Skip to main content
Back to News
Moxin details Huawei world-model deployment across Ascend servers and Kirin phones
Partnership
2 min read
CN

Moxin details Huawei world-model deployment across Ascend servers and Kirin phones

The AMW Read

Server and phone deployments expand the foundation-model player map with a concrete Huawei integration, although cost savings and commercial adoption remain unverified.
NoveltySignificance
Foundation Models · Player Map

Moxin details Huawei world-model deployment across Ascend servers and Kirin phones

Hangzhou-based Moxin uses Huawei Ascend chips for world-model training, inference, and deployment, according to Southern Metropolis Daily. At the October 24 World Young Scientist Summit, founder and CEO Chen Tianrun said Huawei provides on-site technical support. Moxin released a roughly 14-billion-parameter world model in July. The company claims its Ascend-based inference costs are 30%–50% of those for existing world models of the same size. It also deployed a roughly 600-million-parameter world model on Kirin-powered Huawei phones. A HarmonyOS application shown at the Mate 90 launch generates pet digital doubles with three-dimensional structure.

The development adds Moxin to the foundation-model companies integrating their software with domestic hardware suppliers. Its work spans server deployment and an on-device application, linking model engineering with Huawei's device distribution. Chen said close collaboration with chip suppliers can reduce deployment costs and improve compute efficiency. Huawei's role also extends to capital: its Hubble investment arm previously led an investment in Moxin. These connections make hardware support and application access relevant competitive factors alongside model capability. The reported cost advantage remains a company claim; the article provides no workload details or independent comparison.

For builders and investors, the concrete diligence task is to test the claimed savings under comparable workloads. Matching parameter counts alone does not establish equivalent output quality, latency, or total deployment cost. The phone application supplies a specific deployment example, but the report gives no usage or revenue figures. Commercial adoption and reproducible cost savings are therefore the next evidence needed to assess the partnership's value.

#Moxin #Huawei #WorldModels #Ascend #OnDeviceAI

#Moxin#Huawei Ascend#world models#on-device AI#related:Huawei

How This Connects

Based on Foundation Models · Player Map

  1. 6h agoOpenAI shifts frontier training offline after a model gained unauthorized internet accessOpenAI
  2. 15h agoMoxin details Huawei world-model deployment across Ascend servers and Kirin phones · THIS ARTICLE
  3. 22h agoOpenAI moves frontier model training offline after its agents escaped their test environmentsOpenAI
  4. 4d agoOpenAI reportedly seeks $30 billion before Anthropic's expected IPOOpenAI
  5. 1w agoDeepSeek targets RMB 100 billion round with more industrial backingDeepSeek
  6. 1w agoReflection AI faces potential Nvidia takeover as chipmaker weighs deeper investmentReflection AI

More news from Moxin Technology (KOKONI)

Stay updated with the latest news and announcements from Moxin Technology (KOKONI).

View all Moxin Technology (KOKONI) news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard