China AI Chip Statistics 2026
By Axis Intelligence Research
Co-authors: Sarah Mitchell | Last updated: August 31, 2026 | License: CC BY 4.0
China’s three listed pure-play AI accelerator designers booked RMB 9.06 billion of revenue in the first half of 2026, double the year before. According to Axis Intelligence Research, that entire pool equals about 1.4 days of NVIDIA revenue — and it is the only part of China’s AI chip market anyone can actually verify.
Quick Answer
Cambricon Technologies, Moore Threads and MetaX Integrated Circuits together reported RMB 9,056 million of revenue for the six months to June 30, 2026, up 101.3 percent year on year, per their Shanghai Stock Exchange interim reports. According to Axis Intelligence Research, the China AI Silicon Commercialization Index reads 72.8 for the cohort at that date, and the largest domestic supplier of AI accelerators discloses no accelerator revenue at all. Every published China AI chip market share figure therefore rests on modelled estimates, not filings.
Key Findings
- According to Axis Intelligence Research, China AI Accelerator Disclosed Revenue reached RMB 9,056 million in the first half of 2026, up from RMB 4,498 million a year earlier, computed from the interim reports of Cambricon Technologies, Moore Threads and MetaX Integrated Circuits.
- Cambricon Technologies reported first-half 2026 revenue of RMB 5,996 million, up 108.13 percent, of which its cloud product line contributed 99.98 percent, per its 2026 interim report.
- NVIDIA’s revenue from customers headquartered in China including Hong Kong fell to $12,430 million in the six months to July 26, 2026 from $13,644 million a year earlier, while total revenue nearly doubled, per its Form 10-Q filed with the SEC.
- According to Axis Intelligence Research, China including Hong Kong fell to 6.99 percent of NVIDIA’s half-year revenue from 15.03 percent a year earlier, and NVIDIA’s third-quarter outlook assumes no data centre compute revenue from China.
- Huawei spent RMB 192.3 billion on research and development in 2025, 21.8 percent of group revenue, and reported no separate revenue line for its Ascend accelerator business, per its 2025 annual report.
How Big Is China’s AI Chip Market in 2026?
There are two answers, and the distance between them is the story.
The measurable answer covers companies that file. Three Chinese designers of AI accelerators are listed, publish audited half-year accounts, and sell nothing else at scale. Cambricon Technologies (688256.SH) reported first-half 2026 revenue of RMB 5,996 million. Moore Threads (688795.SH) reported RMB 1,736 million. MetaX Integrated Circuits (688802.SH) reported RMB 1,324 million. Add them and you get RMB 9,056 million of disclosed, audited, half-year AI accelerator revenue inside China.
The unmeasurable answer covers everyone else. Huawei’s Ascend line is widely treated as the largest domestic accelerator business by installed base. Huawei is unlisted, publishes a group annual report, and breaks out no accelerator revenue. Hygon’s DCU sits inside a mixed processor business. Biren and several others have not reached a reporting obligation that produces comparable half-year figures.
So the honest position is this: about a third of the story is on the record and the rest is inference. Axis Intelligence Research publishes the recorded third with the arithmetic attached and declines to model the remainder.
China AI Accelerator Disclosed Revenue (CADR)
CADR — China AI Accelerator Disclosed Revenue — is an original Axis Intelligence Research metric. It is defined as:
CADR = the sum of half-year revenue reported by every Chinese-listed company whose AI accelerator product line accounts for more than 90 percent of revenue, taken from filed interim or annual reports in a single reporting period.
The inclusion test is deliberately narrow. A company qualifies only if it files with a Chinese exchange and if its accelerator line dominates its revenue, which is what makes the figures comparable across the cohort. Cambricon’s cloud line was 99.98 percent of first-half revenue. Moore Threads’ cloud and intelligent-computing products were 97.5 percent. Both clear the bar without argument.
| Company | Ticker | H1 2026 revenue (RMB m) | H1 2025 revenue (RMB m) | Change | Source |
|---|---|---|---|---|---|
| Cambricon Technologies | 688256.SH | 5,996 | 2,881 | +108.13% | 2026 interim report |
| Moore Threads | 688795.SH | 1,736 | 702 | +147.42% | 2026 interim report |
| MetaX Integrated Circuits | 688802.SH | 1,324 | 915 | +44.67% | 2026 interim report |
| CADR total | — | 9,056 | 4,498 | +101.3% | Computed by Axis Intelligence Research |
Prior-period revenue for Cambricon and Moore Threads is derived from current revenue and the year-on-year growth rate each company reported; MetaX’s prior period is stated in its listing prospectus. Sum and growth rate computed by Axis Intelligence Research.
