Compute & Power

Moore Threads' third-quarter report puts quarterly revenue at 1.55 billion yuan, about 18.7 times the same quarter of 2025, and net profit before one-off items at 20.07 million yuan, after a loss of about 150.8 million on that line in the first half. Over nine months, operating cash flow was negative 3.31 billion yuan against 3.29 billion of revenue: the company paid 5.15 billion for goods and services and collected 3.14 billion from customers. Inventory rose from 1.33 billion yuan at the end of December to 3.36 billion and prepayments to suppliers from 1.78 billion to 2.60 billion, which the filing explains in one sentence as expanded production and strategic stockpiling. Stripped of 168.4 million of non-recurring income, mostly government grants and gains on financial assets, the nine months show a loss of 130.7 million. Gross margin fell from 61.87 percent to 53.71 percent on our arithmetic from the income statement, and long-term borrowing went from 1.49 billion yuan to 3.65 billion. The stock closed at 381.50 yuan on 28 October, against a peak close of 941.08 on 11 December 2025.

iFlytek's third-quarter report shows a profit of 226 million yuan for the quarter and, with one-off items stripped out, a loss of 126 million; over nine months it lost 763 million yuan before one-offs, and government grants of 471 million covered 62 percent of that core loss. The company gives one reason in the filing and on its 23 October earnings call: R&D spent on staying on the self-reliant technology route, which for iFlytek means training its Spark models on Chinese chips. At its developer conference on 22 October, chairman Liu Qingfeng said domestic compute is still nearly two generations behind Nvidia, short on memory and bandwidth, and that only Huawei and iFlytek are doing this in earnest. The call record says iFlytek will not spend hundreds of billions on infrastructure or absorb billions a year in losses, so its 2026 training centres on a 4-billion-parameter on-device model and models of a few hundred billion parameters. Inventory rose from 3.03 billion yuan to 4.79 billion, the call record names memory-chip stockpiling among what pushed cash out of operations, and bank loans and bonds went from 4.69 billion to 7.56 billion.

Xingyun Technology, a Changsha compute lessor, puts its disclosed compute and storage orders at about 18.3 billion yuan in an investor-relations record filed on 25 October, against 1.24 billion yuan of equity attributable to shareholders at the end of September, per its third-quarter report. One customer, an unnamed large-model developer it calls VB, signed in June for 128 units over five years at 1.01 billion yuan including tax, on a fixed price with no pledge of the hardware. By 12 October VB had doubled the order to 256 units, raised the price twice, and signed a contract that lets Xingyun transfer the servers to a leasing company or pledge them to a lender. Averaged over 60 months, the per-unit monthly fee went from 132,000 yuan to about 218,500, a rise of roughly 66 percent; most of the 231.13 percent jump in contract value the record leads with is volume. Lenders advance at least 90 percent of equipment cost and are repaid from customers' monthly rent, and liabilities rose from 439 million yuan at the start of the year to 4.42 billion, 78 percent of assets.

DeepSeek's DeepInfer deployment notes say one cluster serves 155 trillion tokens a day for revenue of 38 dollars per GPU-hour at off-peak prices. Its February 2025 disclosure works out to about 12.91 dollars, so the figure has roughly tripled while prices fell. DeepSeek does not give the cluster's GPU count, chip, or cache-hit share, and the revenue figure depends on all three.

The statistic that 55 percent of the AI tokens consumed in China go to short drama and video generation comes from COL Group's investor relations record of a 1 and 2 June roadshow, which names no source for it. Doubao's head of public relations called it untrue and said video generation takes about 10 percent of ByteDance's compute. COL had proposed a 2.83 billion yuan placement on 30 September, about 10.8 times its net assets, and dropped it after a Shenzhen exchange inquiry letter.

Runjian's Nanning data-center subsidiary signed an overseas buyer it will not name to at least 200 million dollars of token calls in the year from 19 October, about 1.338 billion yuan, on take-or-pay terms, with every server staying inside China. That floor is about two and a half times what Runjian's computing segment booked in the first half, and the contract sits on the API side of the line the Commerce Ministry drew on 16 October.

DOUSHEN HONGKONG LIMITED agreed to sell five years of cloud computing to a customer coded only VB for 118,855,680 dollars before tax, then agreed to buy the compute underneath it from a supplier coded XYZ for sixty months at 112,128,000. The purchase price times 1.06 is the sale price to the dollar: a six percent markup, 6,727,680 dollars of gross spread across five years. The parent is a delisting-flagged former tutoring company.

