Blaize Unveils Pathfinder and Xplorer: GSP‑Based Edge AI Platforms Deliver 60x Efficiency and Code‑Free Development

In a landmark announcement, Blaize unveiled its Pathfinder and Xplorer platforms—embedded and accelerator solutions built on the company’s Graph Streaming Processor (GSP) architecture. These platforms aim to shatter the limitations of traditional GPU and CPU‑based edge AI by offering “code‑free” development and unmatched performance‑to‑power ratios.

The GSP delivers 16 cores and 16 TOPS of inference throughput while consuming only 7 W of power. Compared to conventional GPUs and CPUs, the architecture achieves up to 60‑fold higher system‑level efficiency, 50‑fold lower memory bandwidth usage, and 10‑fold reductions in inference latency. All features—multi‑threading, true streaming, and 100 % programmability—are implemented in silicon, allowing developers to tailor the processor to any neural network topology.

Blaize’s early‑adopter customers are already moving from proof‑of‑concept to production, deploying real‑time smart‑vision workloads across industry, automotive, retail, and smart‑city environments. CEO and co‑founder Dinakar Munagala explained, “Today’s edge solutions are either too small to handle the computational load or too expensive and complex to ship at scale. Blaize’s products break those barriers, unlocking AI at the edge and enabling a shift away from data‑center‑centric architectures.”

Omdia Research forecasts that edge‑AI chip revenue will hit $51 B by 2025, underscoring the market’s explosive growth as AI becomes a core requirement for emerging devices.

The two flagship products are: the Pathfinder P1600, a standalone system‑on‑module that embeds the GSP directly into sensor or network edge devices—no host processor required—and the Xplorer accelerator family that plugs into a PCIe slot on existing servers or appliances. The X1600E is a small‑form‑factor card designed for constrained spaces such as convenience‑store kiosks or industrial PCs, while the X1600P offers a half‑height, half‑width PCIe enclosure that can replace power‑hungry GPUs in edge servers, delivering 16‑64 TOPS of inference within a low‑power envelope.

The accompanying software suite—Picasso SDK and AI Studio—offers a fully open‑standard, code‑free visual interface that empowers both seasoned developers and domain experts without coding experience. Both tools integrate Blaize Netdeploy, an edge‑aware algorithm engine that optimizes accuracy and performance for on‑premise deployments.

Proof‑of‑Concept Use Cases

• Industrial – Five independent neural networks run at 50 FPS, achieving <100 ms aggregate latency for human and robot pose detection, product identification, and quality inspection.
• Smart City – Three networks at 50 FPS monitor human detection, pose, position, and traffic‑intersection safety, enabling real‑time security analytics.
• Sensor Fusion – High‑resolution FHD video combined with LiDAR/radar streams is processed as a native graph‑based application on the GSP.
• Last‑Mile Delivery – HD camera and LiDAR/radar fusion runs at 30 FPS with <100 ms latency and under 15 W power.
• Retail – Four networks at 50 FPS track human detection, position, mask compliance, theft prevention, and product detection for store safety.

Both product lines are now available in volume pricing: the Xplorer X1600E at $299, the Pathfinder P1600 SOM at $399, and the Xplorer X1600P at $999.