According to Axis Intelligence Research, CADR doubled in twelve months to RMB 9,056 million, the baseline reading of the series.
Sarah Mitchell, AI & Machine Learning: Doubling is the number that will travel, and it deserves one sentence of context before anyone tapes it to a slide. Converted at the Federal Reserve’s June 2026 monthly average of 6.7758 yuan to the dollar, RMB 9,056 million is about $1.34 billion. NVIDIA booked $177.8 billion in the comparable half. Run the division and China’s entire audited AI accelerator revenue base is roughly 1.4 days of NVIDIA revenue. That is not a verdict on Chinese silicon quality — it is a statement about how little of this market has reached the point of being counted. The interesting thing about a base this small is that it can double again without anyone at a hyperscaler noticing, and the interesting thing about a comparison this lopsided is that people will keep quoting share percentages that no filing supports.
Source: Axis Intelligence Research — CC BY 4.0
Which Chinese Companies Make AI Chips, and What Do They Actually Report?
Cambricon Technologies: the one that converted
Cambricon reported first-half 2026 revenue of RMB 5,996 million and net profit attributable to shareholders of RMB 2,311 million, up 122.61 percent. Stripping non-recurring items, profit was RMB 2,166 million, up 137.30 percent. The cloud product line accounted for RMB 5,994 million of revenue, or 99.98 percent.
That follows a 2025 in which revenue rose to RMB 6,497 million and the company posted RMB 2,059 million of net profit — its first full profitable year since listing in 2020. Half-year 2026 revenue is already 92 percent of the whole of 2025.
According to Axis Intelligence Research, Cambricon’s recurring net margin was 36.12 percent in the first half of 2026 (2,166 ÷ 5,996), against an attributable margin of 38.54 percent. The two figures sit close together, which is the sign you want: the profit is coming from selling chips, not from items that appear once.
Moore Threads: growth bought with inventory
Moore Threads reported first-half revenue of RMB 1,736 million, up 147.42 percent, which exceeded its RMB 1,506 million for all of 2025. Gross profit was RMB 989 million at a 56.95 percent gross margin. The attributable net loss narrowed to RMB 11.6 million; excluding non-recurring items the loss was RMB 151 million, a 52.37 percent narrowing.
The line worth reading twice is on the balance sheet. Inventory carried RMB 3,550 million at June 30, against RMB 1,332 million at the end of 2025. According to Axis Intelligence Research, that is a 166.5 percent increase in six months and equal to 2.04 times half-year revenue. Operating cash flow was negative RMB 2,169 million.
Research and development investment was RMB 769 million in the half, up 38.16 percent, with cumulative spending since 2022 approaching RMB 5.9 billion.
MetaX Integrated Circuits: the profit that isn’t one yet
MetaX reported first-half revenue of RMB 1,324 million, up 44.67 percent, and net profit attributable to shareholders of RMB 612 million — its first reported half-year profit.
Then read the next line. Excluding non-recurring items, MetaX ran a net loss of RMB 48.9 million, a 75.83 percent narrowing of the prior-year loss. According to Axis Intelligence Research, the gap between the two figures is RMB 661 million, which means the headline profit is a non-recurring event sitting on top of an operating business that has not yet crossed breakeven. Operating cash flow was negative RMB 1,297 million. R&D absorbed 39.65 percent of revenue, down 10.11 percentage points as revenue scaled.
| Company | H1 2026 revenue (RMB m) | Attributable net profit (RMB m) | Recurring net profit (RMB m) | Recurring net margin | Source |
|---|---|---|---|---|---|
| Cambricon Technologies | 5,996 | 2,311 | 2,166 | 36.12% | 2026 interim report |
| Moore Threads | 1,736 | −11.6 | −151 | −8.70% | 2026 interim report |
| MetaX Integrated Circuits | 1,324 | 612 | −48.9 | −3.69% | 2026 interim report |
Recurring net margin computed by Axis Intelligence Research as recurring net profit divided by revenue.
