> ## Content Index > Fetch the complete content index at: https://www.sotkn.com/llms.txt > Use this file to discover other available public pages before exploring further. # 4,096 AI Cards in 30 Days, a Substation in 18 Months: Who Gets Paid While China’s New Halls Wait for Power - URL: https://www.sotkn.com/markets/china/china-mobile-ai-supernodes-substation-wait-november-2026/ - Published: 2026-11-11T08:53:06.000Z - Updated: 2026-11-12T10:44:04.000Z - Description: China Mobile plans over RMB100bn of compute spending in 2027, and a Dahua unit won 4,096 AI cards with 30-day delivery. Yet only 13.6% of AI compute projects by value are running, and a substation can take about 18 months. The grid link is stuck. - Author: Sean - Tags: #market-china, Markets, #layer-ai-data-centers, #layer-power-energy *China Mobile says it will spend more than RMB100 billion on computing power in 2027, and one of its suppliers has won 4,096 AI cards with delivery in 30 days. The servers arrive fast. The stuck step is the substation: industry sources say it can take about 18 months to approve, against four to five months to build the hall it feeds.* **CHINA PULSE** · AI DATA CENTRES · POWER · Data as at 11 Nov 2026 (China Mobile 6 Nov; Dahua filing 3 Nov; CINNO Q3) · Next update: Dahua contract signing or a named substation approval **Affects:** [AI Data Centers](https://www.sotkn.com/stack/ai-data-centers/) · [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/) · [Huawei](https://www.sotkn.com/stack/hyperscalers/what-does-huawei-do/) — Markets: [China](https://www.sotkn.com/markets/china/) · [Malaysia](https://www.sotkn.com/markets/malaysia/) · [US](https://www.sotkn.com/markets/us/) **In short:** On 6 November 2026 China Mobile chairman Chen Zhongyue said the company will invest more than RMB100 billion (about $15 billion) in computing power in 2027, inside a procurement plan expected to top RMB500 billion. Four days earlier, Dahua Technology said its subsidiary Zhejiang Huaqi had won 4,096 of the 20,480 AI cards in China Mobile’s supernode tender, worth about RMB2.63 billion, with delivery in 30 days. But CINNO Research counts only 13.6% of the RMB728.8 billion of third-quarter AI compute projects as delivered or running, and 71.6% as still planned. Industry sources told Shanghai Securities News that a hall takes four to five months to build while its substation approval takes about a year and a half. Orders are not the wait. Stuck step: the substation. Map of who gets paid — not a buy list. Education only, not advice to buy or sell any security. RMB is the Chinese yuan; China Daily converted RMB100 billion to $15 billion, and we use about RMB6.7 to the dollar for other sums. A megawatt (MW) is a thousand kilowatts, roughly what a large shopping mall draws at full tilt. Our own sums are labelled. This piece is about orders against grid connections. It does not re-tell our 25 October China Pulse on how fast compute capacity grew against power use ([compute up 146%, power use up 35%](https://www.sotkn.com/markets/china/china-ai-compute-146-percent-power-35-percent-october-2026/)). That one asked whether new racks were being used; this one asks whether the next ones can be plugged in. ### The numbers that matter this week China Mobile 2027 plan \>RMB100bn Computing power spend · \~$15bn · inside a >RMB500bn plan New Dahua unit award 4,096 cards \~RMB2.63bn incl. tax · 30-day delivery · no contract yet Won Projects delivered or running 13.6% Of \~RMB728.8bn in Q3 (CINNO) Thin Still planned 71.6% RMB521.7bn in planning or preparation (CINNO) Queue Hall build time 4–5 months Industry sources via Shanghai Securities News Fast Substation approval \~18 months Same sources · “compute waiting for power” Stuck step China Daily, 10 Nov 2026; Dahua filing 2026-079, 3 Nov 2026; CINNO via Zhitong Finance, 14 Oct 2026; Shanghai Securities News, 10 Nov 2026\. Status chips are Second Order’s read. **On this page:** [Why it matters](#why-it-matters) · [China Mobile’s plan](#the-plan) · [The Dahua order](#the-order) · [How much is running?](#pipeline) · [The substation clock](#grid) · [Our math](#our-math) · [Who gets paid](#money-goes) · [What is stuck?](#still-stuck) · [Dates next](#dates) · [FAQ](#faq) ## Why does a telecom’s shopping list matter for the AI buildout? Think of a restaurant chain that has just signed for 500 new ovens. The oven maker will deliver in a month. But each new kitchen also needs a gas connection and an electrical hookup from the city, and the city says that takes a year and a half. The ovens are not the wait. The hookup is. That is the shape of