China’s homegrown AI chipmakers are raising prices across the board, and the reason has nothing to do with demand for their processors alone. A shortage of high-bandwidth memory (HBM), the specialized stacked memory that feeds every modern AI accelerator, has pushed component costs up so fast that Huawei, Cambricon, MetaX, and Iluvatar CoreX all repriced their product lines within the same few weeks in September 2026. The increases range from 20% to more than 50%, and they land at the exact moment Beijing is pushing its chip industry to cut its dependence on Nvidia.
The clearest number in the story belongs to Huawei. The company’s upcoming Ascend 950DT accelerator card is now quoted at more than 250,000 yuan, or roughly $37,255, according to Reuters. That is 20% to 50% higher than the price Huawei was giving customers just two months earlier. Cambricon raised the indicated price of its next-generation 690 processor by 20% to 30%. MetaX and Iluvatar CoreX also moved prices up, though the exact percentages for those two vendors have not been made public.
Timeline: How the September 2026 Price Hikes Unfolded
The repricing did not happen all at once. Reuters first reported Huawei’s higher Ascend 950DT quotes on September 10, 2026, framing it as an early signal of the broader HBM squeeze. A follow-up report a day later, on September 11, filled in Cambricon’s, MetaX’s, and Iluvatar CoreX’s parallel increases, making clear this was an industry-wide move rather than a single company adjusting its price list. Intel’s Lip-Bu Tan added his memory-constraint warning five days later, on September 16, tying the China-specific story to a global pattern that had already been building for months. That sequence, three separate disclosures inside a single week, is itself notable: pricing shifts of this size usually roll out over a full quarter, not a matter of days.
Why HBM Scarcity Is Driving Chinese AI Chip Prices Higher
HBM sits at the center of nearly every AI accelerator built since 2023. It stacks memory dies vertically and connects them to the processor through a silicon interposer, giving GPUs and NPUs the bandwidth they need to keep pace with massive AI models. The problem is that only a handful of companies, chiefly SK Hynix, Samsung, and Micron, can manufacture it at scale, and all three have redirected capacity toward the highest bidders in the Nvidia and AMD supply chains. That leaves Chinese chipmakers competing for whatever HBM output remains, much of it sourced through gray-market channels or domestic substitutes that lag the leading edge.
Intel CEO Lip-Bu Tan added his own warning to the mix on September 16, 2026, saying memory constraints could get worse before they get better. Tan’s comments echoed what shattered.io covered in Intel’s Tan: Memory Up 7x, No Relief Until 2028, where Intel described DRAM and NAND prices climbing five to sevenfold across parts of the memory market. The China-specific squeeze on HBM is a subset of that broader crunch, but it hits AI chipmakers harder because HBM has no substitute in accelerator design. A GPU or NPU without enough memory bandwidth simply cannot run large models at competitive speed, so vendors either pay the higher price or ship a slower product.
The knock-on effect is showing up in ordinary consumer memory too. Reports place a single-week jump in Chinese DRAM prices at 14%, as AI data center buildouts and HBM production absorb wafer capacity that used to go toward standard memory chips. shattered.io detailed a similar dynamic in RAM Prices Up 89%: AI Memory Crunch Hits Gaming and in Budget Phones, Laptops Lose 80% of Cost to Memory, both of which trace how AI’s memory appetite is reshaping pricing for products that have nothing to do with AI at all.
The Numbers: Who Raised Prices and By How Much
Four companies make up the core of this story, and each is responding to the same shortage in a slightly different way. Huawei’s Ascend line carries the highest public number. Cambricon’s 690 processor increase is smaller in percentage terms but still material for a chip aimed at large-scale training and inference. MetaX and Iluvatar CoreX have both confirmed price increases without disclosing exact figures, which itself signals how fluid the pricing environment has become; vendors appear reluctant to commit to a number that might be outdated within weeks.
| Company | Product | Price Change | New Price / Detail |
|---|---|---|---|
| Huawei | Ascend 950DT accelerator | +20% to +50% | More than 250,000 yuan (~$37,255) |
| Cambricon | 690 processor (next-gen) | +20% to +30% | Not publicly disclosed |
| MetaX | AI processor line | Confirmed increase, % undisclosed | Not publicly disclosed |
| Iluvatar CoreX | AI processor line | Confirmed increase, % undisclosed | Doubling ByteDance shipments to 100,000 units in 2026 |
Iluvatar CoreX’s shipment figure is worth pausing on. The company is reportedly on pace to ship 100,000 GPUs to ByteDance in 2026, double its prior volume, even as it raises prices. That combination, higher unit prices alongside rising shipment volume, is a signal that demand inside China for domestic AI silicon is strong enough to absorb cost increases rather than force customers toward alternatives. ByteDance, the parent company of TikTok, has been one of the most aggressive Chinese buyers of AI compute, and its willingness to keep scaling purchases from a domestic vendor despite higher prices says something about how constrained the alternative supply, meaning smuggled or gray-market Nvidia chips, has become.
