Nvidia Corp. (NASDAQ:NVDA) faces a fresh challenge to one of its biggest competitive advantages after DeepSeek teamed up with Huawei Technologies to build open-source software for Huawei's Ascend AI chips.
The effort targets CUDA, the software platform used to program Nvidia's GPUs, which has helped keep millions of developers tied to the company's hardware even as rivals build competing processors.
DeepSeek Targets Nvidia's Software Advantage
DeepSeek is adapting and open-sourcing parts of its software stack for Huawei's Ascend processors, making it easier to run and optimize AI workloads without Nvidia hardware, Reuters reported Wednesday.
That includes TileLang, an open-source programming language DeepSeek says offers a 'simpler programming model' than CUDA.
DeepSeek and Huawei also built a 128-chip Ascend 950 "supernode" and optimized the software used to move data and workloads across those processors.
That matters because Nvidia's dominance extends well beyond its GPUs. Nvidia says more than 7.5 million developers use CUDA and its other software tools, a developer base that makes rival chips harder to adopt.
Nvidia Has Already Warned About the Risk
Nvidia said in its latest quarterly filing that U.S. export restrictions and Chinese rules have effectively shut it out of competing broadly in China's data-center compute market.
The company warned that its absence has helped rivals build "larger developer and customer ecosystems to challenge us worldwide."
In April 2025, Gregory Allen of the Center for Strategic and International Studies said in congressional testimony that DeepSeek's technical expertise and open-source community could help improve Huawei's CANN software ecosystem, its closest answer to CUDA.
Allen said shifting workloads away from CUDA would likely take years. Wednesday's announcement suggests that effort is moving forward.
CUDA Still Has a Huge Lead
TileLang is not a full replacement for CUDA, which has accumulated nearly two decades of libraries, developer tools and infrastructure.
The threat is therefore less about replacing Nvidia overnight and more about reducing the software barriers that make its GPUs difficult to leave.
Polymarket traders see little near-term danger to the wider AI trade, putting the chance of the AI bubble bursting by Dec. 31 at roughly 9%, with about $2.4 million traded on the contract.
The market requires at least three of six stress events to occur within the 90-day measurement window in its rules. Those include Nvidia shares falling 50% from their all-time high or H100 rental prices falling to $1 an hour for five consecutive days.
Meanwhile, demand for Nvidia GPU capacity remains strong enough that customers are signing multiyear contracts for older H100 chips.