China’s AI buildout this week. China Mobile is the largest of the three state telecom operators, and TechNode reported that telecom operators took more than half of the big data centre tenders (RMB100 million or more) in the first seven months of the year. When China Mobile says what it will buy, it shows where a big share of the country’s AI hardware money is headed. For anyone mapping AI infrastructure, the point is simple. An AI data centre turns electricity into computing, so chips that cannot be switched on are just inventory ([what an AI data center is](https://www.sotkn.com/explained/what-is-an-ai-data-center/); [how much power one uses](https://www.sotkn.com/explained/how-much-power-does-an-ai-data-center-use/)). The question for the next year is not whether China will order the machines. It is how fast each order becomes a running hall. That is also why this piece is not about chips. Chinese chip supply, export controls and domestic accelerators matter, and we have covered them in earlier China Pulses. Here the hardware has been bought and is coming. We follow the step after the box arrives. ## What did China Mobile say it will spend? China Daily reported on 10 November that China Mobile will invest more than RMB100 billion, about $15 billion, in computing power in 2027\. That sits inside a broader procurement plan the company expects to exceed RMB500 billion next year. Chairman Chen Zhongyue called computing power the company’s largest incremental investment and said it aims to double its intelligent computing capacity for two consecutive years. Shanghai Securities News dates the remarks to 6 November. On the physical side, Chen said China Mobile broke ground on three gigawatt-level computing center parks this year and will build two more next year, bringing the total to seven. He said several 10,000-card clusters are already complete and the company is developing several ultra-large-scale ones. To see how big that jump is, take the first half of this year. Shanghai Securities News reported that China Mobile’s compute network capital spending in the first half of 2026 was RMB15.6 billion, up 71% from a year earlier, and that China Unicom and China Telecom’s compute investment rose by more than 80% and 97%. On our arithmetic, doubling that first-half figure gives about RMB31.2 billion a year, so RMB100 billion in 2027 would be roughly three times that pace. That is not a like-for-like comparison, since China Mobile has not given a 2026 full-year compute figure in the sources we read, and the 2027 number may include items counted differently. But it shows the order of magnitude. Chen also said the shift is from internet data centres to AI data centres, from central processors to graphics processors, and from cloud pools to what he called token factories. Shanghai Securities News added that China Mobile has made the token, the unit an AI model reads and writes, its fourth core business measure after voice, text messages and mobile data. Daily token use passed 140 trillion in March 2026, the same report said, up more than a thousand times from early 2024. Official policy points the same way. The National Development and Reform Commission (NDRC) says the 2026–30 plan period will bring RMB4 trillion in new direct investment in computing network construction, mostly from companies. The Ministry of Industry and Information Technology (MIIT) said China’s intelligent computing capacity reached 2,592 exaflops by the end of September, up 146% from a year earlier. We covered that figure and what it means for power use in an earlier Pulse ([compute up 146%, power use up 35%](https://www.sotkn.com/markets/china/china-ai-compute-146-percent-power-35-percent-october-2026/)). ## What exactly did the Dahua unit win? On 2 November Dahua Technology, a Shenzhen-listed camera and video-systems company, told investors that its subsidiary Zhejiang Huaqi Smart Technology had won a share of China Mobile’s 2026–2027 new-generation AI supernode equipment tender. Its formal announcement, number 2026-079, followed on 3 November. The filing gives the details. China Mobile’s tender covers 20,480 AI supernode cards, plus 448 RoCE network switches and 82 storage systems. Huaqi won a share of 4,096 cards, worth about RMB2.63 billion including tax, with a delivery time of 30 days. As of the filing, no formal contract had been signed, and Dahua warned that the award is not the revenue it will book in 2026. Two other points matter. First, China Mobile is a shareholder of more than 5% in Dahua, so the deal counts as a related-party transaction. Dahua’s board approved the