Historical Context: How China’s AI Chip Push Got Here
China’s push toward domestic AI silicon accelerated sharply after US export controls tightened access to Nvidia’s top-tier accelerators. Huawei’s Ascend series, Cambricon’s processor line, and newer entrants like MetaX and Iluvatar CoreX all emerged or scaled up as direct responses to that squeeze. The goal was straightforward: build enough local capacity that Chinese cloud providers, social platforms, and state-backed AI labs would not depend on chips that Washington could cut off at any point.
That strategy worked well enough to create real demand, evidenced by ByteDance’s shipment numbers, but it ran straight into a second bottleneck that has nothing to do with export policy: the physical scarcity of HBM. Unlike logic chips, which China has made real progress fabricating domestically through SMIC and other foundries, advanced memory manufacturing remains dominated by SK Hynix, Samsung, and Micron, all companies headquartered outside China and all currently prioritizing shipments to Nvidia, AMD, and hyperscale cloud buyers in the US and elsewhere. China’s chip self-sufficiency drive solved the logic problem before it solved the memory problem, and September 2026’s price hikes are the clearest evidence yet of that gap.
This is not the first time HBM scarcity has rewritten pricing across the industry. shattered.io covered Nvidia’s own version of this squeeze in Nvidia Hikes AI Server Prices 15%+ on Memory Crunch, and the yield side of the story in HBM4 Memory Hits 80% Yield, Powers Nvidia Rubin. What’s different about the Chinese chipmakers’ situation is that they sit lower in the queue than Nvidia when memory suppliers allocate limited HBM output, which explains why their percentage increases are, in several cases, sharper than what Nvidia itself has passed on to customers.
Competitive Comparison: Chinese Accelerators vs. Nvidia and AMD
Even after the price increases, Huawei’s Ascend 950DT and Cambricon’s 690 remain positioned as lower-cost alternatives to Nvidia’s export-restricted data center GPUs inside China, though the gap has narrowed. A $37,255 price tag for the Ascend 950DT puts it in a similar range to some configurations of Nvidia’s China-specific compliant chips, eroding what used to be a clearer cost advantage for domestic silicon. The performance gap has not closed at the same pace as the price gap, which is the uncomfortable part of this story for Chinese buyers: they are paying closer to Nvidia-adjacent prices without getting Nvidia-adjacent performance.
| Vendor | Primary Market | HBM Supply Position | Recent Pricing Trend |
|---|---|---|---|
| Huawei (Ascend line) | China domestic AI training/inference | Constrained, competing with global buyers | +20% to +50% (Ascend 950DT) |
| Cambricon | China domestic AI training/inference | Constrained | +20% to +30% (690 processor) |
| MetaX / Iluvatar CoreX | China domestic, hyperscaler supply (ByteDance) | Constrained | Confirmed increases, undisclosed % |
| Nvidia (global, ex-China compliant SKUs) | Global hyperscalers, China-compliant exports | Priority allocation from SK Hynix/Samsung/Micron | +15%+ on AI server pricing, per shattered.io reporting |
| AMD (Instinct/Helios line) | Global hyperscalers | Priority allocation, large supply deals (e.g. Oracle) | Selective price hikes, Ryzen consumer line spared so far |
AMD’s approach offers a useful contrast. As shattered.io reported in AMD Preps 10% Price Hike, Spares Ryzen for Now, AMD has chosen to concentrate its price increases on data center and AI-facing products while shielding its consumer Ryzen line, at least for now. Chinese chipmakers don’t have that luxury in the same way, since their entire product lines are built around AI acceleration rather than a mix of consumer and enterprise silicon. When HBM costs rise, there is no lower-margin consumer segment to absorb the hit; the increase flows straight through to the AI chip price.
