VENDORiQ: AI Dominance – A Shifting Landscape

Why it Matters:

The current announcements from NVIDIA occur within a competitive and geopolitically charged environment. While NVIDIA maintains market leadership in AI processing hardware, particularly with its CUDA ecosystem, the landscape is evolving and becoming a geopolitical battleground. 

Huawei, based in China, has developed Ascend AI chips, particularly the latest 910D, that compete with Nvidia’s H100 chips. While Huawei chips are not yet at the same performance level as Nvidia’s, they are significantly less costly, presenting a credible challenge to the US-based dominant provider. Huawei has also announced investments to double the production output, which will help alleviate the current global shortage of AI chips.

The AI competition from Huawei is not limited to chips. It covers an entire chip and photonics ecosystem, including: EUV-related lasers, Gallium-nitride chips, Photoresists (still dominated by Japan), Silicon carbide (SiC) wafers (with 32% global share taken in just a few years).

This competition is notable as the US government has advocated for a global ban on Huawei AI chips. The formal reason for the US DoC’s ban on Chinese AI chips worldwide is the claim that these chips were developed and made using American technologies. The US suggests implementing sanctions on any nation or firm using competitive chips, citing the AI Diffusion Rule¹, issued January 15, 2025, with compliance requirements that took effect on May 15, 2025. Given their recent introduction, these rules have not been tested in the US or international courts, nor put to arbitration.

This US governmental posture has not been accompanied by publicly available evidence supporting the US claims, contributing to a perception of geopolitical and protectionist motivation rather than purely technical justification.  

The irony here is that with increased restrictions from the US, its global technology competitors are forced to accelerate investments in sovereign alternatives, which positions them as much more compelling competitors on the global stage. For example, Huawei replaced 13,000 foreign-made parts and developed MetaERP, Harmony OS, and Gauss DB to survive sanctions. China also made several breakthroughs in its chip supply chain, such as LDP-based EUV development at Harbin Institute, which challenges ASML’s lithography dominance and shows China’s expansion beyond inference chips.

Furthermore, there is a growing decline in trust in US AI firms, stemming from concerns over the US administration’s actions regarding AI policy discussions on safety, intellectual property, and privacy.  This environment could potentially empower countries seeking alternatives to dominant US technology suppliers, including China, which may see an opportunity to advance its domestic chip industry and international market share. 

Recent history also suggests that the current US position has accelerated China’s efforts to develop its own AI and chip design and manufacturing, which looks set to accelerate. 

The effectiveness of NVIDIA’s pervasive software ecosystem, specifically CUDA-X, in retaining developers and fostering application development becomes critical in this evolving geopolitical situation. 

However, there are now three competing frameworks: CUDA, CANN, and OpenCL. CUDA remains the undisputed leader—it provides a deep, mature ecosystem with optimised support across major AI toolchains, making it a compelling moat for NVIDIA. However, CANN is improving under pressure in China, and OpenCL offers vendor-neutral flexibility that appeals to sovereignty-focused markets. Framework fragmentation will shape developer loyalty and long-term platform control—it deserves dedicated analysis alongside hardware.

The long-term implications for market diversification and supply chain resilience are substantial. Australian firms and the government must evaluate the use of non-US AI chips and other advanced technologies more broadly. 

If the current trend continues, the US will likely lose its ‘soft power’ lead in global technology as nations migrate to lower-cost, stable suppliers. 

However, proceeding with Chinese AI technologies may trigger a US response. Exactly what that response will be is anyone’s guess and will likely change rapidly.

IBRS expects that the current protectionist trends around AI and related technologies will see a substantial rise in interest in sovereign AI. In the Australian context, this will likely involve adopting open source large language models (LLMs), machine learning models, and graph technologies, running as-a-service on locally cloud computing platforms. The chipsets these platforms use will be driven by economics: the potential cost impact of US retaliation for not using US chips, the costs of power consumption, and the cost and availability of the chips.

The big three hyperscale cloud providers will also promote ‘in-country, sovereign AI’ capabilities, though all of these players will opt for US chips.

Who’s Impacted?

• Chief Technology Officer (CTO): Evaluate the technical capabilities of competitive chip architectures and their integration with current AI development frameworks. Understand the cost-to-risk tradeoffs and be prepared to provide input to legal and procurement executives.
• Chief Procurement Officer (CPO): Navigate the growing risks of AI procurement, including cost benefits, long-term supply stability, and decisions to adhere to US demand when sourcing AI technologies. 
• Developers and AI Engineers: Directly impacted by the availability of development tools and software ecosystems (e.g., CUDA vs CANN vs OpenCL). Codebases and workflows must be adapted to new hardware platforms as geopolitical fragmentation increases. Portability, performance tuning, and framework compatibility become major bottlenecks when working across heterogeneous AI stacks.
• Chief Information Security Officer (CISO): Assess hardware-level security risks introduced by foreign chips and toolchains, including firmware integrity, patching procedures, and supply chain exposure.
• Chief Risk Officer (CRO): Model geopolitical and operational risks related to chip vendor changes, export restrictions, and retaliation scenarios. Evaluate indirect risks like public backlash, compliance violations, and vendor lock-in.
• Head of Infrastructure: Plan around hardware deployment constraints, including power draw, thermal footprint, form factor changes, and compatibility with cooling or racking systems, especially as non-traditional or sovereign chip options emerge.
• Legal and Compliance Officers: Interpret complex new regulations (e.g., AI Diffusion Rule) and IP-related restrictions on AI hardware. Understand legal exposure from adopting chips or lithography tools that may be contested by the U.S. or EU.

Next Steps

• Evaluate the long-term implications of geopolitical tensions on AI hardware supply chains and develop contingency plans for diversified sourcing. 
• Monitor public and private sector discussions regarding AI policy, including safety, intellectual property, and privacy, to anticipate future regulatory environments.
• Assess the strategic risks and opportunities associated with reliance on specific AI hardware ecosystems, considering both technical merit and geopolitical factors.

1. The ‘Framework for Artificial Intelligence Diffusion’, is a policy that aims to manage the global spread of advanced AI technology, particularly chips and model weights, to balance national security interests with promoting AI innovation. The rule, issued by the Department of Commerce in January 2025, establishes a tiered system for exporting these technologies based on national security risks, with countries categorised into three tiers. ↩︎