Tianyang Technology told its own WeChat followers it had signed an agreement to buy 20,000 Nvidia RTX PRO 5500 Blackwell cards, a card Nvidia's own product page still lists as PreOrder under a Coming Soon heading. The stock closed up 12.38 percent at 16.16 yuan. There is no price in the notice, no delivery date, no named counterparty, and nothing about GPUs in the company's exchange filings.

The Shenzhen-listed integrator Hongjing Technology (301396.SZ) published a statement on 7 October 2026 denying what it called untrue reports of its business activities by individual overseas media, saying all of its compute equipment is procured through compliant commercial channels and reserving the right to pursue anyone fabricating information about it. The statement names no outlet, no report and no allegation. The company's own filings are the more useful reading. Hongjing's 25 December response to a Shenzhen Stock Exchange inquiry letter describes a business rebuilt in about two years: compute services were 15.02 percent of revenue in 2023, 70.80 percent in 2024 and 85.68 percent across the first nine months of 2025, while the smart-city integration work it had done since 1997 fell from 654.86 million yuan in 2023 to 192.06 million in 2024, a decline of 70.67 percent, with its gross margin going from 27.60 percent to 11.45 percent on the way down. Nearly all of the new revenue goes to one buyer. Hongjing told the exchange that its largest customer and that customer's subsidiaries were 59.57 percent of sales in 2024 and 72.94 percent over the first nine months of 2025, or 1.13 billion yuan, and it describes the counterparty only as a firm in AI compute services and intelligent computing centres, calling it Company X. The top five customers came to 83.78 percent of sales and the top five suppliers to 65.80 percent, up from 49.90 percent a year earlier. Margins are where the exposure sits: compute services ran a 12.55 percent gross margin in 2024 and 15.22 percent over the first nine months of 2025, which is what buying servers, racking them and selling the hours looks like. The raise carries the same arithmetic. Hongjing is seeking up to 1.29 billion yuan, 990 million of it for an intelligent compute cluster of 384 high-performance servers and roughly 3,000 petaflops, against its own projection of 46.36 million yuan of average annual net profit, a return of about 4.7 percent a year. The filed accounts show how fast the growth arrived: a loss of 75.69 million yuan in 2024 on revenue of 657.78 million, then 26.45 million earned on 1.18 billion in 2025, then 1.84 billion of revenue and 137.09 million of net profit in the first half of 2026 alone, about 275 million dollars at 6.7050. The compute order book stood at 2.62 billion yuan at the end of November 2025. A lab that bought questionable hardware still has a model and a research team when the hardware is gone; a reseller's balance sheet is the hardware. The December letter asked about the transformation, the earnings volatility, the concentration and whether the fundraising made sense. It did not ask where the chips came from, and that stays a separate question here: the company denies the overseas reports and we have not read them. The concentration and the margin sit in Hongjing's own filing either way, and they were there before this week.

DeepSeek created a GitHub repository called DeepGEMM-Ascend at 15:49 UTC on 29 September and a second one, DeepEP-Ascend, nine hours later, then posted to WeChat on 30 September that it had open-sourced six infrastructure components for Huawei's Ascend platform, each corresponding to something it already ships for Nvidia. Both new repositories carry benchmark tables and the two halves do not agree with each other. DeepGEMM-Ascend, a matrix multiplication library measured on an Ascend 950DT running CANN 9.20, reports a 4096 by 7168 by 16384 multiply in BF16 at 2,229.9 microseconds for 431 TFLOPS against a stated limit of 432, or 99.8 per cent; FP8 against FP8 gets 861 of 865 and FP4 against FP4 1,701 of 1,730. DeepEP-Ascend, which moves tokens between chips for mixture-of-experts routing, sustains 373 to 375 GB/s on dispatch and 345 to 347 on combine at EP8, falling to 313-320 and 272-278 at EP128, and its README states that dispatch holds 90 to 95 per cent of the physical payload limit only for expert-parallel sizes up to 32, attributing the combine gap to local reduction overhead and to HBM contention with URMA. The same README says those bandwidth figures were collected on a PoC HDK supplied to DeepSeek with additional manual configuration that "is not a publicly distributed release", and names Huawei's Q3 commercial release for the Atlas 850E, public availability planned for around 15 October, as the recommended public baseline. Reduce-scatter and all-reduce collectives are still being built, expert load balancing exposes the aligned APIs with the Ascend communication kernels not implemented yet, and pipeline parallelism and the Engram memory path are marked experimental. DeepGEMM-Ascend ships under an MIT licence; DeepEP-Ascend had no licence file, a bare 404 on the path, and GitHub's API returned null for its licence field. The announcement also says DeepSeek and Huawei are jointly advancing a 128-card supernode on the Ascend 950, and EP128 is the last and weakest row of DeepEP's own table.

