For more than a decade, NVIDIA’s CUDA platform has been the backbone of modern artificial intelligence. It is not just a software framework but a deeply embedded ecosystem that powers how AI models are trained, deployed, and scaled. China now recognises that NVIDIA’s dominance does not come from hardware alone but from this tightly integrated ecosystem built over decades. As a result, Huawei is developing CANN as an alternative, and emerging models such as DeepSeek V4 may soon deploy on it.

CUDA enables developers to harness the parallel processing power of GPUs for complex computations. Unlike traditional CPUs that handle tasks sequentially, GPUs process thousands of operations simultaneously. This makes them ideal for AI workloads, particularly deep learning, which relies heavily on matrix computations. However, CUDA’s strength lies beyond raw performance. It integrates hardware, compilers, libraries, and developer tools into a unified ecosystem. Frameworks such as PyTorch and TensorFlow are deeply optimised for CUDA, making it the default platform for AI development globally. Over time, this has created a powerful lock in effect.

Developers, enterprises, and researchers are all deeply embedded with in the CUDA ecosystem. Geopolitical pressures, particularly restrictions on advanced GPU exports, have accelerated China’s push for technological independence. Huawei’s CANN platform is at the centre of this effort. Designed as a CUDA like environment, it provides libraries, compilers, and runtime systems tailored for AI workloads. Unlike CUDA , which evolve d as a general-purpose parallel computing platform, CANN is designed primarily for artificial intelligence. It is tightly integrated with Huawei’s Ascend chips, creating a vertically aligned stack.

The goal is clear: reduce dependence on foreign technology and build a self-sufficient AI ecosystem. Despite strong state support and rapid progress, CANN faces significant challenges. CUDA’s maturity is the result of years of development, optimisation, and community building. Replicating this depth will take time. One of the biggest barriers is the developer ecosystem. Millions of developers are trained in CUDA, and shifting them to a new platform requires retraining and rebuilding tools. In addition, performance gaps remain. While China’s chips are improving quickly, they still lag behind the most advanced GPUs in efficiency and scalability.

Another challenge is fragmentation. Multiple Chinese companies are developing their own chips and frameworks, which may lead to a less unified ecosystem compared to CUDA’s global standard. DeepSeek V4 represents a potential turning point. By focusing on algorithmic efficiency rather than brute force computing, it aims to reduce the cost of training and inference. Techniques such as mixture of experts and optimised architectures allow mo dels to achieve high performance with fewer computational resources. If DeepSeek V4 is successfully integrated with CANN and Huawei’s hardware, it could demonstrate that competitive AI systems can be built without relying on CUDA.

This would not only validate China’s approach but also reshape the economics of AI by making it more accessible. The emerging divide in AI infrastructure offers important lessons. First, control over the software ecosystem is as important as hardware leadership. Second, technological dependence can become a strategic vulnerability. Third, building an alternative ecosystem requires long term investment across multiple layers including hardware, software, and talent. For countries like India, the challenge is to balance participation in global ecosystems with the development of domestic capabilities. Investing in open standards, fostering developer communities, and supporting indigenous innovation will be critical.

China’s effort to build an alternative to CUDA is not just a technological project but a strategic shift in the global AI landscape. While CUDA remains the dominant standard today, the rise of CANN and innovations like DeepSeek V4 indicate that the future may be shaped by competing ecosystems rather than a single global platform. The battle for AI supremacy is no longer limited to models or data. It has moved to the foundational infrastructure that powers the entire ecosystem. And in this deeper contest, the stakes are far higher.

(The writer is director-Mrikal (AI/Data Center) and a young alumni member, Government Liaison Task Force, IIT Kharagpur. He tweets as @ipravinkaushal.)

India, Japan to intensify cooperation in AI to attain national goals

Bikini, beer, big opinions: Viral ‘MAGA nurse’ turns out to be an Indian med student running an AI money machine

Millions followed her patriotic posts and paid for exclusive content, believing she was a real American influencer. The truth revealed a carefully crafted AI persona by an Indian national.

Obscene images of BJP candidate appears on buses in Durgapur, MLA lodges plaint

Lakshman Ghorui, BJP candidate for the Durgapur West Assembly constituency and sitting MLA, has lodged a complaint with the Election Commission of India (ECI), alleging that morphed nude images of him have been pasted on buses using artificial intelligence (AI) technology in an attempt to defame him ahead of the polls.

You might be interested in

US-Israel-Iran war LIVE UPDATES: Iran hardliners tighten grip as US enforces naval blockade; Israel-Hezbollah clashes continue

US-Israel-Iran war LIVE UPDATES: Iran hardliners tighten grip as US enforces naval blockade; Israel-Hezbollah clashes continue

Talks or a workaround? US sends envoys to Islamabad; Iran says no meeting planned

Talks or a workaround? US sends envoys to Islamabad; Iran says no meeting planned

Big political jolt to AAP: Raghav Chadha joins BJP; announces two-thirds of AAP’s Rajya Sabha MPs merger with BJP

Big political jolt to AAP: Raghav Chadha joins BJP; announces two-thirds of AAP’s Rajya Sabha MPs merger with BJP

© The Statesman