Market Impact: What Rising Chinese AI Chip Prices Mean for Buyers
For Chinese cloud providers and AI labs, the immediate effect is straightforward: training and inference costs go up. Companies like ByteDance, Alibaba, and Tencent have built AI roadmaps around a mix of restricted Nvidia imports and domestic silicon, betting that local chips would eventually close the cost gap even if the performance gap persisted. September’s price increases undercut half of that bet. If Ascend and Cambricon chips cost nearly as much as Nvidia-adjacent options while still trailing on raw performance, the calculus for large Chinese buyers gets more complicated, not less.
There’s a second-order effect worth watching too. Higher component costs for AI accelerators tend to ripple into cloud service pricing, since compute providers pass hardware costs on to customers renting GPU or NPU time. Chinese AI startups that rely on rented compute from domestic cloud vendors could see their own costs climb in the coming quarters, potentially slowing the pace of model releases from smaller labs that don’t have the balance sheet of a ByteDance or Alibaba.
Globally, the story adds pressure to an HBM market that was already tight. TrendForce and other supply chain analysts have tracked HBM allocation as one of the defining bottlenecks of 2026’s AI buildout, and a fourth major buyer segment, Chinese domestic accelerator makers, competing harder for the same limited output only tightens that market further. That has implications for Samsung, SK Hynix, and Micron’s pricing power, and for how quickly next-generation HBM4 capacity (see shattered.io’s coverage of Samsung Builds 8-Layer HBM4E for Nvidia at 18Gbps) can come online to relieve the squeeze.
How This Fits Into Beijing’s Chip Self-Sufficiency Strategy
Beijing has treated domestic AI chip capacity as a strategic priority for years, and nothing about the September price increases changes that policy direction. What it does change is the near-term economics facing the companies Beijing is counting on to execute that strategy. Huawei, Cambricon, MetaX, and Iluvatar CoreX are not hobbyist operations; they are the vendors China is relying on to keep its AI industry running without Nvidia. If HBM scarcity keeps pushing their prices toward parity with restricted foreign alternatives, the pressure to solve the memory bottleneck domestically will intensify.
That pressure is already visible in China’s push to scale up its own memory manufacturers, chiefly CXMT and YMTC, though neither has yet demonstrated HBM output at a scale that could meaningfully offset what SK Hynix, Samsung, and Micron currently supply to the rest of the world. Closing that gap is widely viewed as a multi-year undertaking, not something that resolves within a single product cycle. In the meantime, Chinese AI chipmakers are left absorbing global memory prices while selling into a market that expects domestic alternatives to be cheaper than Nvidia, not comparably priced.
DRAM Spillover: Why Ordinary Memory Prices Are Climbing Too
The HBM shortage doesn’t stay contained to AI accelerators. Wafer capacity is finite, and when memory manufacturers redirect production lines toward HBM, the conventional DRAM used in phones, laptops, and servers loses capacity too. That’s the mechanism behind the 14% single-week jump in Chinese DRAM pricing referenced in recent reporting, and it lines up with what shattered.io covered in Memory Chip Shortage: Stockpiles Fall Below 10 Days, where distributor inventory buffers across the memory supply chain had shrunk to levels that leave almost no cushion against a demand spike.
For device makers outside the AI chip conversation entirely, this is where the story becomes tangible. Smartphone and laptop makers sourcing standard DRAM are now competing with AI accelerator vendors for the same fabs. Apple’s own supply chain has felt this, as detailed in shattered.io’s report on iPhone 18 DRAM Shortage Strands $1B in Chips at TSMC. The Chinese AI chip price increases are one visible symptom of a supply chain squeeze that touches far more products than most buyers realize.
Risks If the HBM Shortage Drags Into 2027
The scenario Chinese chipmakers are trying to avoid is a prolonged squeeze that erases the price advantage domestic silicon was supposed to offer. If Ascend, Cambricon, MetaX, and Iluvatar CoreX products keep climbing toward Nvidia-adjacent pricing while still lagging on raw throughput, large buyers gain a real incentive to push harder for import workarounds or delay compute-heavy projects rather than commit to domestic hardware at a shrinking discount. That would undercut the core premise of Beijing’s chip self-sufficiency push, which has always depended on domestic silicon being cheap enough to offset any performance gap.
There’s also a financing risk for the chipmakers themselves. Huawei, Cambricon, MetaX, and Iluvatar CoreX have all built growth plans around expanding shipment volumes, evidenced by Iluvatar CoreX’s ByteDance ramp. If rising HBM costs force further price hikes and volume growth slows in response, that combination could squeeze margins at exactly the moment these companies need capital to invest in next-generation designs. A memory shortage that outside observers see as a pricing story is, from inside these companies, a solvency-adjacent risk if it persists long enough.
