How does this company make money?
Cambricon earns money each time it sells a chip to a cloud provider like Alibaba Cloud or Tencent Cloud, or to an OEM customer building hardware. It also charges annual licensing fees per developer seat for access to the Neuware software development kit and for technical support tied to that software.
What makes this company hard to replace?
Customers who want to leave must first rewrite and retrain their AI models using a different software development kit — the Neuware SDK compiles code in a way that only runs on MLU processors, so none of that work transfers. On top of that, MLU-Link uses its own connection protocol that is incompatible with standard interfaces like PCIe or Nvidia's NVLink, meaning the physical server hardware must also be redesigned before a replacement chip can even be plugged in.
What limits this company?
The hard ceiling is how many chips TSMC can make for Chinese customers at the 7nm node. U.S. export controls already block Cambricon from using any more advanced process, and if that allocation shrinks — or if the controls tighten further — fewer chips can be produced, because no other available factory can meet the electrical density that MLU-Link's design requires.
What does this company depend on?
Cambricon cannot operate without TSMC and SMIC for chip fabrication, ARM for the instruction set architecture licenses that underpin its processor cores, Cadence and Synopsys for the EDA software tools used to design chips, ASE Group for advanced chip packaging, and SK Hynix and Samsung for the high-bandwidth memory that goes into every chip.
Who depends on this company?
Alibaba Cloud and Tencent Cloud rely on Cambricon chips to run AI inference for their recommendation engines and computer vision services — without them, that processing slows or stops. Baidu Apollo uses Cambricon hardware for real-time perception in its autonomous vehicle systems, and degraded performance there affects driving safety decisions. Xiaomi and other Chinese smartphone makers use Cambricon chips for on-device AI tasks like camera processing and voice recognition.
How does this company scale?
New chip variants and expansions of the Neuware software stack can be rolled out to many more customers without significant additional engineering cost — the architecture replicates cheaply once it exists. What cannot scale freely is foundry access: TSMC's capacity available to Chinese customers is capped by geopolitics and export controls, so every unit of growth ultimately runs into the same physical production ceiling.
What external forces can significantly affect this company?
U.S. Entity List designations and CHIPS Act export controls already block Cambricon from the most advanced manufacturing nodes and threaten access to EDA software tools like those from Cadence and Synopsys. Chinese government policy pushes for domestic alternatives to foreign technology, which creates pressure to reduce reliance on ARM, TSMC, and foreign memory suppliers. When the yuan falls against the dollar, the cost of paying TSMC in Taiwan rises automatically, squeezing margins on every chip produced.
Where is this company structurally vulnerable?
If U.S. export controls were extended to cover 7nm chips — not just the more advanced nodes they already restrict — TSMC could no longer legally manufacture Cambricon's chips at all. MLU-Link exists only as a physical chip; if that chip cannot be made, the interconnect disappears, and the customer lock-in it created has nothing left to run on.
CompanyGraph