Qingmao Intelligence plans to create a network of Ascend 950DT tokens for 10,000 cards by 2027
Read more
Pandaily
pandaily.com

Qingmao Intelligence plans to create a network of Ascend 950DT tokens for 10,000 cards by 2027

Qingmao Intelligence has presented a plan for joint operations to create an artificial intelligence inference network based on Huawei Ascend 950DT equipment. According to the plan, the network should include 10,000 accelerator cards by 2027. A longer-term goal, described in the November QbitAI report, envisions expanding this network to 100,000 cards within five years, covering various regions and chip platforms.

This initiative was announced on November 17th with the participation of Ascend ecosystem partners and other industry participants. The proposed 'Token Factory' represents an operational model for providing tokens generated by models, rather than an announcement about the operation of an existing 10,000-card facility.

Division of Responsibilities

The plan distributes tasks among four groups. Ascend ecosystem partners are responsible for supplying chips, supernodes, and technical support. Data center operators provide computing resources, power, and infrastructure. Model developers and traffic partners ensure demand and assist in assessing service quality. Qingmao takes charge of coordinating software optimization, planning, unified accounting, and daily operations.

The company states that service level agreements and overall metrics will define each party's contribution, operational duties, and profit-sharing mechanism. The goal is to provide a unified inference service on resources owned or managed by different participants, rather than simply summing up the total number of installed accelerators.

Qingmao points out that its previous experience includes collaborating with over 20 data centers, planning nearly 20,000 internal accelerator cards, and adapting more than 15 types of chips. Furthermore, the company reports deploying about 10 Ascend 910C supernodes. These figures relate to previously completed work on various platforms and should not be considered the completed capacity for the planned 950DT program.

The company views software as the connecting layer: the inference engine improves performance on individual hardware platforms, while the management software coordinates resources, monitoring, and token accounting across data centers. Acceptance criteria include effective throughput, first-token latency, output speed, accuracy consistency, availability, and cost per million effective tokens.

This roadmap differs from the inference optimization solution based on BAAI and Qingmao FlagOS, which was presented at the same event and concerned specific Ascend 910C and 910B workloads. The new approach focuses on how client capacity and demand will be organized on a larger scale.

It is important to note that achieving the 2027 goal and the five-year expansion remain company plans. The report does not confirm that the proposed 10,000-card 950DT pool is operational, nor does it conduct an independent audit of its future utilization, service quality, or economic viability of service provision.

Similar stories

China Mobile Hunan and Huawei recorded 1.86 Gbps speed in 5G-Advanced test field at 2.3 GHz frequency
Read more
pandaily.com

China Mobile Hunan and Huawei recorded 1.86 Gbps speed in 5G-Advanced test field at 2.3 GHz frequency

China Mobile Hunan, in collaboration with Huawei, announced the achievement of a peak downlink speed of 1.86 Gbps during a field test of 5G-Advanced technology. The testing was conducted in the city of Changsha using the aggregation of three carriers in the 2.3 GHz and 2.6 GHz frequency bands.

It is important to note that the obtained result pertains to a specific internal network test, not a nationwide deployment or an announcement of a new chip. In Chinese reports, the 2.3 GHz frequency band is referred to as the E-band, which has a bandwidth of 50 MHz and is combined with 160 MHz in the 2.6 GHz D-band. This local term should not be confused with backbone millimeter-wave E-band channels.

The value of 1.86 Gbps was achieved through the operation of a three-carrier network. When using only the carrier at 2.3 GHz, the project demonstrated downlink speeds of about 400 Mbps and uplink speeds of over 100 Mbps. These figures do not guarantee that an individual subscriber will achieve the same speed outside of measurement conditions.

The test was carried out on the internal distributed antenna network of Hunan Agricultural University, where high load from teaching, video streaming, and file transfers can create throughput issues. The combination of available spectrum aims to eliminate this bottleneck while utilizing existing internal coverage infrastructure.

Over the three days of the project, a 4% increase in downlink traffic and a 10% increase in uplink traffic were registered. Changes in traffic volume are not equivalent to a controlled comparison of individual download speeds, and the presented reports do not contain independent testing methodology confirming a broader performance increase.

China Mobile Hunan stated that its current network includes approximately 13,500 E-band modules that can gain these capabilities through a software update. This indicates potential for modernization rather than confirmation that all modules have already been converted or that similar equipment elsewhere supports the same path.

Another limitation is terminal availability: estimates suggest that about 10% of devices in the existing network support this feature. Actual access also depends on supported bands, carrier aggregation, local coverage, signal condition, and the number of simultaneous users. The presence of a 5G-A icon alone does not guarantee sustained peak performance.

The next planned step is to expand internal tests across provinces and assess wider activation. Thus, the published measurements confirm a local engineering demonstration, as practical coverage is still determined by network and device conditions.

