Back to News
ProductAI Understanding briefing

DeepSeek partners with Huawei to open‑source TileLang for Ascend chips

DeepSeek announced a partnership with Huawei to open‑source TileLang, a high‑level programming language for Huawei's Ascend AI chips, positioning it as a simpler alternative to Nvidia's CUDA and aiming to reduce reliance on Nvidia's ecosystem.

4 min readRead the linked source
Source-provided image accompanying DeepSeek partners with Huawei to open‑source TileLang for Ascend chips
Source referenceSource recorded
Publisher
economictimes.indiatimes.com
Source link
economictimes.indiatimes.comhttps://economictimes.indiatimes.com/tech/artificial-intelligence/deepseek-partners-with-huawei-to-develop-chip-programming-tools-reducing-reliance-on-nvidia/articleshow/134579562.cms
Source type
Linked source — primary-source status has not been established.
Also cited

Story last revised

ContextUnderstand this in 60 seconds

Start here

Key terms

Compute
The processing resources required to train and run models, often measured in FLOPS or GPU hours.
Test yourselfAI Models Explained Quiz

What changed since publication

  1. First published
  2. DeepSeek announced a partnership with Huawei to open‑source TileLang, a high‑level programming language for Ascend AI chips, positioning it as a simpler alternative to Nvidia's CUDA and aiming to build an independent GPU software ecosystem.

What happened

DeepSeek said it is collaborating with Huawei Technologies to develop and open‑source programming tools for Huawei’s Ascend AI chips. The joint effort includes releasing TileLang, a high‑level language designed to simplify GPU programming while preserving performance, along with associated and communication libraries. DeepSeek framed the move as a step toward an independent GPU software ecosystem that can rival Nvidia’s CUDA stack.

In a post on its official WeChat account, DeepSeek announced that it is open‑sourcing infrastructure for Huawei’s Ascend platform. The core of the release is TileLang, a high‑level programming language intended to be universal and easy to program while still extracting the full performance of the hardware.

DeepSeek described TileLang as offering a "simpler programming model" compared with Nvidia’s CUDA, aiming to improve development efficiency and simplify code logic for AI workloads. The release also includes related and communication libraries that developers can use to build and optimize AI models on Ascend chips.

The collaboration is framed as part of a broader effort to create an independent, self‑controlled GPU software ecosystem, reducing reliance on Nvidia’s ecosystem and fostering a more diversified AI hardware market.

Source details: economictimes.indiatimes.com ↗

Why it matters

The partnership signals a strategic push by Chinese AI firms to build a home‑grown software stack for AI hardware, potentially lowering dependence on Nvidia’s dominant CUDA ecosystem. By offering a language that claims to be easier to use yet performance‑competitive, TileLang could accelerate development of AI models on Ascend chips, making Huawei’s hardware more attractive to developers worldwide. If successful, the move may reshape the competitive dynamics of AI hardware and software, encouraging diversification of tooling and reducing the market power of a single vendor.

Reducing reliance on Nvidia’s CUDA stack is significant because Nvidia currently dominates the AI hardware tooling market, which can limit competition and increase costs for developers. An alternative like TileLang could lower barriers to entry for AI developers, especially those targeting Huawei’s Ascend chips.

If TileLang delivers on its promise of simplicity without sacrificing performance, it could accelerate the adoption of Ascend chips in data centers and edge devices, expanding the ecosystem of AI hardware beyond the Nvidia‑centric model.

The open‑source nature of the tools encourages community contributions and transparency, potentially leading to faster innovation and broader compatibility across different AI frameworks.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Model Parameter Size:8B Parameters
VRAM Required5.5 GBGPU memory footprint
Target HardwareMacBook / Single GPUDeployment tier
Privacy100% Air-GappedLocal device capability
Core takeaway: Small, quantized models (3B–8B) now run directly inside smartphones and laptops with complete data privacy, while mammoth 400B+ models remain the domain of datacenter clusters.
Interactive Concept Check+10 Points
AI Models Explained Quiz

In AI, what are a model's "parameters"?

What to watch next

Future updates on TileLang’s adoption rates, performance benchmarks against CUDA, and the availability of the open‑source libraries will indicate whether the initiative can gain traction. Watch for Huawei’s roadmap on Ascend chip releases, any developer outreach programs, and potential responses from Nvidia or other hardware vendors. Additionally, monitor whether other AI firms join the effort or create competing toolchains.

Performance benchmarks comparing TileLang‑based implementations with CUDA‑based equivalents will be crucial to assess real‑world benefits.

Developer uptake and community contributions to the open‑source repositories will indicate the health of the emerging ecosystem.

Huawei’s upcoming hardware releases and any announced pricing or licensing terms for the Ascend chips will affect the commercial viability of the new software stack.

Reactions from Nvidia and other hardware vendors could lead to competitive responses, such as new tooling or pricing adjustments.

Related guides & quizzes

AI Models ExplainedFuture of AITest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI model release tracker

Updates and corrections

This canonical story is updated in place when the developing event materially changes. Its URL and original publication date never change.

  • DeepSeek announced a partnership with Huawei to open‑source TileLang, a high‑level programming language for Ascend AI chips, positioning it as a simpler alternative to Nvidia's CUDA and aiming to build an independent GPU software ecosystem.
See the public corrections log
Found this useful?