What Is Happening With China AI Chipmakers?
China AI chipmakers are raising prices because the cost of securing high-bandwidth memory has increased sharply, while access to advanced HBM supplies has become more difficult.
Huawei, Cambricon, MetaX and Iluvatar CoreX are among the domestic companies trying to supply China’s rapidly growing AI-computing market. Their products have become increasingly important as U.S. export controls have restricted Chinese companies’ access to Nvidia’s most advanced processors.
But replacing Nvidia hardware requires more than producing an AI processor.
An AI accelerator needs high-speed memory to handle the enormous amount of data involved in training and running AI models. When that memory becomes scarce or expensive, the final price of the accelerator card can rise substantially.
Definition + Expansion: What Is HBM?
High-bandwidth memory (HBM) is a type of memory designed to move very large amounts of data between memory and a processor at extremely high speeds.
Unlike conventional memory configurations, HBM uses vertically stacked memory chips. This architecture allows AI processors to access large volumes of data quickly, making HBM particularly important for demanding AI workloads.
In an AI accelerator, the processor may perform the calculations, but the memory supplies the data needed for those calculations. If the memory becomes a bottleneck, improving the processor alone cannot solve the entire performance problem.
That is why HBM has become strategically important in the global AI-chip race.
Question → Direct Answer: Why Does HBM Affect AI Chip Prices So Much?
Because HBM represents a significant portion of an AI accelerator’s production cost. When suppliers have to pay substantially more to obtain the memory, those additional costs can flow directly into the price of the finished accelerator card.
Reuters reported that HBM obtained through grey-market channels in China can cost several times more than what buyers outside China pay.
That creates a difficult equation for domestic chipmakers: they are expected to produce more AI hardware while simultaneously paying more for one of its most important components.
How Much Have Huawei AI Chip Prices Increased?
Huawei’s upcoming Ascend 950DT has become one of the clearest examples of how HBM shortages are affecting China’s AI-chip market.
According to Reuters, Huawei has raised the indicated price of the accelerator card to above 250,000 yuan, or approximately $37,255 based on the exchange-rate figure cited in the report.
The increase is particularly notable because the new price is reportedly 20% to 50% higher than quotations offered to customers only two months earlier.
The exact increase depends on the terms of individual contracts.
Huawei has publicly said that its Ascend 950 series will become available in the fourth quarter of 2026.
Huawei’s Ascend 950 Series
The Ascend 950 family includes different products aimed at different parts of the AI workload.
| Chip/product | Reported role | Pricing development |
| Ascend 950DT | AI model development and generating responses | Above 250,000 yuan; up 20%–50% from recent quotes |
| Ascend 950PR | Processes user requests before they reach the model | Rose from about 60,000 yuan to more than 80,000 yuan |
| Ascend 910C | Earlier-generation AI accelerator | Rose from about 90,000 yuan to more than 110,000 yuan |
| Cambricon 690 | Planned next-generation AI chip | Indicated price up 20%–30% |
| Cambricon 590 | Current flagship | Current flagship before 690 launch |
The price increases are not limited to Huawei’s newest products.
Reuters reported that Huawei’s Ascend 950PR, which sold for roughly 60,000 yuan per card at the beginning of 2026, now costs more than 80,000 yuan, representing an increase of about 30%.
The earlier-generation Ascend 910C has also reportedly increased from approximately 90,000 yuan to more than 110,000 yuan.
Question → Direct Answer: Is Only Huawei Raising Prices?
No. Cambricon has reportedly raised the indicated price of its planned 690 chip by 20% to 30%, while smaller domestic rivals MetaX and Iluvatar CoreX have also increased prices, according to sources cited by Reuters.
That matters because it suggests the problem is broader than the pricing strategy of a single company.
When multiple suppliers face the same shortage of a critical component, higher prices can become a market-wide issue.
Why Is HBM So Difficult for China to Obtain?
The HBM shortage is closely connected to international export controls.
The advanced HBM market is dominated by SK Hynix and Samsung Electronics of South Korea, and Micron Technology of the United States, according to the Reuters report.
China’s access to certain advanced HBM products became more restricted after Washington tightened export controls in December 2024.
As a result, Chinese chipmakers have increasingly relied on grey-market channels to obtain supplies, according to sources cited by Reuters.