bid in October with an upper limit of RMB2.05 billion; the actual award came in higher, which is why the board met again on 2 November. Dahua has asked the Shenzhen exchange to waive a shareholder vote. Second, Shanghai Securities News called it the largest AI supernode tender in China Mobile’s history, and said all five bidders pointed to domestic computing suppliers. The filing does not say which chips are in the cards, how much power each draws, or which halls will receive them. We do not guess at the first. For the other two, we show a simple, labelled scale in the next sections. ![Chart of the build clocks: server delivery 30 days, a compute hall four to five months, a substation approval about 18 months](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/china-mobile-ai-supernodes-substation-wait-november-2026-shift.png) The servers arrive in a month. The substation takes a year and a half. · Education only. ## How much of China’s AI compute build is actually running? Not much, by value. CINNO Research, a market research firm, counted 62 intelligent compute centre projects in the third quarter of 2026, with a combined investment of about RMB728.8 billion, up 34.4% from RMB542.4 billion a year earlier. Zhitong Finance carried the figures on 14 October. CINNO Q3 2026: where the RMB728.8bn sits Planning or preparationRMB521.7bn · 71.6% · up 99.1% on a year earlier Delivered or runningRMB99.4bn · 13.6% · up 30.5% on a year earlier Started or under constructionRMB91.7bn · 12.6% · up 3.2% on a year earlier Topped out or completeRMB16.0bn · 2.2% · down 86.1% on a year earlier Total, 62 projectsRMB728.8bn · up 34.4% on a year earlier CINNO Research via Zhitong Finance, 14 Oct 2026\. Figures as published; CINNO rounds to RMB0.1bn. Percentages are shares of project investment, not of capacity. Read that slowly. The pile waiting to be built has roughly doubled. The running pile grew about 30%. And the middle, projects under construction, barely moved, up 3.2%. CINNO’s own reading is that Q4 2026 to 2027 should bring a wave of topping-out and start-ups. TechNode, writing on 11 November, said the same: 2027 could be the major year for delivery of new computing infrastructure in China. A caution. These are shares of project money, not megawatts or cards, and a project counts as “delivered or running” by CINNO’s classification, which the summary does not spell out. The measure also covers only 62 projects, not every compute centre in the country. It does not show that the planned projects are late, only that most of the money has not yet reached the stage where it produces computing. Demand, meanwhile, is not the issue where capacity exists. Shanghai Securities News quoted the operator of a Wuxi compute centre, with 2,000 Moore Threads GPUs lit in June, saying cluster usage is close to 100% with about 60% of customers paying. At China Telecom’s Hohhot site, a manager said rack rental is already 100% and signed customers are queuing for new rooms to open. TechNode reported 15.56 million standard racks in operation in June, and 158 big tenders (RMB100 million or more each) worth over RMB113 billion in the first seven months of the year. So the gap between orders and running capacity is a supply-chain-after-the-box story. The same article also said general-purpose compute centres still run with about a 30% usage gap, so the squeeze is in AI halls in particular. Our earlier Pulse showed the other side: new racks were being added faster than work reached them ([compute up 146%, power use up 35%](https://www.sotkn.com/markets/china/china-ai-compute-146-percent-power-35-percent-october-2026/)). Today’s numbers say the next batch is queuing for something other than work. ## How long does a substation take against a data hall? Shanghai Securities News gave the clearest statement. It quoted industry sources saying a compute centre takes only four to five months to build, while approval of the substation that serves it takes about a year and a half. That mismatch, the paper said, produces what the industry calls “compute waiting for power.” The paper also said a data centre’s build cost splits roughly 1:5:30 between civil works, electrical and mechanical systems, and IT equipment, but once it runs, electricity is 50% to 60% of operating cost. These are industry figures, not government statistics, and the paper did not say how many halls are waiting or in which provinces. But the direction is plain: the IT gear is fast and the grid link is slow. The same article added a second problem. The way electricity is priced does not match how AI halls use it, and