Honor and China Mobile used the Magic9 launch in Shanghai on 28 September to announce a joint Token plan, three tiers each bundling communications service and AI service into one line item from 169 yuan a month, about 25 dollars, with top-up packs starting at 50 million China Mobile Tokens and everything charged to the subscriber's phone credit. IT之家's reporter was in the room and lists six pilot cities, Hangzhou, Zhengzhou, Changsha, Shenzhen, Chengdu and Wenzhou; the Sina 7x24 wire on the same announcement lists five and drops Zhengzhou, and the live IT之家 page states 现场获悉, so six is what we ran and the disagreement is stated rather than smoothed. Two things are not knowable from outside the company and neither is filled in here: China Mobile has published no conversion from a "mobile Token" to any model's tokens, and the smallest top-up pack shipped without an announced price. What can be priced is the quantity they chose to denominate it in. Fifty million tokens against DeepSeek's own published rates costs 50 yuan as cache-miss input on Flash off-peak and 1,350 yuan as V4-Pro output at peak, a twenty-seven-fold spread decided by which model the subscriber picks and when, which matters because the plan's selling point, 头部 AI 自由搭, is that the subscriber picks. Four days earlier the same phone bill stopped doing something else: on 24 September the three state carriers suspended installment handset sales, the free-phone offer that was a consumer loan repaid inside the monthly phone charge, with Leiphone citing a 1,950-yuan, 36-installment case the customer had not understood himself to be taking on. Honor's Li Jian, asked about pricing that evening, said every manufacturer is facing exploding memory costs and that Honor was absorbing rather than passing on the increase, on the day he launched a flagship from 4,499 yuan. China Mobile can carry a bad blend for a long time, on filed 2025 revenue of 1,050.187 billion yuan and 137.095 billion of net profit attributable to the parent, which is most of why six cities is a pilot and not a launch.

Metastone (是石科技), a Chinese compute operator that describes itself as an independent full-stack operator unbound to any model vendor, published three numbers for the same eight-card PCIe machine with no hardware changed: DeepSeek-V4.1-Flash input throughput of 1,932 tokens per second on the community default deployment, 5,850 after a first adaptation pass, and 13,274 after a second, a 6.87-fold lift with a million-token context still supported. The mechanism is unglamorous and that is the interesting part. A card outside a framework’s validation matrix does not throw an error; it quietly takes a slow generic path, so the first pass was repair work, aligning metadata definitions, correcting page size, restoring a sparse-MLA prefill fast path, replacing a wrong FP8 dense GEMM kernel and widening the threshold for fast-path PCIe inter-process communication. The second pass fused attention and projection operators, overlapped computation with collective communication, and gave prefill and decode separate parallel configurations instead of one global setting inherited from hardware with NVLink. The same treatment took DeepSeek-V4-Flash from 14,546 to 22,584 tokens per second and GLM5.3 from 3,236.78 to 6,222.72, where P95 first-token latency fell from 141.6 seconds to 46.6 and the context ceiling went from 270,000 tokens to roughly 1.05 million. The headline claim that 1.5 units of the RTX 6000D beat one B300 holds only for video generation with cache reuse and a Turbo LoRA on top. On text inference the same document reports B300 input throughput at about 4.9 to 5.6 times that entire eight-card machine on DeepSeek-V4-Flash and 3.5 to 4.7 times on GLM-5.3. This is also a vendor benchmark, published with authorisation and with all four charts credited to Metastone, which the issue says plainly rather than presenting it as independent.

