Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline. This would go beyond the September report of a planned 160,000 chip Ascend 950DT deployment, which focused on operating models rather than establishing that the order would train V5.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer

Create a landscape editorial hero image for this Studio Global article: What do reports from STAR Market Daily, DigiTimes, TrendForce citing Cailian Press, and Bloomberg say about DeepSeek’s unconfirmed plan to t. Article summary: The reports describe an unconfirmed move toward training DeepSeek V5 primarily—and reportedly natively—on Huawei Ascend 950 chips, rather than merely running an already-trained model on Ascend hardware, with a possible 2. Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Reports from TrendForce and DigiTimes describe a possible change in DeepSeek’s hardware plans: using Huawei Ascend 950 chips not only to run models, but as the primary compute for training its next-generation V5 model. The reported target is 2027, but neither the plan nor a V5 release date is confirmed. 1
12
5
TrendForce, citing Caiwen News and industry sources, says DeepSeek plans to use Ascend 950 as V5’s primary training compute, marking a move beyond the chips’ reported earlier role, which was mainly inference. DigiTimes also reports that the companies’ cooperation could expand from inference to model training. 1
12
Some coverage describes the proposed approach as “native” training on Ascend. That wording signals a reported change in the hardware used for training; it does not establish that DeepSeek has completed a full training run on those chips. The underlying claims rely on industry sources, and a secondary account says neither company has confirmed the plan. 1
5
The timeline is tentative. TrendForce reports that the project could materialize in 2027, while another summary of its reporting says V5 remains at an early selection stage, with specifications and capabilities undisclosed. That is not a firm public release date. 1
7
DeepSeek’s publicly reported software work provides context for the proposed shift. In September, the company said it was open-sourcing programming infrastructure for Huawei’s Ascend platform, including compute and communication libraries, and highlighted TileLang.
A separate report says DeepSeek and Huawei had carried out low-level adaptation work during the V4 phase. That detail comes from secondary reporting, however, and the available sources do not establish that DeepSeek’s Ascend software stack is at parity with its Nvidia infrastructure. 9
In September, Bloomberg reported that DeepSeek planned to deploy at least 160,000 Huawei accelerators at a data center in Inner Mongolia. That report concerned a planned deployment to operate models; it did not establish that the order was for V5 pre-training. Other coverage characterized the planned use as inference, with training remaining on Nvidia hardware at that time.
The newer V5 reports describe a proposed expansion into training. But they do not show that the September order was repurposed for that work, or that the Ascend 950 and Ascend 950DT references describe one confirmed, unified deployment plan. 1
12
The available reporting does not confirm V5’s final specifications, capabilities, or release date. Nor does it demonstrate that DeepSeek has successfully completed frontier-scale pre-training on Huawei chips. The careful reading is that a move toward Ascend-based training is being reported—not that a completed training run or a replacement for Nvidia-based training has been proven. 1
5
7
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline.
Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline. This would go beyond the September report of a planned 160,000 chip Ascend 950DT deployment, which focused on operating models rather than establishing that the order would train V5.
Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline. This would go beyond the September report of a planned 160,000 chip Ascend 950DT deployment, which focused on operating models rather than establishing that the order would train V5.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer

Create a landscape editorial hero image for this Studio Global article: What do reports from STAR Market Daily, DigiTimes, TrendForce citing Cailian Press, and Bloomberg say about DeepSeek’s unconfirmed plan to t. Article summary: The reports describe an unconfirmed move toward training DeepSeek V5 primarily—and reportedly natively—on Huawei Ascend 950 chips, rather than merely running an already-trained model on Ascend hardware, with a possible 2. Topic tags: general, general web, user generated, news. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Reports from TrendForce and DigiTimes describe a possible change in DeepSeek’s hardware plans: using Huawei Ascend 950 chips not only to run models, but as the primary compute for training its next-generation V5 model. The reported target is 2027, but neither the plan nor a V5 release date is confirmed. 1
12
5
TrendForce, citing Caiwen News and industry sources, says DeepSeek plans to use Ascend 950 as V5’s primary training compute, marking a move beyond the chips’ reported earlier role, which was mainly inference. DigiTimes also reports that the companies’ cooperation could expand from inference to model training. 1
12
Some coverage describes the proposed approach as “native” training on Ascend. That wording signals a reported change in the hardware used for training; it does not establish that DeepSeek has completed a full training run on those chips. The underlying claims rely on industry sources, and a secondary account says neither company has confirmed the plan. 1
5
The timeline is tentative. TrendForce reports that the project could materialize in 2027, while another summary of its reporting says V5 remains at an early selection stage, with specifications and capabilities undisclosed. That is not a firm public release date. 1
7
DeepSeek’s publicly reported software work provides context for the proposed shift. In September, the company said it was open-sourcing programming infrastructure for Huawei’s Ascend platform, including compute and communication libraries, and highlighted TileLang.
A separate report says DeepSeek and Huawei had carried out low-level adaptation work during the V4 phase. That detail comes from secondary reporting, however, and the available sources do not establish that DeepSeek’s Ascend software stack is at parity with its Nvidia infrastructure. 9
In September, Bloomberg reported that DeepSeek planned to deploy at least 160,000 Huawei accelerators at a data center in Inner Mongolia. That report concerned a planned deployment to operate models; it did not establish that the order was for V5 pre-training. Other coverage characterized the planned use as inference, with training remaining on Nvidia hardware at that time.
The newer V5 reports describe a proposed expansion into training. But they do not show that the September order was repurposed for that work, or that the Ascend 950 and Ascend 950DT references describe one confirmed, unified deployment plan. 1
12
The available reporting does not confirm V5’s final specifications, capabilities, or release date. Nor does it demonstrate that DeepSeek has successfully completed frontier-scale pre-training on Huawei chips. The careful reading is that a move toward Ascend-based training is being reported—not that a completed training run or a replacement for Nvidia-based training has been proven. 1
5
7
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline.
Reports say DeepSeek may train V5 primarily on Huawei Ascend 950 chips, with a possible 2027 timeline. This would go beyond the September report of a planned 160,000 chip Ascend 950DT deployment, which focused on operating models rather than establishing that the order would train V5.