the system for tracing and trading green power is still being finished. On our arithmetic, a 30-day delivery against an 18-month substation means the servers are ready about 17 months before the grid link, if the two clocks start together. They do not always start together, of course. A hall can be planned around an existing substation. But when a new campus needs a new substation, the gap is the planning problem. We explain why the first power connection is the long pole in our Explained pieces ([how long it takes to build an AI data center](https://www.sotkn.com/explained/how-long-does-it-take-to-build-an-ai-data-center/); [how power gets from the grid to the GPU](https://www.sotkn.com/explained/data-center-power-distribution-explained/)). Beijing knows. The NDRC is setting up a coordination mechanism called “2+3+N”: the two state grid companies, the three telecom operators and N compute-related companies. The National Energy Administration published a list of 93 “AI plus energy” pilot projects on 5 November. The Ministry of Industry and Information Technology has run a campaign to link compute with power. China Daily reported that NDRC will focus on coordination among computing networks, new power grids and next-generation communications. These are plans and pilots. None of the sources we read gives a date by which a substation approval will take less than a year and a half. China Daily’s own commentary on 10 November makes a similar point in friendlier words: the buildout is “demand-driven, self-reliant and coordinated with the power grid,” it said, and then added that fast growth can invite overheating, so the country should build what is needed where it is needed, with the chips, power and users to make it work. ![Chart: where China's RMB728.8bn of third-quarter intelligent compute projects sit by stage, from 71.6% planning to 13.6% delivered or running, 12.6% under construction and 2.2% topped out](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/china-mobile-ai-supernodes-substation-wait-november-2026-print.png) Seven in ten yuan are still on paper. · Education only. ## What does the arithmetic say? Everything below is our arithmetic from the published figures. None of it is a company forecast. The first sums use only numbers in the Dahua filing. - Share of the tender: 4,096 of 20,480 cards is exactly one fifth, 20%. The other 16,384 cards went to other winners that the filing does not name. - Price per card, upper bound: RMB2.63bn divided by 4,096 is about RMB642,000 (about $96,000) including tax. That is an upper bound for the card alone, because the filing does not say how many of the 448 switches and 82 storage systems are in Dahua’s share. - If the whole tender were priced like Dahua’s share: five times RMB2.63bn is about RMB13bn, roughly 13% of China Mobile’s 2027 compute plan, or less if other winners were cheaper. This is an illustration, not a disclosed number. - The award against the board’s ceiling: RMB2.63bn against RMB2.05bn is about 28% higher. - Power scale, assumption labelled: the filing gives no power figure. If each card drew about 1.5 kW including its share of servers and networking, a purely illustrative figure, 20,480 cards would need about 31 MW and Dahua’s 4,096 cards about 6 MW. A different chip changes this a lot. - Clock gap: 18 months minus 1 month is about 17 months. Four to five months for a hall against 18 months for a substation means the grid link takes roughly 3.6 to 4.5 times as long. - Running share: RMB99.4bn of RMB728.8bn is 13.6%. The RMB521.7bn still in planning is about 5.2 times the RMB99.4bn already delivered or running. The point of the power line is only to show the scale of the connection problem: a 30 MW load is one large substation job, and China Mobile is building gigawatt-level parks. A gigawatt is a thousand megawatts. Three parks of that size this year and two more next year are the real test of whether the grid clock can be shortened. Build clocks and sums: China Mobile’s AI supernode tender Tender20,480 cards · 448 switches · 82 storage systems (Dahua filing) Dahua unit share4,096 cards · \~RMB2.63bn incl. tax · 30-day delivery ContractNot yet signed at 3 Nov (Dahua) Price per card, upper bound\~RMB642,000 (our arithmetic) Hall build4–5 months (industry sources) Substation approval\~18 months (industry sources) Delivered or running share13.6% of RMB728.8bn (CINNO) Dahua filing 2026-079, 3 Nov 2026; Shanghai Securities News, 10 Nov 2026; CINNO via Zhitong Finance, 14 Oct 2026\. Sums marked as ours are Second Order arithmetic. ## Who gets paid, and who waits? 