What Comes Next: Predictions
- Further price increases through Q4 2026. If HBM allocation stays as tight as current reporting suggests, expect Huawei, Cambricon, MetaX, and Iluvatar CoreX to announce additional increases before the end of the year rather than holding at September’s levels.
- Chinese cloud providers diversify supplier mix further. ByteDance’s decision to double Iluvatar CoreX shipments despite price hikes suggests large buyers will keep spreading orders across multiple domestic vendors to hedge against any single company’s pricing or supply problems.
- Beijing accelerates domestic HBM investment. Expect increased state-backed funding announcements aimed at CXMT, YMTC, or new entrants specifically targeting HBM production, even though meaningful output remains years away.
- Smaller Chinese AI labs face rising compute costs. Startups renting GPU/NPU time from domestic cloud vendors are likely to see their own bills climb as accelerator costs get passed through, potentially slowing release cadence for smaller players.
- Global HBM allocation stays a seller’s market into 2027. With Nvidia, AMD, and now a fourth major buyer segment in Chinese domestic chipmakers all competing for the same constrained HBM supply, SK Hynix, Samsung, and Micron are positioned to keep pricing power well into next year.
Why This Matters Beyond China’s Borders
It’s tempting to read this as a purely domestic Chinese story, but the mechanics apply everywhere AI accelerators get built. Every AI chip on the market today, whether it’s Nvidia’s Blackwell architecture, AMD’s Instinct line, or China’s Ascend and Cambricon families, depends on the same narrow pool of HBM suppliers. When one buyer segment absorbs more capacity, prices move for everyone downstream, including gamers buying graphics cards, cloud customers renting GPU instances, and phone buyers picking up a new device with pricier DRAM inside. The Chinese AI chip price hikes are a symptom of a global supply constraint, not an isolated regional event.
That’s also why the story is worth tracking even for readers who never touch a Chinese-made accelerator. Memory economics set a floor under hardware pricing across categories, and 2026 has made clear that AI demand alone is enough to push that floor higher, month over month, regardless of which country or company is doing the buying. Further reading on the supply chain mechanics behind these numbers is available from TrendForce, Tom’s Hardware, DigiTimes, Nikkei Asia, and The Register, all of which have tracked the HBM allocation squeeze through 2026.
Frequently Asked Questions
Why are Chinese AI chip prices rising in September 2026?
A shortage of high-bandwidth memory (HBM) has pushed production costs up for AI accelerators. Huawei, Cambricon, MetaX, and Iluvatar CoreX all raised prices within weeks of each other as HBM became harder and more expensive to source.
How much did Huawei raise the price of the Ascend 950DT?
Reuters reported the indicated price rose to more than 250,000 yuan, roughly $37,255, which is 20% to 50% higher than the quote Huawei was giving customers two months earlier.
What is HBM and why does it matter for AI chips?
High-bandwidth memory stacks memory dies vertically next to the processor, giving AI accelerators the data throughput they need to run large models efficiently. It has no practical substitute in current AI chip designs, which is why a shortage hits accelerator pricing directly rather than pushing vendors toward a cheaper alternative.
Which companies control the global HBM supply?
SK Hynix, Samsung, and Micron are the three manufacturers capable of producing HBM at scale, and all three have prioritized shipments to the largest global buyers, including Nvidia and AMD’s supply chains.
Is ByteDance still buying Chinese AI chips despite the price increases?
Yes. Iluvatar CoreX is reportedly on pace to double its GPU shipments to ByteDance, reaching 100,000 units in 2026, even as it raises prices, indicating strong underlying demand for domestic AI compute.
Does this affect memory prices outside of AI chips?
Yes. Because HBM production competes with standard DRAM for the same wafer capacity, the shortage has contributed to broader memory price increases, including a reported 14% single-week jump in Chinese DRAM pricing.
How does this compare to Nvidia and AMD’s pricing moves in 2026?
Nvidia has raised AI server pricing by more than 15% citing the same memory crunch, while AMD concentrated its price increases on data center and AI products while sparing its consumer Ryzen line. Chinese AI chipmakers don’t have a comparable consumer segment to absorb costs, so their increases apply across their entire accelerator lineup.
When might HBM supply loosen up for Chinese chipmakers?
Analysts don’t expect meaningful relief in the near term. Domestic Chinese memory manufacturers like CXMT and YMTC are investing in HBM capability, but reaching production scale comparable to SK Hynix, Samsung, or Micron is widely viewed as a multi-year effort.