Definition + Expansion: What Is a Grey Market?
A grey market is a supply channel in which products are obtained and resold outside the manufacturer’s normal authorized distribution network.
Grey-market transactions are different from ordinary procurement because buyers may have to rely on intermediaries and alternative supply routes rather than established supplier relationships.
In this case, Reuters reported that HBM acquired through such channels can cost Chinese buyers several times more than the prices paid by buyers outside China.
That premium becomes particularly significant when the component itself represents a large share of an AI accelerator’s cost.
Question → Direct Answer: Why Can’t Chinese Chipmakers Simply Buy More HBM?
Because access to advanced HBM products has been constrained by export controls, while the global market is also experiencing strong demand.
Chinese companies therefore face both availability and pricing problems.
Even if a domestic AI-chip designer can develop a new processor, obtaining enough suitable high-bandwidth memory at a competitive price can remain difficult.
This creates a supply-chain bottleneck that can slow down the broader development of domestic AI infrastructure.
How Is China Building Alternatives to Nvidia?
China has been pushing domestic AI hardware as access to Nvidia’s most advanced processors becomes more restricted.
This has created an opening for companies such as Huawei, Cambricon, MetaX and Iluvatar CoreX.
The opportunity is substantial. Reuters described China’s AI-chip market as worth approximately $50 billion, highlighting the size of the domestic demand that Chinese chipmakers are attempting to serve.
Question → Direct Answer: Why Are Domestic AI Chips So Important to China?
Domestic AI chips can help Chinese technology companies build AI-computing capacity without relying as heavily on restricted foreign hardware.
That makes companies such as Huawei strategically important to China’s AI ecosystem.
However, domestic substitution is not simply a question of designing a chip that performs calculations. Companies also need memory, manufacturing capacity, packaging, software ecosystems and reliable supply chains.
The HBM shortage demonstrates how these different parts of the technology stack are connected.
Why Is the HBM Shortage a Bigger Problem Than Just Higher Prices?
Higher chip prices can have consequences far beyond individual purchases.
AI models require enormous computing resources. Companies building AI infrastructure typically need large numbers of accelerators rather than just a handful of chips.
If each accelerator becomes more expensive, the total cost of expanding AI-computing capacity rises.
For Chinese technology companies, that can make large-scale AI infrastructure more expensive at precisely the time when demand for computing is increasing.
The Supply Chain Has Several Pressure Points
The current situation illustrates several connected challenges:
- HBM availability: Advanced memory is difficult for Chinese companies to obtain.
- Higher procurement costs: Grey-market HBM can reportedly cost several times more.
- AI accelerator pricing: Higher component costs are being reflected in chip and card prices.
- Export restrictions: Controls limit access to some advanced foreign technologies.
- Growing domestic demand: Chinese technology companies continue to require AI computing capacity.
- Supplier allocation: Chipmakers must decide how to distribute limited GPU supplies among customers.
- Scaling pressure: Higher prices make it more expensive to build large AI clusters.
The important point is that an AI chip is not an isolated product.
It sits inside a much larger hardware supply chain.
How Are Chinese Chipmakers Allocating Scarce AI GPUs?
The shortage is also changing how domestic suppliers distribute their products.
Reuters reported that Iluvatar CoreX has doubled its GPU shipments to ByteDance to 100,000 units in 2026, according to one source.
ByteDance, the technology company behind TikTok, is one of the major consumers of AI computing in China.
Iluvatar CoreX has reportedly diverted GPUs originally intended for internal use to help meet ByteDance’s needs.
Question → Direct Answer: Why Does ByteDance Matter in This Story?
ByteDance is a major AI-computing customer, so its demand can influence how scarce domestic GPUs are allocated.
When a supplier redirects hardware from internal use toward a major customer, it shows how intense computing demand has become.
It also highlights an important trade-off for smaller AI-chip companies: using chips for their own development versus selling them to customers who urgently need computing capacity.
Reuters reported that Huawei is ByteDance’s largest domestic AI-chip supplier, followed by Cambricon and Iluvatar CoreX.
ByteDance did not respond to Reuters’ request for comment.
What Do Huawei’s HBM Technologies Mean for the Ascend 950?
Huawei has said that its Ascend 950 series will use two proprietary HBM technologies.