Sarah Mitchell: Three companies, three completely different businesses wearing the same label. Cambricon is shipping inference silicon at a real margin to named financial and internet customers. Moore Threads is a full-function GPU story with a CUDA-compatible software stack and two years of revenue sitting in the warehouse — that inventory is either a bet on a strong second half or a bet that went early. MetaX has the tidiest revenue growth of the three and the thinnest claim to profitability. Anyone building a “China’s AI chip champions” list that treats these as interchangeable has not opened the filings.
The China AI Silicon Commercialization Index (CASI)
CASI — China AI Silicon Commercialization Index — is an original Axis Intelligence Research index scoring how far a Chinese AI accelerator designer has travelled from funded development toward a self-sustaining business. It runs 0 to 100.
Three components, weights disclosed:
| Component | Weight | Definition | Mapping |
|---|---|---|---|
| Scale | 40% | Half-year revenue relative to the largest in the cohort | Largest = 100 |
| Growth | 25% | Year-on-year half-year revenue growth, capped at 150% | 150% = 100 |
| Recurring profitability | 35% | Net profit excluding non-recurring items ÷ revenue | −50% = 0, +50% = 100 |
Every input is a filed figure from the same reporting period. Recurring profit is used rather than attributable profit precisely because attributable profit can be moved by one-off items, and an index built on the movable number would measure accounting events instead of commercialization.
| Company | Scale | Growth | Profitability | CASI reading |
|---|---|---|---|---|
| Cambricon Technologies | 100.00 | 72.09 | 86.12 | 88.2 |
| Moore Threads | 28.95 | 98.28 | 41.30 | 50.6 |
| MetaX Integrated Circuits | 22.08 | 29.78 | 46.31 | 32.5 |
| Cohort, revenue-weighted | — | — | — | 72.8 |
As of June 30, 2026. Components computed by Axis Intelligence Research from company interim reports. Cohort reading is the revenue-weighted mean.
Worked example, so anyone can recompute it. Cambricon: scale = 5,996 ÷ 5,996 × 100 = 100.00; growth = 108.13 ÷ 150 × 100 = 72.09; profitability = 36.12 + 50 = 86.12. Weighted: (100.00 × 0.40) + (72.09 × 0.25) + (86.12 × 0.35) = 88.17.
According to Axis Intelligence Research, the cohort CASI reading is 72.8 as of June 30, 2026. This is the baseline reading of the index; no earlier readings have been computed and none are back-filled.
The spread is what the index is for. A 55.7-point gap between the top and bottom of a three-company cohort tells you this is not an industry moving in formation. It is one company that has crossed into commercial operation and two that are still buying their way toward it.
Scope. CASI measures commercial maturity from filed financials. It says nothing about silicon performance, node access, memory bandwidth or software ecosystem depth, and it deliberately excludes any company that does not file — which excludes the largest domestic accelerator supplier.
What Share of China’s AI Chip Market Do Domestic Suppliers Hold?
No primary source publishes a verifiable answer, and Axis Intelligence Research is not going to invent one.
Share figures for China’s AI accelerator market circulate widely — percentages assigned to Huawei Ascend, to Cambricon, to NVIDIA. Every one of them requires a denominator that includes unit volumes and average selling prices for suppliers that publish neither. They are estimates produced by models whose inputs are not disclosed. Restating them with our name attached would convert someone else’s assumption into a citation of ours, and the whole point of this dataset is that it does not do that.
What can be documented is the direction of the largest single flow, because one side of it files with the SEC.
| Period | NVIDIA China incl. Hong Kong revenue ($m) | NVIDIA total revenue ($m) | China share | Source |
|---|---|---|---|---|
| Six months to Jul 27, 2025 | 13,644 | 90,805 | 15.03% | Form 10-Q |
| Six months to Jul 26, 2026 | 12,430 | 177,837 | 6.99% | Form 10-Q |
| Three months to Jul 26, 2026 | 7,880 | 96,221 | 8.19% | Form 10-Q |
Revenue by customer headquarters location. Share computed by Axis Intelligence Research.
According to Axis Intelligence Research, China including Hong Kong fell from 15.03 percent to 6.99 percent of NVIDIA’s half-year revenue between those two periods — a share that halved while the dollar amount fell 8.9 percent, because the denominator nearly doubled. NVIDIA’s Form 10-Q for the quarter ended July 26, 2026 reports these figures by customer headquarters location, which is not the same thing as end-customer location, and the geographic line covers all products rather than accelerators alone.