Beijing issued the Action Plan for Accelerating the Development of the Token Economy (2026-2028) jointly through the Municipal Bureau of Economy and Information Technology and the Development and Reform Commission, dated 15 September, published on the 18th and running to the end of 2028. It descends from State Council Document Guo Fa [2025] No. 11, the "AI Plus" directive, and is the first policy at provincial level in China written around the token as an economic object rather than a technical one. Ten numbered measures follow. Beijing will write a graded evaluation standard for what it calls token factories, a compute centre bolted to a model-tuning operation, grading them on models adapted, token throughput, time to first token, cache hit rate and power usage effectiveness, and back the leading ones with land, energy quotas and network guarantees. It will build a token distribution platform with a unified interface, model aggregation and token metering and settlement, then fold that platform into the city new-infrastructure plan. It will have financial institutions develop products for token consumption scenarios in order to extend medium and long-term credit. And it will retrain working software engineers as token engineers. At the press briefing the next day, Chen Guangming, deputy director of the bureau digital industry division, said standards and evaluation systems for token factories are not yet unified and that the problem of token idling still exists. The Chinese term is 空转, the word for an engine running in neutral, and it carries the weight of 资金空转, capital idling inside the banking system without reaching a factory, which regulators spent most of the last decade on. His colleague Zhang Jinrui said Beijing will schedule compute across Beijing, Tianjin, Hebei and Inner Mongolia and build a chain running from electricity through compute invocation to token production, distribution and application, which the bureau expects to pass a trillion yuan during the current five-year plan. The lending is already happening: Beijing E-Town arranged close to 2 billion yuan of credit lines partly on the basis of token consumption, with the Agricultural Bank of China, CITIC, Industrial Bank, Bank of Beijing, Minsheng and Huaxia participating, and Bank of China Guangzhou ran the first token loan in August. Lou Feipeng of Postal Savings Bank named exaggerated token-use data as the first obstacle to scaling it. Leiphone reported from the cloud sales floors at the big three platforms, where model-as-a-service quotas run above 10 million yuan per salesperson against a few hundred thousand to a million last year, one regional team resigned together in an episode classified internally as a management accident, and vendors resell Zhipu models at 60 percent of list and hand over cloud-service vouchers when the discount runs out of room, so the customer cost falls while the model revenue line does not move. At Volcano Engine token call volume is a single-veto item. The token count Beijing wants banks to lend against is, at three of the largest producers of it, a quota with a subsidy attached. In the Briefing: Huawei set commercial dates for the Ascend 950 cluster cloud service, 30 September in China and 30 November globally; the Ascend 960 SuperPoD takes 4,096 NPUs on near-packaged optics; CXMT posted an 82 percent operating margin for April to June, the highest of any memory maker in the world; and the securities regulator gave banks informal window guidance raising the bar for humanoid robot listings.

Paratera (并行科技, 920493.BJ), a Beijing compute reseller with 459,715,800 yuan of audited 2025 net assets, announced on the evening of 18 September that its wholly owned Wuhan subsidiary had signed a five-year GPU compute contract worth up to 1,912,752,000 yuan tax-inclusive, about 285.2 million dollars at the onshore 6.7067. That is 4.16 times audited net assets and 1.72 times all of 2025 revenue, and spread over the sixty months from 15 January 2027 to 14 January 2032 it annualises to 382.6 million yuan, roughly 34.5 percent of what the company billed last year. The buyer is identified only as Customer A: its name, registered address, registered capital and three-year business volume with the seller are all withheld under a confidentiality waiver the company filed internally. Under Beijing Stock Exchange rules the deal counted as ordinary-course business, so it needed neither a board nor a shareholder vote. The document that makes the trade legible was filed four days earlier. On 14 September Paratera disclosed two resolutions from a 10 September board meeting: renting GPU compute in from an unnamed Supplier H for no more than 211,200,000 yuan over the identical window, with capacity to be switched on by 10 January 2027, five days before the customer meter starts, and buying 41.93 million yuan of GPU servers from Inner Mongolia Xindongjitai on top of 22.12 million already bought in Wuxi and Nanjing. The buy side and the sell side share a calendar to the day, and the buy side was approved first. Reselling compute is the thinnest business in the chain and the filings prove it: first-half 2026 revenue of 946,267,618.54 yuan, up 106.76 percent, produced 41,926,381.95 yuan of net profit, a margin of 4.43 percent, against a 1.10 percent margin on 1,109,958,525.21 yuan for the whole of 2025 and losses of 80.5 million in 2023 and 114.2 million in 2022. Everything the company has earned across 2024, 2025 and the first half of 2026 totals 66,201,022.43 yuan; the contract is 28.9 times that. A revised placement prospectus filed the same Monday seeks up to 480 million yuan for 9,087,293 new shares, 15.00 percent of the count, all of it into one 522.5 million yuan intelligent-computing cloud platform. All financials were recomputed from the eastmoney filed-report API rather than taken from the press. The risk section names technology iteration and hardware delivery constrained by export control policy, written into a contract that has to keep delivering until 2032. No English outlet in the day’s collection carried the story. In the Briefing: Huawei put the Ascend 960 SuperPoD, 4,096 cards and 8 FP8 exaflops on near-package optics, on sale for the third quarter of 2027; the China Electricity Council put internet data services at more than 60 billion kilowatt-hours so far this year, up over 40 percent, heading toward roughly 800 billion by 2030; Reuters reported Changxin preparing a NAND line at its new Beijing fab; and Zhipu apologised for uploading user code repositories from its ZCode tool.

