Huawei and DeepSeek release open source AI tools in a bid to lower Nvidia exposure — but will programmers make the switch? (www.techradar.com)

DeepSeek, in collaboration with Huawei, has open-sourced a suite of AI infrastructure components aimed at reducing reliance on Nvidia's platforms by facilitating the use of Huawei's Ascend accelerators. This release includes new repositories and updates to existing tools like TileKernels and DeepSelect, with a primary focus on niche applications due to their hardware limitations, which require the Ascend 950 series and specific connectivity options. Notably, while the new offerings are not groundbreaking in performance terms, they strive to ease the transition for developers accustomed to Nvidia's CUDA architecture by maintaining similar API structures. The significance of this move lies in the increasing constraints on Nvidia's chip exports amid a tense geopolitical landscape, positioning Huawei's Ascend as a potential alternative as developers search for practical solutions. Despite the advantages of familiar programming environments, the uptake of these tools faces significant hurdles—DeepGEMM-Ascend and DeepEP-Ascend attract considerably fewer user engagements on platforms like GitHub compared to their Nvidia counterparts, illustrating the formidable "switching costs" Nvidia's ecosystem represents. The reputation and usability challenges associated with Huawei's technologies further complicate the narrative, suggesting that while this initiative aims to broaden market choices, widespread adoption may remain elusive.