Cambricon’s Profit Surge Faces an Inventory Test

#ai-chips#inventory#chip-supply#financial-resultscambricon-ai-chipscambricon

Cambricon’s AI-chip growth is strong—but unfinished inventory could determine whether demand becomes deployable hardware

30-Second TL;DR

What Changed

First-half revenue reached approximately 6 billion yuan, nearly doubling year over year.

Why It Matters

For AI infrastructure buyers, strong financial growth may indicate rising demand for Cambricon accelerators, but unfinished inventory raises questions about production throughput and delivery timing. Customers should distinguish booked demand from chips that are actually completed and deployable.

What To Do Next

Ask Cambricon suppliers for confirmed delivery dates and completed-chip availability before committing production workloads to its accelerators.

Who should care:Enterprise & Security Teams

Key Points

•First-half revenue reached approximately 6 billion yuan, nearly doubling year over year.
•Net profit more than doubled during the same period.
•A sharp rise in unfinished inventory creates execution risk for second-half chip deliveries and revenue recognition.

Deep Insight

AI-generated analysis for this event — not the original article.

Enhanced Key Takeaways

•Cambricon's inventory buildup is largely attributed to the aggressive procurement of advanced packaging materials and high-bandwidth memory (HBM) components to mitigate potential supply chain restrictions.
•The company's reliance on TSMC and other advanced foundries remains a critical bottleneck, with inventory levels reflecting a 'stockpile-first' strategy ahead of anticipated geopolitical export control tightening.
•A significant portion of the revenue growth is driven by the MLU (Machine Learning Unit) series, specifically the MLU590, which has seen increased adoption in domestic large-scale model training clusters.
•Despite the profit surge, Cambricon continues to face high R&D expenditure ratios, with capital allocation heavily skewed toward the development of next-generation interconnect technologies and proprietary software stacks.
•Market analysts note that the 'unfinished inventory' consists primarily of work-in-progress (WIP) wafers that require complex post-processing, making the conversion rate to finished goods highly sensitive to foundry yield rates.

Competitor Analysis

Architecture

Cambricon (MLU590): Proprietary MLU
Huawei Ascend (910B): Da Vinci
NVIDIA (H20): Hopper (Modified)

Primary Market

Cambricon (MLU590): Domestic Cloud/Data Center
Huawei Ascend (910B): Domestic Cloud/Government
NVIDIA (H20): China-specific Export Market

Software Ecosystem

Cambricon (MLU590): Cambricon Neuware
Huawei Ascend (910B): CANN
NVIDIA (H20): CUDA (Limited)

Manufacturing

Cambricon (MLU590): Advanced Foundry (Restricted)
Huawei Ascend (910B): SMIC (Domestic)
NVIDIA (H20): TSMC (Export Restricted)

Feature	Cambricon (MLU590)	Huawei Ascend (910B)	NVIDIA (H20)
Architecture	Proprietary MLU	Da Vinci	Hopper (Modified)
Primary Market	Domestic Cloud/Data Center	Domestic Cloud/Government	China-specific Export Market
Software Ecosystem	Cambricon Neuware	CANN	CUDA (Limited)
Manufacturing	Advanced Foundry (Restricted)	SMIC (Domestic)	TSMC (Export Restricted)

Technical Deep Dive

MLU590 Architecture: Utilizes a multi-core chiplet design to improve yield and scalability for large-scale AI training tasks.
Interconnect Technology: Employs proprietary high-speed chip-to-chip interconnects designed to reduce latency in multi-node training clusters.
Memory Integration: Supports high-capacity HBM3 integration to address memory bandwidth bottlenecks common in transformer-based model training.
Software Stack: Relies on the Neuware platform, which provides optimized kernels for deep learning frameworks like PyTorch and TensorFlow, though it lacks the mature ecosystem depth of CUDA.

Future ImplicationsAI analysis grounded in cited sources

Cambricon will face margin compression in Q4 2026.

The high cost of stockpiling advanced components and the potential for lower-than-expected yield rates on unfinished inventory will likely increase the Cost of Goods Sold (COGS).

Domestic market share will stabilize as supply chain constraints persist.

Continued reliance on external foundries for advanced nodes limits the company's ability to scale production to meet the full demand of the domestic AI infrastructure market.

Timeline

2016-03

Cambricon Technologies is founded as a spin-off from the Chinese Academy of Sciences.

2020-07

Cambricon completes its IPO on the Shanghai Stock Exchange STAR Market.

2023-04

Release of the MLU590 series, marking a significant shift toward high-performance training chips.

2025-02

Company reports a strategic pivot to prioritize domestic supply chain partnerships for packaging and assembly.

2016-03
Cambricon Technologies is founded as a spin-off from the Chinese Academy of Sciences.
2020-07
Cambricon completes its IPO on the Shanghai Stock Exchange STAR Market.
2023-04
Release of the MLU590 series, marking a significant shift toward high-performance training chips.
2025-02
Company reports a strategic pivot to prioritize domestic supply chain partnerships for packaging and assembly.