The 950PR will use HiBL 1.0, while the 950DT will use HiZQ 2.0.
Huawei has not detailed where or how those memory technologies are manufactured.
This is significant because the memory question sits at the heart of the current pricing pressure.
Developing proprietary technologies can help a company reduce dependence on external components over time, but it does not automatically eliminate the need for a broader manufacturing and supply ecosystem.
Question → Direct Answer: What Is the Difference Between the 950DT and 950PR?
The Ascend 950DT is designed mainly for developing AI models and generating responses, while the 950PR processes user requests before they are passed to the model.
In simple terms, they target different stages of AI workloads.
That distinction matters because AI systems increasingly divide work across different types of computing tasks rather than relying on a single processor for everything.
What Does This Mean for China’s Nvidia Challenge?
The price increases reveal both progress and difficulty in China’s attempt to build an AI-chip ecosystem that can compete with or substitute for Nvidia hardware.
On one hand, domestic companies are developing new processors and expanding their role in China’s AI market.
On the other hand, they are encountering bottlenecks in components such as HBM.
Definition + Expansion: What Is an AI Accelerator?
An AI accelerator is specialized computing hardware designed to perform the mathematical operations used heavily by artificial intelligence systems.
GPUs are one type of AI accelerator and are widely used for training and running AI models because they can perform many calculations in parallel.
AI accelerator cards combine processors with memory and other components. That means their overall performance and cost depend on the entire package, not just the processor itself.
This is why an AI-chip race can quickly become a race over memory, manufacturing and packaging as well.
Question → Direct Answer: Does Higher Pricing Mean Chinese AI Chips Are Failing?
Not necessarily.
Higher prices show that Chinese chipmakers are facing significant supply and cost pressures, but they do not by themselves establish that the underlying processors are unsuccessful.
In fact, the reported price increases also reflect strong demand for domestic AI hardware in China.
The bigger issue is whether Chinese suppliers can eventually increase production, secure affordable memory and build a competitive ecosystem at scale.
How Does This Affect Chinese AI Companies?
For Chinese AI companies, cloud providers and technology firms, the immediate issue is the cost of computing.
AI development requires repeated access to accelerators for activities such as model training, inference and experimentation.
When accelerator prices rise, companies may have to spend more to obtain the same amount of hardware.
That can make computing budgets more difficult to manage.
Question → Direct Answer: Could Higher Chip Prices Slow AI Development?
They could increase the cost of expanding computing capacity, particularly for companies that need large numbers of accelerators.
However, the actual impact will depend on chip availability, demand, production growth and the ability of suppliers to secure memory and other components.
The Reuters report does not establish that Chinese AI development will slow, but it clearly shows that hardware costs are becoming an important constraint.
What Should Developers and AI Students Learn From This?
The biggest lesson is that AI progress depends on infrastructure.
It is easy to think of AI as primarily a software story: better models, smarter algorithms and improved applications.
But underneath those applications is a huge hardware ecosystem.
A shortage of one component such as HBM can affect accelerator prices, infrastructure budgets and the ability of companies to deploy AI at scale.
For students and early-career professionals, understanding this broader ecosystem can be valuable.
Three Layers to Watch
1. AI processors
These perform the calculations required by AI workloads. Nvidia is a major global player, while companies such as Huawei and Cambricon are developing domestic alternatives in China.
2. High-bandwidth memory
HBM supplies processors with data at very high speeds. It has become one of the most strategically important components in AI computing.
3. AI infrastructure
Processors and memory ultimately become part of servers, data centres and computing clusters used to train and run AI models.
A bottleneck in any one layer can affect the others.
China AI Chipmakers Face a Hardware Ecosystem Test
The current price increases show why replacing Nvidia is a much larger challenge than creating another processor.
China AI chipmakers must compete within an ecosystem where memory, manufacturing, software and supply-chain access all influence the final product.
Huawei’s Ascend 950 series illustrates the scale of the ambition. The company is preparing new AI accelerators while also developing proprietary memory technologies.
Cambricon is preparing its next-generation 690, while MetaX and Iluvatar CoreX are also expanding their presence.
But the HBM shortage shows that hardware independence cannot be achieved through processor design alone.
Question → Direct Answer: What Is the Biggest Bottleneck Right Now?