1 · The card and server suppliers **Who:** Dahua unit: 4,096 cards, \~RMB2.63bn incl. tax; four other winners **Paid:** Paid on delivery once the contract is signed Paid 2 · China Mobile (the buyer) **Who:** \>RMB100bn compute spend in 2027; RMB15.6bn in H1 2026 **Paid:** Pays now, earns when the halls run Spends 3 · Hall builders and rack operators **Who:** Hohhot rack rental 100%; customers queue for new rooms **Paid:** Paid as halls open Next 4 · Grid and substation builders **Who:** Substation approval \~18 months vs 4–5 months for a hall **Paid:** The slow, regulated clock Next 5 · The grid connection **Who:** No count of waiting halls published; 93 pilot projects listed 5 Nov **Paid:** Substation approval and the first power-on Stuck step First, the suppliers who win the cards. The Dahua unit is paid once a contract is signed and the goods arrive; the filing says that has not happened yet, and a related-party rule applies. It is also worth saying that a camera maker winning a slice of a supernode tender is a sign of how crowded the field is for domestic AI servers. Shanghai Securities News said all five bidders pointed to the domestic ecosystem. Second, China Mobile. It is the one paying out RMB100 billion, so it is also the one who has to wait for the return. It is a telecom with a stable cash business, so it can fund ahead of the grid. The sources do not say how much of its 2027 plan is already tied to a named site with a power approval. Third, the hall and rack operators. If a hall opens, a customer is already queuing, according to the operators Shanghai Securities News quoted. These are the people paid once the lights go on, which is exactly why the grid link matters more to them than the chip supply. Fourth, the makers of grid equipment. Transformers, switchgear and cables are paid as substations are built, but the approval clock sets the pace. The same pattern shows up elsewhere. In Malaysia the wires are on track but the power plants are the harder part ([TNB needs 9,000 MW more](https://www.sotkn.com/markets/malaysia/malaysia-tnb-9000mw-power-plants-november-2026/)), and in the United States turbine makers are booked into the 2030s ([GE Vernova’s turbine slots](https://www.sotkn.com/markets/us/us-ge-vernova-q3-turbine-slots-november-2026/)). China’s version is the substation. Who waits. Anyone who has already bought cards and does not yet have a powered hall. Chinese cloud and AI companies short of capacity. And smaller buyers, who sit behind the big telecom tenders. The accelerator pool the Huawei ecosystem is building depends on the same halls ([Ascend 950DT’s 10,000-card pool](https://www.sotkn.com/markets/china/china-ascend-950dt-10000-card-pool-october-2026/); [what Huawei does](https://www.sotkn.com/stack/hyperscalers/what-does-huawei-do/)). ## What is stuck, exactly? The stuck step is the grid connection: the substation approval and the line that brings power to the hall. China Mobile has the money and an order book. The Dahua unit has a 30-day delivery time. What the sources describe as slow is the part controlled by the planning and approval system for power equipment, about a year and a half. It is not the money: RMB100 billion for 2027 is announced. It is not the machines: delivery is in a month. And it is not demand: operators report full racks and queues for new rooms. The sources do not say it is the chips either, though chip supply is a separate constraint we have covered elsewhere. The honest limit of this read is that we do not have a count. Shanghai Securities News said “compute waiting for power” happens “from time to time.” That is a description, not a statistic. The CINNO 13.6% share is a money share and does not say why the other 86.4% is not running: much of it is simply early. A planning-stage project is not late. So the stuck step is the thing that decides the timing of the projects that are under way, not an explanation of the whole 71.6%. This is the same shape we keep seeing across AI infrastructure: the party that owns the slow step does not get paid on time, while the parties around it are paid on their own clocks ([where the stuck step moves](https://www.sotkn.com/analysis/where-does-the-stuck-step-move/)). In the US this week the slow step was the site for generators ([Anthropic’s 1.7 GW of generators](https://www.sotkn.com/markets/us/us-anthropic-1-7gw-generators-site-november-2026/)). In China it is the substation. Contracted power is not always used power, and an order is not a running hall ([contracted power explained](https://www.sotkn.com/explained/contracted-power-explained/)). ## Which dates test the