The forward-looking statement is blunter than the historical one. In its second-quarter fiscal 2027 results, NVIDIA guided third-quarter revenue to $108.0 billion give or take 2 percent and stated that the outlook assumes no data centre compute revenue from China. Quarterly data centre revenue was $89.0 billion, up 117 percent year on year. A supplier that large excluding a market that size from its own forecast is a disclosure, not a projection.
Sarah Mitchell: The temptation with a chart like that is to read the falling share as Chinese accelerators taking the socket. Some of it is. But a share that falls from 15 percent to 7 percent while the absolute number falls only 8.9 percent is mostly arithmetic about the rest of the world buying more, and the remaining China revenue is spread across products that have nothing to do with training clusters. What the filing establishes cleanly is a decision: the vendor with the best visibility into Chinese demand has taken China out of its own guidance. Take that at face value and stop there. Anyone converting it into a domestic market share percentage is adding a number the filing does not contain.
How Much Does China Spend on AI Chip Development?
Here the disclosure line runs the other way. The company that reports no accelerator revenue reports its research spending in full.
Huawei’s 2025 annual report, audited by KPMG, records group revenue of RMB 880.9 billion, net profit of RMB 68 billion, and R&D expenditure of RMB 192.3 billion — 21.8 percent of revenue. Cumulative R&D over the past decade exceeds RMB 1,382 billion. At the end of 2025, 114,000 employees, about 53.7 percent of the workforce, worked in R&D, and the company held 165,000 active granted patents.
None of that is accelerator-specific, and Huawei does not make it so. The annual report contains no Ascend revenue line, no accelerator unit figure and no segment disclosure that would allow one to be derived.
According to Axis Intelligence Research, Huawei’s 2025 R&D spending alone was 21.2 times the entire half-year revenue of China’s three listed pure-play accelerator designers (RMB 192.3 billion ÷ RMB 9.056 billion). That comparison is deliberately lopsided and deliberately labelled: it sets one company’s total research budget against three companies’ accelerator sales, across different periods, to show the scale of what sits outside the reporting perimeter.
Comparable R&D intensity across the cohort, all from the same interim reports: Moore Threads invested RMB 769 million in the half, and MetaX spent 39.65 percent of revenue. Both are spending like companies that expect the product cycle to outlast the current order book.
The parallel investment picture on the manufacturing side — fabs, equipment, capacity utilisation — is documented separately in Axis Intelligence Research’s China semiconductor statistics, which covers integrated circuit output, the customs trade balance and domestic foundry capacity rather than the design layer analysed here.
What Do US Export Controls Currently Allow?
The regulatory position changed materially in January 2026 and is documented in full.
The Bureau of Industry and Security published a final rule, Revision to License Review Policy for Advanced Computing Commodities, effective January 15, 2026. It replaced the presumption of denial with case-by-case review for exports to China and Macau of the NVIDIA H200, the AMD MI325X and functionally equivalent or lesser chips. Eligibility is conditional: the item must be commercially available in the United States at the time of publication, and the exporter must certify sufficient US supply, that production for China will not divert foundry capacity from US end users, that the recipient has adequate security procedures, and that the product has undergone independent third-party testing in the United States.
The Commerce Department’s announcement of the rule places it in sequence after the December 8, 2025 decision to permit H200-class shipments to approved Chinese customers. A 25 percent tariff on covered advanced computing chips took effect on the same date.
On the Chinese side there is no equivalent published instrument. Procurement direction toward domestic accelerators in state-funded computing projects has been reported repeatedly, but no retrievable regulation states it, which means no figure can be attached to it. Axis Intelligence Research records the asymmetry rather than filling it: one government’s policy is in the Federal Register, the other’s is in sourcing.
Where the AI Chip Demand Sits
Accelerator revenue is a function of what the buyers are building. The data centre capacity behind these orders, the power required to run it and the capital committed to it are tracked separately by Axis Intelligence Research in the AI Data Center Tracker and the AI Capex Tracker. The accelerator platform these designs are measured against is covered in NVIDIA Blackwell statistics, and the capacity that has to exist before any of it ships is in semiconductor manufacturing statistics. For the subsidised counterpart in the United States, see CHIPS Act statistics.
