Pango Micro Positions for AI Computing: The Golden Window for Domestic FPGA

FPGA: The "Third Pole" of AI Computing

While the world fixates on NVIDIA's GPU shortage, the AI computing infrastructure is undergoing a structural shift. Large model training still relies on GPU clusters, but on the inference side—especially for edge inference and private deployments—FPGA is emerging as the "third pole" of AI computing, alongside GPUs and ASICs, thanks to its low latency, high energy efficiency, and reconfigurability.

According to Frost & Sullivan, the global FPGA market is projected to grow from approximately $8 billion in 2023 to over $15 billion by 2028, with a CAGR of about 13%. AI inference acceleration represents the fastest-growing segment, particularly in 5G base stations, industrial vision, autonomous driving, and smart grid applications where FPGA's real-time processing capabilities are hard to replace.

In the Chinese market, the convergence of domestic substitution policies and AI computing demand has created a rare "golden window" for local FPGA vendors. Pango Micro (紫光同创)—with the largest shipment volume and most comprehensive product portfolio among domestic FPGA companies—is accelerating its strategic positioning within this window.

Pango Micro's AI Computing Product Portfolio

Pango Micro currently offers three FPGA product families covering AI inference scenarios from cloud to edge:

• Titan Series (High-Performance): Targeting data centers and cloud inference, with logic capacity up to 1.8M LUTs, PCIe Gen3 support, and 28 Gbps SerDes. AI inference accelerator cards are already available, supporting CNN and Transformer model deployment at INT8/FP16 precision.
• Logos Series (Mid-Range): Positioned for edge AI and smart manufacturing, with power consumption of 5–15W. Suitable for real-time inference in industrial inspection, smart cameras, and similar scenarios.
• Compact Series (Low-Power): Designed for IoT endpoint inference, supporting TinyML model deployment with extremely low static power consumption, ideal for battery-powered devices.

On the software ecosystem front, Pango Micro has launched PDS (Pango Design Suite), which integrates a deep learning compiler capable of directly mapping ONNX-format models to FPGA logic, substantially lowering the barrier for migrating from GPU to FPGA. The maturity of this toolchain will be decisive in whether FPGA can truly enter the AI mainstream ecosystem.

DeepSeek Partnership: A Key Validation for FPGA-Based LLM Inference

In the second half of 2024, Pango Micro's collaboration with DeepSeek became a landmark event in the industry. Pango Micro's Titan series FPGA successfully adapted DeepSeek-V2/V3 LLM inference tasks and completed pilot deployment in a telecom operator's intelligent customer service scenario.

This partnership validated several key hypotheses:

• FPGA can run large language models: A single PCIe accelerator card can handle 7B–13B parameter models with first-token latency below 50ms;
• Significant energy efficiency advantage: Compared to GPU solutions at equivalent throughput, energy efficiency is improved by approximately 3–4×, making FPGA ideal for 24/7 online inference;
• Private deployment-friendly: The two parties plan to jointly launch FPGA-based inference appliances, offering turnkey on-premise AI deployment for government and enterprise customers.

Amidst the DeepSeek-driven wave of domestic LLM adoption, FPGA's role as a domestic computing foundation is being redefined—it is no longer merely a supporting player for signal processing and network acceleration, but a core option in AI inference infrastructure.

Next Stop: Advanced Process Nodes and Chiplet Architecture

Pango Micro's technology roadmap points toward an even more aggressive AI positioning:

• Titan-9000 Series (14nm): Doubles logic capacity to over 3M LUTs, with 64GB HBM2e integrated on-package, purpose-built for LLM inference;
• Edge AI SoC "Cuihu" (Emerald Lake): Integrates a proprietary AI acceleration engine with a RISC-V processor, consuming less than 8W while supporting real-time analysis of 16 channels of 1080p video, targeting industrial vision and intelligent transportation;
• 7nm Chiplet Solution: Using the UCIe standard to interconnect external HBM and compute dies, enabling flexible compute scaling—positioned to compete with Xilinx's Versal architecture.

From 28nm to 14nm, and with 7nm on the roadmap, Pango Micro's process node evolution clearly demonstrates its long-term commitment to the AI computing track.

Capital Support: STAR Market IPO on the Horizon

Pango Micro filed its IPO prospectus with the Shanghai Stock Exchange STAR Market in late 2023, seeking to raise approximately RMB 12 billion for advanced process R&D and AI product line development. The review process has progressed smoothly; if plans proceed as expected, a formal listing in 2026 would make Pango Micro the largest FPGA stock on the A-share market. The capital markets will provide ample financial ammunition for Pango Micro's AI computing ambitions.

Competing in a Three-Way Race: The Domestic FPGA Landscape

China's domestic FPGA market currently features a three-way rivalry among Pango Micro, Anlogic (安路科技), and Fudan Microelectronics (复旦微电). Pango Micro's core competitive moats include:

• Shipment leadership: Highest market share in domestic telecommunications and industrial sectors;
• Most complete product portfolio: Full coverage from low-power Compact to high-performance Titan series;
• Parent company synergy: The Unigroup ecosystem spans chip design, cloud services, and telecommunications equipment, offering natural advantages for system-level integration of AI computing solutions.

However, Xilinx (AMD) and Intel (Altera) still hold dominant positions in the high-end segment. Pango Micro's breakthrough path lies in using AI inference as the beachhead, establishing irreplaceability in mid-range and edge markets, then gradually moving up to challenge the high end.

HSY Perspective

Pango Micro's AI computing strategy carries multiple practical implications for Hengsen Technology's customers:

First, supply chain substitution window—Against the backdrop of ongoing US-China technology competition, demand for domestic FPGA substitution in telecommunications, power systems, and industrial control continues to grow. As the domestic FPGA leader, Pango Micro's product maturity and supply stability are rapidly improving. Downstream equipment makers should proactively evaluate migration roadmaps.

Second, edge AI deployment acceleration—Pango Micro's Logos series and "Cuihu" edge AI SoC offer cost-effective inference hardware options for industrial vision, smart surveillance, and predictive maintenance. For manufacturing enterprises planning AI upgrades, FPGA solutions' advantages in real-time performance and power efficiency merit serious attention.

Finally, milestones to watch—If Pango Micro successfully lists on the STAR Market in 2026, it will trigger a valuation reassessment of the entire domestic FPGA sector. Meanwhile, a successful 7nm chiplet tape-out would mark domestic FPGA's formal entry into global tier-one competition. We recommend that Hengsen customers continuously track Pango Micro's new product launches and ecosystem development, seizing the first-mover advantage in the restructuring of domestic computing supply chains.

This article is compiled from Pango Micro's official website, Frost & Sullivan industry reports, public media coverage, and industry research.