Moore Threads announces MDC 2026 for December 29-30 in Beijing, unveiling anticipated S6000 cards and Lushan GPUs
Read more
pandaily.com

Moore Threads announces MDC 2026 for December 29-30 in Beijing, unveiling anticipated S6000 cards and Lushan GPUs

Moore Threads has announced its MUSA Developer Conference 2026 (MDC 2026) to be held in Beijing on December 29th and 30th. The company made this announcement on November 11th, stating that over ten new products, technologies, and solutions will be presented at the event.

Founder, Chairman, and CEO Zhang Jianzhong will deliver a keynote speech on the company's plans for full-function computing based on so-called full-function GPUs, as well as on the architecture roadmap.

Anticipated New Products

Although Moore Threads has not yet confirmed the launch list, reports cited by Sina Tech and Jiemian indicate that an AI computing card, the S6000, and the 'Lushan' series of graphics GPUs built on the next-generation 'Huagang' architecture are expected to appear on the market by the end of the year. The S6000 is the successor to the S5000 series.

Chinese media describes the S6000 as a key element in the company's plans to create clusters of 100,000 accelerator cards, but the specifications for the S6000 have not yet been published by the company. Furthermore, the company demonstrated the C128 and C256 supernodes, with more details expected at the conference.

Lushan GPU Specifications

Regarding Lushan, reports reiterate data first presented by Moore Threads when introducing the Huagang architecture last December: 16 times higher geometric processing performance, 64 times higher AI computing performance, and 50 times higher ray tracing performance, along with improved texture filling, atomic memory access, and increased memory capacity. The base level of these multipliers was not restated in recent reports.

Lushan integrates generative AI rendering, a rendering architecture called UNITE, and a new hardware ray tracing engine, and is designed for high-performance AAA gaming and professional graphics work.

Event Scale

According to Shanghai Securities News and the industry site ijiwei, this year's event is larger in scale. It will feature 26 thematic forums across 11 areas, including 'factories' for model training and token and agent production, edge intelligence, supernodes, and AI infrastructure. The exhibition area has increased from 1,000 to 3,000 square meters, and over 4,000 developers are expected to attend, compared to more than 2,000 at the previous event. The conference website is available at mdc.mthreads.com.

Information regarding pricing, delivery times, and independent tests for both product lines remains unpublished. The romanizations 'Lushan' and 'Huagang' are Chinese names; Moore Threads' official English styling for them, as well as the full name of the S6000 product, could not be confirmed.

Lenovo launches Wanquan hybrid token factory and selects Zhangjiakou for the first flagship site
Read more
pandaily.com

Lenovo launches Wanquan hybrid token factory and selects Zhangjiakou for the first flagship site

Lenovo introduced its proprietary Wanquan Hybrid Token Factory during an event held on November 10 in Zhangjiakou, Hebei Province. As part of this event, a cooperation agreement was signed with local authorities, which stipulates the creation of the first 'mother factory'—a term Lenovo uses to denote the main flagship facility of its network.

Lenovo defines artificial intelligence computing power as a factory whose product is tokens—text units that the model reads and generates. For Zhangjiakou, the company plans to build a heterogeneous computing cluster with a capacity of 15,000P and an intelligent computing center with a total capacity of about 100 MW by 2027. Although reports do not specify the numerical accuracy of the 'P' rating, Lenovo aims to achieve an annual output exceeding 2,000 quadrillion tokens, a peak speed of 200 million tokens per second, and support more than 100,000 concurrent users.

By 2030, the company intends to increase the metrics of supernode clusters, data centers, and annual token output by more than ten times. Lenovo also provided efficiency data: a fivefold increase in token generation speed, an 'output rate' above 99.9%, a 70% reduction in overall cost per million tokens, and energy consumption efficiency below 1.1. It should be noted that these figures are presented as company statements, as the reports lack a baseline or definition of the token output metric.

The selection of Zhangjiakou is due to its location within the national plan 'Data East, Computing West' for the Beijing-Tianjin-Hebei region, as well as the availability of wind and solar energy. Liu Jun, Lenovo Vice President and China President, described a three-layer 'hybrid power' model. The base layer represents Lenovo's AI infrastructure for computing. The middle layer is the token factory, which combines the construction and operation of the intelligent computing center with Lenovo's xCloud service. The top layer includes agent solutions and services that deliver tokens into customer workflows. Lenovo also offers agent-as-a-service and token-as-a-service packages.

Zhangjiakou is the first of a series of planned locations. Lenovo intends to create factories for transforming heavy industry in the northeast, smart manufacturing in northern China, education in the northwest, short drama series generation in central China, and advanced manufacturing in the east. The next target will be Changsha, where joint work with Mango TV will take place on a token factory for new media and an AI film center, aiming to reduce production costs by more than 50%, shorten cycles by more than 60%, achieve a monthly output of 1,200 hours of finished material, and attract over 10,000 content creators.

Information regarding clients, chip suppliers, energy agreements, and construction phases until 2027 has not been disclosed.

Popular