read? ![Calendar of dates that test the read: Dahua contract signing, China Mobile's two extra parks in 2027, NEA monthly power data, CINNO next quarter, and 2027 as the expected delivery year](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/11/china-mobile-ai-supernodes-substation-wait-november-2026-calendar.png) Dates that test the read. · Education only. - 2–3 November 2026 — Dahua board confirms the RMB2.63 billion award; no formal contract yet (Dahua filing). - 6 November 2026 — China Mobile’s 2027 compute plan, over RMB100 billion (China Daily, 10 Nov). - Date not set — formal contract between the Dahua unit and China Mobile; the first sign the order is firm. - About 20 November — monthly power-use figures for data services from the National Energy Administration; the date is our expectation from the monthly pattern, not confirmed. - Q4 2026 to 2027 — CINNO expects a wave of topping-out and start-ups; TechNode calls 2027 a delivery year. - 2027 — China Mobile builds two more gigawatt-level parks, taking the total to seven, and plans to double intelligent computing capacity again. - Any named substation approval or grid-connection date for a China Mobile park — the datapoint we are missing. Dates from Dahua’s filing (3 Nov 2026), China Daily (10 Nov), CINNO via Zhitong Finance (14 Oct) and TechNode (11 Nov). The 20 November item is our expectation. ## How does this compare with Malaysia and the US? In all three countries the order book is bigger than the running base, and the slow step sits outside the chip. Malaysia’s utility says it needs 9,000 MW more and the plants are the hard part ([TNB needs 9,000 MW more](https://www.sotkn.com/markets/malaysia/malaysia-tnb-9000mw-power-plants-november-2026/)). In the US, Anthropic paid for generators and still has to name sites ([Anthropic’s generator deal](https://www.sotkn.com/markets/us/us-anthropic-1-7gw-generators-site-november-2026/)). China differs in one way the sources point to: Beijing is already building a national coordination mechanism between the grid companies and the compute buyers, so a fix, if it comes, could be set centrally. That is a possibility, not a forecast. At the chip end, China’s mature-node factories are full, a different bottleneck ([Hua Hong’s specialty slots](https://www.sotkn.com/markets/china/china-hua-hong-q3-specialty-slots-november-2026/)). The two do not cancel each other. Full factories and full order books are good for the makers, and the grid link decides when the customer gets to switch the machines on. Second-order read. China Mobile’s over-RMB100bn plan and the Dahua unit’s 4,096-card award show that the money and the machines are lining up, with delivery in 30 days. The wait sits one step later: the substation, which industry sources put at about 18 months against four to five months for the hall. Suppliers are paid first, the buyer funds the gap, and the hall operators are paid when the power is on. Watch for the Dahua contract and the first named grid-connection date. **Get the Wire.** Second Order’s free daily brief on what moved in the AI chip map. [Read the Wire →](https://www.sotkn.com/the-wire/) · [Subscribe free →](https://www.sotkn.com/#/portal/signup) **How we check this.** China Mobile’s 2027 plan, the three-plus-two gigawatt-level parks and the national figures come from China Daily’s 10 November report of China Mobile’s 6 November remarks. The tender size, the 4,096-card share, the RMB2.63 billion, the 30-day delivery and the absence of a signed contract come from Dahua Technology’s own filing of 3 November. The split of projects by stage is CINNO Research’s third-quarter count, as carried by Zhitong Finance. The build clocks (four to five months for a hall, about 18 months for a substation) are what industry sources told Shanghai Securities News; they are not official statistics and we say so. Every sum we worked out is labelled “our arithmetic”. Where a source is silent, such as how many halls are waiting for power, we say that instead of guessing. Price targets and ratings are left out on purpose. The references we used are listed at the end. [Our editorial standards →](https://www.sotkn.com/editorial-standards/) ## FAQ **What did China Mobile announce?** On 6 November 2026 China Mobile chairman Chen Zhongyue said the company will invest more than RMB100 billion (about $15 billion) in computing power in 2027, as part of a procurement plan expected to exceed RMB500 billion, according to China Daily on 10 November. He said China Mobile broke ground on three gigawatt-level compute parks this year, will build two more next year for seven in total, and aims to double its intelligent computing capacity two years in a row. **What did the Dahua unit win?