According to the Reuters report, high-bandwidth memory is a major bottleneck because export restrictions and limited supply have made advanced HBM significantly more expensive for Chinese chipmakers.
That bottleneck is feeding into accelerator prices.
The result is a striking contradiction: China’s push for greater AI-chip independence is increasing the importance of domestic hardware at the same time that some critical components remain difficult to source.
What Happens Next for China’s AI Chip Market?
The next phase of China’s AI-chip development will likely depend on how quickly domestic suppliers can address supply-chain constraints.
Three questions will be particularly important.
Can Chinese companies secure enough advanced memory?
The availability and price of HBM will directly affect the economics of AI accelerators.
Can domestic chipmakers scale production?
Demand is clearly growing, but meeting that demand requires reliable manufacturing and component supplies.
Can domestic hardware become economically competitive?
Performance matters, but so does total cost. A domestic AI accelerator that is difficult or expensive to produce may face a different challenge from one that can be manufactured at scale.
The reported price increases therefore offer a snapshot of a much bigger transition in China’s AI industry.
China is building alternatives to Nvidia, but doing so requires control over an increasingly complex technology stack.
Key Takeaways: China AI Chipmakers and the HBM Shortage
Here are the most important points from the Reuters report:
- Huawei’s Ascend 950DT has reportedly been priced above 250,000 yuan ($37,255).
- The new 950DT price is 20% to 50% higher than quotes given about two months earlier.
- Cambricon’s planned 690 chip has reportedly been repriced 20% to 30% higher.
- Huawei’s Ascend 950PR has risen from roughly 60,000 yuan to more than 80,000 yuan since the beginning of 2026.
- Huawei’s Ascend 910C has reportedly increased from around 90,000 yuan to more than 110,000 yuan.
- Advanced HBM is dominated by SK Hynix, Samsung Electronics and Micron Technology.
- Chinese chipmakers have increasingly relied on grey-market HBM supplies after tighter U.S. export controls.
- Reuters reported that grey-market HBM can cost Chinese buyers several times more than prices paid outside China.
- Iluvatar CoreX has reportedly doubled GPU shipments to ByteDance to 100,000 units in 2026.
- The developments show that China’s AI-chip challenge involves memory, supply chains and production costs, not just processor design.
FAQ: China AI Chipmakers, Huawei and HBM
Why are China AI chipmakers raising AI-chip prices?
China AI chipmakers are raising prices largely because high-bandwidth memory has become more difficult and expensive to obtain. Export restrictions and reliance on grey-market supplies have increased procurement costs, and those costs are being reflected in AI accelerator prices.
How much does Huawei’s Ascend 950DT cost?
According to Reuters, Huawei has raised the indicated price of the Ascend 950DT accelerator card to more than 250,000 yuan, or about $37,255. The reported price is 20% to 50% higher than quotations offered two months earlier, depending on contract terms.
What is HBM and why is it important for AI chips?
High-bandwidth memory, or HBM, is high-speed memory that allows AI processors to access large quantities of data quickly. Because AI accelerators depend heavily on fast memory, HBM availability and pricing can have a major impact on the cost and performance of AI hardware.
Which companies dominate the advanced HBM market?
According to the Reuters report, the advanced HBM market is dominated by South Korea’s SK Hynix and Samsung Electronics and U.S.-based Micron Technology.
Is Cambricon also increasing its AI-chip prices?
Yes. Reuters reported that Cambricon has raised the indicated price of its planned 690 chip by approximately 20% to 30% compared with levels indicated two months earlier. The company has not formally launched the 690, and its current flagship is the 590.
What does the HBM shortage mean for China’s AI industry?
The shortage increases the cost of building AI computing capacity and creates a supply-chain bottleneck for domestic chipmakers. It also shows that reducing dependence on Nvidia requires progress across the entire AI hardware ecosystem, including processors, memory and manufacturing.
Final Takeaway
The latest price increases from China AI chipmakers reveal a central reality of the global AI race: the future of AI depends on memory and supply chains just as much as processors. China is expanding its domestic AI-chip ecosystem, but the HBM shortage shows how difficult it can be to build an alternative technology stack when access to critical components is constrained.
For more explainers on AI chips, semiconductor technology and the global AI industry, explore Kalinga.ai’s latest AI and technology coverage.