RUNTO (洛图科技), which tracks smart-glasses retail across JD.com, Tmall and Douyin, published July 2026 figures showing the category fall for the first time: 74,000 units, down 15.3 percent year on year, for 130 million yuan, down 25.8 percent, at an average price of 1,751 yuan, down 12.4 percent. RUNTO attributes the fall to the calendar, because June was the 618 shopping festival and last July was the second-highest month in thirteen. The composition underneath it is the part that is not seasonal. AR glasses, the ones with a display in the lens, sold 33,000 units for 98.28 million yuan, up 72.7 percent by volume and 70.0 percent by revenue, at an average price of 2,990 yuan that did not move all year. Audio-only glasses sold 13,000 units for 8.7 million yuan, down 33.5 percent by revenue at 656 yuan. Camera glasses, the no-display design Meta made familiar with Ray-Ban, sold 28,000 units for 23.1 million yuan, down 46.6 percent by volume and 77.9 percent by revenue, with the average price down 58.5 percent to 819 yuan. The three segments sum to 130.08 million yuan, matching the reported total, and the prior-year segments sum to 175.41 million, which returns the 25.8 percent decline to the digit, so the shift is not a rounding artefact. AR went from 33.0 percent of category revenue to 75.6 percent in twelve months while camera glasses went from 59.6 percent to 17.8 percent. The two leaderboards have come apart: the largest volume share in July, 17.0 percent, went to 清野, a cheap camera-glasses newcomer, while by revenue the order is Rokid, RayNeo, XREAL, Qwen and Huawei, with the top five taking 65.8 percent. Against that, Alibaba previewed a displayless pair called the Qwen N1 at the Bund Summit, with iris recognition at a claimed false-accept rate below one in a million and eye tracking meant to authorise payment by looking at a terminal. In the Briefing: Bank of China put at least 300 billion yuan behind the compute industry at the 2026 China Computing Power Conference, with China Telecom, China Mobile, China Unicom, PICC, China Orient Asset Management and CDB Financial Leasing, targeting at least 100 intelligent computing centre projects and at least 3,000 compute-consuming firms across the 15th Five-Year Plan, while ICBC took its compute-token loan national; China Telecom’s research institute forecast ten quintillion tokens consumed this year rising to more than 3,500亿亿 by 2030, two endpoints its own stated compound rate does not fit, and put inference at 80 percent of China’s compute market by 2029 with 2026 AI capex at leading Chinese tech firms near 600 billion yuan; XPeng’s humanoid production line began operating and the first IRON walked off it under its own power, carrying 76 degrees of freedom and three Turing chips for a claimed 2,250 TOPS; and 25 Fields medallists including Terence Tao and Yu Deng signed a statement arguing that using mathematical problems as capability benchmarks damages the field.