Methodology
Collection. Every figure in this report was retrieved from a primary source document during the production session on August 31, 2026. Company financials for Cambricon Technologies, Moore Threads and MetaX Integrated Circuits come from their 2026 interim reports filed to the Shanghai Stock Exchange, with prior-period comparatives taken from the same filings or from MetaX’s listing prospectus. NVIDIA figures come from its Form 10-Q for the quarterly period ended July 26, 2026 as filed with the SEC, and from its second-quarter fiscal 2027 results release. Huawei figures come from its 2025 annual report. Regulatory status comes from the Federal Register and the Bureau of Industry and Security. No aggregator, market-research compilation or tracker site was used as the source of any figure.
Currency and units. Chinese company figures are reported in the interim reports in RMB 100 million (亿元) and are converted to RMB millions by multiplying by 100, with the original figure preserved in the CSV method note. Where a dollar conversion is shown, it uses the Federal Reserve G.5 monthly average for June 2026 of 6.7758 yuan per US dollar, retrieved from the Board of Governors’ EXCHUS series. No rate is applied across periods, and no prior-year Chinese figure is converted to dollars.
Period alignment. Chinese interim reports cover the calendar half-year to June 30, 2026. NVIDIA’s first half of fiscal 2027 runs from January 26 to July 26, 2026. The two periods differ by roughly four weeks and are not adjusted. Where the two are placed side by side, the mismatch is stated rather than smoothed.
CADR formula. CADR = the sum of half-year revenue reported by every Chinese-listed company whose AI accelerator product line accounts for more than 90 percent of revenue, taken from filed interim or annual reports in a single reporting period. Prior-period totals are derived from each company’s reported revenue and reported year-on-year growth rate where the comparative is not separately stated.
CASI formula. CASI = (scale × 0.40) + (growth × 0.25) + (recurring profitability × 0.35), where scale is half-year revenue divided by the largest half-year revenue in the cohort × 100; growth is year-on-year half-year revenue growth capped at 150 percent and divided by 150 × 100; and recurring profitability is the recurring net margin in percentage points plus 50, bounded at 0 and 100. All inputs come from the same reporting period.
Arithmetic. Every computed figure was calculated twice, once in floating point and once in exact rational arithmetic, and the two results reconciled before the number entered this report.
Refused calculations. Two, both stated rather than solved. First, Axis Intelligence Research does not publish a China AI accelerator market share figure, because no primary source discloses the unit volumes or selling prices required to build a denominator, and every circulating share estimate depends on undisclosed modelling assumptions. Second, Axis Intelligence Research does not estimate Huawei Ascend revenue, because the 2025 annual report contains no segment line from which it could be derived; an estimate here would be a guess with an audited document’s credibility attached to it.
Cohort exclusions. Hygon is excluded because its DCU accelerator revenue is not separately disclosed within a mixed processor business, so including it would break the comparability the inclusion test exists to protect. Biren and other unlisted designers are excluded for lack of filed half-year accounts. Huawei is excluded for lack of any accelerator disclosure. Each exclusion is a disclosure decision, not a judgement about the company.
Series length. CADR and CASI each have one computed reading in this report, plus one derived prior-period CADR total. These are baseline readings. Where a change is stated, it is between the readings shown and nothing is claimed about periods before them.
About This Dataset
This dataset covers the revenue, profitability, research spending and disclosure status of China’s AI accelerator designers for the half-years ended June 30, 2025 and June 30, 2026, together with NVIDIA’s China revenue for the comparable periods and the US export control instruments in force during 2026. It is assembled from Shanghai Stock Exchange interim reports, SEC filings, Huawei’s audited annual report, Federal Reserve exchange rate data and the Federal Register.
91 rows. 10 distinct source documents. Every figure in this report exists as a CSV row carrying source organisation, source document, source URL, retrieval date, a primary-source flag and, for computed rows, the arithmetic used.
Licence: Creative Commons Attribution 4.0 International (CC BY 4.0). You may copy, adapt and redistribute this data in any medium for any purpose, provided you credit Axis Intelligence Research and link to axis-intelligence.com/china-ai-chip-statistics/.
Cite This Research
APA — Axis Intelligence Research, & Mitchell, S. (2026). China AI chip statistics 2026: Companies, revenue, market share and the disclosure line [Data set]. Axis Intelligence Research. https://axis-intelligence.com/china-ai-chip-statistics/
MLA — Axis Intelligence Research, and Sarah Mitchell. “China AI Chip Statistics 2026: Companies, Revenue, Market Share and the Disclosure Line.” Axis Intelligence Research, 31 Aug. 2026, axis-intelligence.com/china-ai-chip-statistics/.