** In a filing dated 3 November 2026, Dahua Technology said its subsidiary Zhejiang Huaqi won 4,096 cards of China Mobile's tender for 20,480 AI supernode cards, with 448 switches and 82 storage sets, for about RMB2.63 billion including tax and 30-day delivery. No formal contract had been signed, and Dahua said the award is not the revenue it will book in 2026. **Is China's AI compute actually running?** Mostly not yet, by project value. CINNO Research counted 62 intelligent compute centre projects in the third quarter, worth about RMB728.8 billion, with 13.6% delivered or operating, 12.6% under construction, 2.2% topped out and 71.6% still planned or in preparation. Operators report full racks where capacity exists, so demand is not the gap. **What is the stuck step?** The grid connection. Industry sources told Shanghai Securities News that a compute centre can be built in four to five months, while approval of its substation takes about a year and a half, so halls sometimes wait for power. The sources did not give a count of halls waiting, so we cannot say how many are affected. **Is this investment advice?** No. Education only. It maps who gets paid and which step is stuck. It is not a recommendation to buy or sell any security, and we leave out price targets and ratings on purpose. --- ### References Outbound links below are nofollow. We list only the sources actually used. Internal sotkn.com links in the body are separate. 1. China Daily, China Mobile to invest big in computing power, 10 Nov 2026\. [chinadaily.com.cn](https://www.chinadaily.com.cn/a/202611/10/WS6af275a2e4b03ebf8519b75b.html?ref=sotkn.com) — China Mobile's plan to put more than RMB100bn ($15bn) into computing power in 2027 within a procurement plan above RMB500bn, three gigawatt-level parks started this year and two more next, the NDRC's RMB4 trillion estimate for 2026–30 and MIIT's 2,592 EFLOPS figure. 2. Shanghai Securities News via China Economic Net, 算力网建设须“协同发力” (computing network needs coordinated effort), 10 Nov 2026\. [ce.cn](http://www.ce.cn/xwzx/gnsz/gdxw/202611/t20261110%5F3289274.shtml?ref=sotkn.com) — China Mobile's first-half compute capex of RMB15.6bn (+71%), the Dahua award, industry sources putting a data hall at 4–5 months against about 18 months for substation approval, full rack leasing at Hohhot, and the NDRC “2+3+N” grid, telecom and compute coordination. 3. Dahua Technology (Shenzhen Stock Exchange 002236), Announcement 2026-079, bidding progress and related-party amount, 3 Nov 2026\. [static.cninfo.com.cn](http://static.cninfo.com.cn/finalpage/2026-11-03/1225653806.PDF?ref=sotkn.com) — the tender for 20,480 AI supernode cards, 448 switches and 82 storage sets; the subsidiary's 4,096-card share at about RMB2.63bn including tax; 30-day delivery; no formal contract yet; and the earlier RMB2.05bn estimate. 4. Zhitong Finance, CINNO Research: Q3 China intelligent compute centre investment about RMB728.8bn, 14 Oct 2026\. [zhitongcaijing.com](https://www.zhitongcaijing.com/content/detail/1507206.html?ref=sotkn.com) — CINNO's split of 62 projects by stage, with 71.6% planning, 13.6% delivered or running, 12.6% under construction and 2.2% topped out, and the year-on-year changes. 5. TechNode, China accelerates data center expansion as the West faces mounting opposition, 11 Nov 2026\. [technode.com](https://technode.com/2026/11/11/china-accelerates-data-center-expansion-as-the-west-faces-mounting-opposition/?ref=sotkn.com) — 15.56 million standard racks in June, 158 tenders of RMB100m or more worth over RMB113bn in seven months, telecom operators taking over half of them, and the view that 2027 will be the big delivery year. 6. China Daily, Slow and steady characterizes nation's rational AI buildout, 10 Nov 2026\. [chinadaily.com.cn](https://www.chinadaily.com.cn/a/202611/10/WS6af30654e4b03ebf8519b9e1.html?ref=sotkn.com) — the official line that the buildout is demand-driven and coordinated with the grid, plus its own caution that fast growth can invite overheating. Not investment advice. Do your own research. Education only. Map of who gets paid and which step is stuck — not a buy list. More on this desk: [China Markets](https://www.sotkn.com/markets/china/) · [AI Data Centers](https://www.sotkn.com/stack/ai-data-centers/) · [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/) · [all Markets](https://www.sotkn.com/markets/) · [Malaysia Markets](https://www.sotkn.com/markets/malaysia/) · [US Markets](https://www.sotkn.com/markets/us/).