfinder P1600 system-on-module and Xplorer accelerator cards in EDSFF, PCIe and M.2 form factors, up to a four-chip PCIe card with 64-80 TOPS. The Picasso SDK programs streaming AI applications, while the code-free Blaize AI Studio covers the workflow from idea to deployed application. The newer Blaize AI Services Platform delivers AI applications as APIs for cloud providers, system integrators and enterprises, and hybrid servers complement GPU clusters for smart city video analytics.

• Graph Streaming Processor: programmable architecture with 16 GSP cores and 16 TOPS per Blaize 1600 SoC.
• Pathfinder P1600 SoM: embedded module with MIPI camera input, H.264 and H.265 video and industrial grade.
• Xplorer cards: EDSFF, PCIe and M.2 accelerators, up to a four-chip card with 64-80 TOPS.
• Picasso SDK: development platform for streaming AI applications that can be updated as models evolve.
• Blaize AI Studio: code-free workflow from idea to deployed edge AI application.
• AI Services Platform: delivers AI applications as APIs for cloud providers, integrators and enterprises.

A Vietnamese smart-city or traffic integrator can add Xplorer cards to video servers for multi-stream vehicle and incident analytics. A retail chain can deploy low-latency edge AI for security and shopper analytics with AI Studio. A factory or irrigation operator can run predictive maintenance and resource optimisation models at the edge.

Blaize claims up to 60 times higher system-level efficiency than CPU or GPU systems for edge inference, lowering energy use and total cost of ownership. Programmable GSP hardware can be updated as neural networks change, protecting the hardware investment. Code-free AI Studio lets non-specialist teams build and deploy vision applications. Hybrid deployments add inference capacity next to existing GPU infrastructure for multi-camera analytics.

Asian cities and infrastructure operators are building AI computing capacity for multi-camera video analytics, traffic and public safety, but GPU-only clusters are costly to power. Hybrid designs that place efficient inference accelerators next to GPUs are gaining traction in India, Indonesia, Japan and Korea, and Southeast Asian smart-city programmes face similar cost and energy constraints.

Listed on Nasdaq as BZAI, headquartered in El Dorado Hills, California, with more than 200 employees and offices in San Jose, Cary, Hyderabad, Leeds, Kings Langley and Abu Dhabi. Investors include DENSO, Samsung, Mercedes-Benz, Temasek and Franklin Templeton. In July 2025 it signed a contract worth at least 120 million USD with Starshine Computing Power for deployments in India, Indonesia, Japan, South Korea and China. The P1600 won the 2022 Edge AI and Vision Product of the Year award.