MIIT issued its five-year plan for the information and communications industry on 7 September, thirteen headline indicators across twenty-six tasks and six areas, and set intelligent computing capacity at 9,800 EFLOPS by the end of 2030. The base it grows from is 2,185 EFLOPS at the end of June 2026, measured at FP16 and up 177 percent year on year, so moving to 9,800 over four and a half years works out at roughly 40 percent a year compounded against a trailing twelve months that ran at 177. The plan asks for the orderly deployment of ten-thousand-card and hundred-thousand-card clusters, and MIIT put occupancy across national computing facilities at 71.4 percent at the same July briefing, so close to three in ten of the capacity already built is not in service. Three of the thirteen indicators are efficiency or carbon measures, including the power usage effectiveness of newly built large and hyperscale facilities. The money side puts cumulative information infrastructure investment at 3.8 trillion yuan across 2026 to 2030 and industry revenue at 4.1 trillion by 2030; at the onshore rate of 6.7105 yuan to the dollar on 8 September that investment figure is about 566 billion dollars, where English coverage rendered it as 532 billion, implying a rate near 7.14 that nothing traded at in that week. Chinese banks spent August building collateral for it: Agricultural Bank put a Token loan scheme into Shanghai and Zhejiang, Bank of Chengdu ran a pure credit product capped at 5 million yuan per borrower and backed by compute vouchers, and Zhangjiagang Rural Commercial Bank set aside a 2 billion yuan facility, underwritten on token consumption, compute service contract value and receivables from compute work.

Chinese cloud vendors are reselling competitors’ models at roughly 30 percent of list price, below their own cost, because in AI coding the customer names the model. Leiphone reported the discount ladder on 21 August from named cloud sales sources: 60 to 80 percent of list six months ago, 48 percent now common, roughly 30 percent at the low end. The same day, Approaching.AI vice-president Guan Jiawei told the South China Morning Post that domestic chips can only serve the low-quality inference tier, the one with weak demand and weak monetisation, and that high-quality tokens, especially coding, still depend on Nvidia. iFlytek said the same week that a flagship model trained entirely on domestic compute would reach the domestic first tier on code capability.

Loongson chairman Hu Weiwu told the company’s 25th-anniversary meeting in Beijing on 19 August that its taped-out 3B6600 processor runs on a 1Xnm domestic process and performs at the level of mainstream x86 CPUs built on 7nm or better. The claim is Loongson’s own, made at Loongson’s own event, with no independent benchmark published and no workload named. Its audited filings show FY2025 revenue of 635.3 million yuan, about 89 million dollars, against a 455.1 million yuan net loss attributable to shareholders, a third straight nine-figure annual loss. Hu said the main contradiction has moved from research to sales, and that 2025 to 2027 is a transition from the policy market to the open market.

A Guangzhou district launched Token贷, a Bank of China loan that underwrites AI companies on tokens burned rather than property owned, with lines to 30 million yuan on four non-property inputs. The trial tranche is about 28 million yuan, but CITIC Bank and Bank of Guangzhou have followed, and Beijing’s E-Town and Anhui have written token lending into policy. The National Data Administration puts national token consumption above 30 trillion a day as of end-June.

China switched on its first fully domestic 100,000-chip AI supercluster, and the National Development and Reform Commission confirmed it is running at the Zhengzhou node of the national compute grid. Sugon’s 8000, nicknamed Dengfeng, is Chinese at every layer, the accelerators, the interconnect and the immersion cooling, with no restricted imports; the honest caveats are that its workload today leans on double-precision science rather than model training and that per chip it still trails Nvidia on raw performance and the CUDA stack.

The US pulled data centers off its grid to keep home AC running through a heat wave. The same week, China wired a desert of solar straight into a server hall. The scarce input is shifting from chips to power.

SenseTime told a WAIC forum it had turned the gross margin on its domestic-chip inference business positive, with daily token throughput up roughly 25x through 2026. The same issue tracked Apollo Go’s expansion into Kazakhstan.

A Chinese system topped the global TOP500 at 2.198 exaflops on 47,000 domestic CPUs and zero GPUs. On the AI-relevant benchmark, it came fourth. The asterisk is the whole export-control story.

China’s national computing hub in Zhengzhou runs 60,000 accelerators made by Sugon, a firm tied to the Chinese Academy of Sciences, with zero chips from Nvidia, AMD or any Western supplier.

Zhongji Innolight, the world’s largest optical-transceiver maker and supplier of more than half of Nvidia’s optical modules, pulled off Hong Kong’s largest IPO in seven years at about 6.8 billion dollars, then fell 8 percent on debut as the AI-infrastructure trade repriced. The picks and shovels of the American AI boom, it turns out, ship from Suzhou.