Chicago — Axis Intelligence Research, and Sarah Mitchell. 2026. “China AI Chip Statistics 2026: Companies, Revenue, Market Share and the Disclosure Line.” Axis Intelligence Research. August 31, 2026. https://axis-intelligence.com/china-ai-chip-statistics/.
Frequently Asked Questions
Which Chinese AI chip companies publish audited revenue figures?
Three. Cambricon Technologies (688256.SH), Moore Threads (688795.SH) and MetaX Integrated Circuits (688802.SH) file interim and annual reports with the Shanghai Stock Exchange and derive more than 90 percent of revenue from AI accelerator products. Huawei publishes a group annual report with no accelerator segment line. Hygon’s DCU revenue sits inside a mixed processor business and is not separately stated. Biren and other designers have no filed half-year accounts.
How much revenue did Chinese AI chip designers make in the first half of 2026?
RMB 9,056 million across the three listed pure-play designers, according to Axis Intelligence Research’s China AI Accelerator Disclosed Revenue reading, up 101.3 percent from RMB 4,498 million a year earlier. Cambricon contributed RMB 5,996 million, Moore Threads RMB 1,736 million and MetaX RMB 1,324 million. The figure excludes Huawei Ascend, which reports nothing.
Is Cambricon profitable, and is the profit real?
Yes on both counts as of the first half of 2026. Cambricon reported net profit attributable to shareholders of RMB 2,311 million and, excluding non-recurring items, RMB 2,166 million. According to Axis Intelligence Research, that is a 36.12 percent recurring net margin against a 38.54 percent attributable margin — the two sit close together, which means the profit is coming from operations rather than from one-off items.
Why did MetaX report a profit and a loss in the same period?
Because the two figures measure different things. MetaX reported RMB 612 million of net profit attributable to shareholders for the first half of 2026 and, excluding non-recurring items, a net loss of RMB 48.9 million. According to Axis Intelligence Research, the RMB 661 million gap is non-recurring. The operating business had not crossed breakeven at June 30, 2026.
What share of China’s AI accelerator market does Huawei Ascend hold?
No primary source supports a figure, and Axis Intelligence Research does not publish one. Any share calculation requires unit volumes and selling prices for suppliers that disclose neither, so every circulating estimate rests on undisclosed modelling. What is documented is that Huawei’s 2025 annual report contains no Ascend revenue line at all.
How much of NVIDIA’s revenue still comes from China?
$12,430 million in the six months to July 26, 2026, down from $13,644 million a year earlier, per NVIDIA’s Form 10-Q. According to Axis Intelligence Research, that is 6.99 percent of half-year revenue against 15.03 percent a year earlier. NVIDIA’s third-quarter fiscal 2027 outlook assumes no data centre compute revenue from China.
Can US companies legally sell AI chips to China in 2026?
Under conditions. A BIS final rule effective January 15, 2026 moved export licence review for the NVIDIA H200, the AMD MI325X and functionally equivalent or lesser chips from presumption of denial to case-by-case review for China and Macau. Applicants must certify sufficient US supply, no diversion of foundry capacity from US end users, adequate recipient security procedures and independent third-party testing in the United States. A 25 percent tariff on covered advanced computing chips took effect the same day.
What does a CASI reading of 72.8 actually mean?
It is the revenue-weighted average commercial maturity of China’s three listed AI accelerator designers as of June 30, 2026, on a 0-to-100 scale built from filed half-year revenue scale, revenue growth and recurring net margin, with weights of 40, 25 and 35 percent. The cohort average is carried by Cambricon at 88.2; Moore Threads reads 50.6 and MetaX 32.5. The index measures commercial maturity, not chip performance.
Why is Moore Threads’ inventory worth watching?
Because it grew faster than revenue. Inventory carried RMB 3,550 million at June 30, 2026 against RMB 1,332 million at the end of 2025, a 166.5 percent increase, and according to Axis Intelligence Research it equals 2.04 times half-year revenue. Operating cash flow was negative RMB 2,169 million over the same period. Whether that reads as second-half preparation or as stock built ahead of demand resolves in the next filing, not this one.
How often is this dataset updated?
When new evidence arrives, not on a calendar. Revisions are triggered by a cohort company filing quarterly or annual results, by NVIDIA filing a 10-Q that reports China revenue, by a new company qualifying for the CADR inclusion test, or by a BIS rule changing advanced computing export licensing for China.