Broadcom, ARM, Huawei, Meta and other alternative AI silicon providers and interconnect technology initiatives

The 2026 AI hardware landscape is witnessing a strategic shift driven by diversification, geopolitical tensions, and technological innovation. Leading silicon providers beyond Nvidia—such as Broadcom, ARM, Huawei, and Meta—are expanding their AI accelerator portfolios and forging new supply agreements to challenge traditional dominance and foster a more resilient ecosystem.

Expansion of AI Accelerators and Supply Agreements

• Broadcom has secured key supply chains by locking in HBM memory and TSMC capacity through 2028, ensuring a stable supply of essential components for AI chips. This move not only strengthens Broadcom’s position but also directly challenges Nvidia’s supply chain dominance in high-performance AI hardware. CEO Jay Liu has publicly stated that copper interconnects will remain viable through 2028, emphasizing that traditional interconnect technologies continue to play a vital role despite emerging optical solutions.

• ARM, renowned for its high-margin licensing model, is quietly increasing its influence in the AI domain. Its architectures are increasingly integrated into AI chips designed for inference and training, contributing to a diversified vendor landscape that reduces reliance on a single dominant player.

• Huawei continues to demonstrate technological prowess with its Atlas 950 Super Pod, a large-scale AI system comprising over 8,192 chips. Recent testing of a 1nm chip using domestic EUV lithography signals a major breakthrough in China’s efforts to develop independent manufacturing capabilities, aiming to disrupt the global supply chain and reduce reliance on foreign technology.

• Meta has doubled down on its in-house AI hardware with the release of four new MTIA chips, designed specifically for inference workloads. These chips, announced to be launched on a six-month cadence, exemplify the company's strategy to reduce dependence on third-party hardware—primarily AMD GPUs—and optimize for large-model inference at scale.

Additionally, industry collaborations are actively shaping the future of AI hardware. Major players like AMD, Broadcom, Nvidia, along with hyperscalers such as Meta, Microsoft, and OpenAI, are working to standardize high-speed optical interconnects—aiming for data transfer speeds of up to 3.2 Tb/s. These efforts are crucial as co-packaged optics transition from prototypes to commercial deployment, promising reduced latency, lower power consumption, and higher bandwidth for large-scale AI clusters.

New Optical and Interconnect Standards

• Optical interconnects, particularly co-packaged optics, are emerging as a disruptive technology to traditional copper cabling. Ayar Labs, in partnership with Wiwynn, has introduced co-packaged optical solutions directly into rack-scale systems, enabling faster data transfer with lower latency and energy consumption. This is vital for scaling AI training to unprecedented levels.

• The debate between photonic interconnects and copper persists. While Broadcom’s CEO advocates for the continued viability of copper through 2028, industry insiders see co-packaged optics as a game-changer that will reshape interconnect standards in the near term.

China-Focused GPU Launches and Domestic Innovation

Amid export restrictions on advanced chips like Nvidia’s H200, Chinese firms are accelerating their domestic GPU development. Lisuan’s G100 series GPUs, fabricated at 6nm with up to 12GB VRAM, are designed to run modern games and AI workloads, with a scheduled launch in China on June 18. This aligns with China’s broader strategy to achieve semiconductor independence through full domestic lithography and self-reliant manufacturing.

Huawei’s testing of a 1nm chip using domestic EUV lithography exemplifies the progress China is making toward disrupting global supply chains and gaining technological sovereignty. The Chinese government’s five-year plan emphasizes reducing reliance on foreign technology in lithography, chip design, and resource procurement, further fueling domestic innovation.

Supply Chain and Geopolitical Dynamics

The ongoing U.S. export controls on high-end AI chips—requiring licenses for sales to China—have prompted Chinese firms to ramp up indigenous manufacturing and develop local resources. Companies like SMIC and Huawei are investing heavily in domestic EUV lithography and chip testing capabilities, aiming to climb the technological ladder.

Meanwhile, supply chain resilience remains a key focus for hyperscalers and chipmakers worldwide. TSMC continues to expand capacity at 3nm and 2nm nodes, with commitments through 2028. Chinese firms, supported by government funding, are working toward full lithography independence, with breakthroughs like Huawei’s 1nm chip signaling a potential disruption in global manufacturing dominance.

Conclusion

The AI hardware race in 2026 is characterized by vendor diversification, advances in interconnect technology, and geopolitical strategies aimed at self-reliance. Chinese firms’ push for domestic lithography and GPU development, combined with industry efforts to establish high-speed optical standards, are reshaping the landscape. While Nvidia remains a dominant force, its position is increasingly challenged by regional startups, alternative architectures, and disruptive innovations.

The industry’s focus on scalable interconnect solutions and local manufacturing will determine whether legacy giants can maintain their lead or if emerging players and technological breakthroughs will redefine the future of AI hardware. The coming years will be pivotal in shaping global semiconductor leadership, security, and economic power—with geopolitical ambitions and innovation at the core of this high-stakes competition.