AMD, Broadcom and Google Intensify Anti-Nvidia Offensive as AI Semiconductor Landscape Faces Potential Realignment

Cracks Emerging in AI Chip Monopoly
AMD Expands Supply Chain and Packaging Push, Broadcom Accelerates ASIC Offensive
Google and Blackstone Forge TPU Alliance

Signs of fragmentation are beginning to emerge in the artificial intelligence (AI) semiconductor market long dominated by Nvidia. As major players including AMD, Broadcom and Google accelerate efforts to penetrate the AI infrastructure sector through divergent strategies, pressure is mounting on the GPU-centric market order. AMD has committed massive capital toward advanced packaging and supply-chain expansion, while Broadcom is aggressively targeting hyperscale customers with customized AI semiconductors. Google, meanwhile, is mounting a direct challenge to Nvidia’s ecosystem through a vertically integrated strategy combining proprietary semiconductors with cloud infrastructure.

AMD Expands Investment in Taiwan’s AI Supply Chain

According to CNBC on the 25th, AMD announced in a statement on the 21st that it plans to invest more than $10 billion in expanding advanced packaging capacity and strengthening strategic partnerships for next-generation AI infrastructure. The company intends to collaborate with major Taiwanese semiconductor backend firms to advance technologies capable of dramatically improving power efficiency and bandwidth between semiconductor chips.

AMD will first work with Taiwanese semiconductor packaging companies ASE Technology and SPIL to develop and validate next-generation wafer-based 2.5D bridge interconnect technology (EFB) aimed at improving power efficiency in AI systems and processors. Together with PTI (Powertech Technology), the company has already completed mass-production validation of the industry’s first 2.5D panel-based EFB interconnect. AMD’s rack-scale AI platform “Helios,” equipped with the sixth-generation EPYC CPU codenamed “Venice” and the Instinct MI450X GPU, is scheduled for large-scale deployment beginning in the second half of this year. Major ODM partners including Sanmina, Wiwynn, Wistron and Inventec will support manufacturing of Helios-based systems.

The investment initiative was announced while AMD Chief Executive Officer Lisa Su was visiting Taiwan. Su arrived in Taiwan on the 20th and completed a series of local engagements through the 22nd. On the 20th, she held talks with TSMC Chairman C.C. Wei, and on the 21st she conducted closed-door meetings with executives from major Taiwanese suppliers. Speaking to reporters outside a hotel on the evening of the 21st, Su stated, “TSMC is a great partner for us,” adding that “we plan to continue expanding production capacity in both Taiwan and Arizona.” She further emphasized that “we are always expanding our business in Taiwan,” noting that “a significant number of AMD engineers are based here, and all AMD product R&D is conducted in Taiwan.”

Broadcom Accelerates Customized AI Chip Strategy

AMD’s large-scale investment announcement aligns with a broader industry trend in which multiple semiconductor companies are intensifying efforts to challenge Nvidia. Broadcom, the U.S. semiconductor and infrastructure software company, has also rapidly expanded its custom AI semiconductor (ASIC) business in recent months, placing growing pressure on Nvidia’s GPU-centered market structure.

Broadcom has deliberately avoided direct confrontation with Nvidia in the general-purpose GPU market. Instead, the company is concentrating on co-designing AI semiconductors with hyperscale customers including Google, Meta and OpenAI. By supplying chips optimized for specific AI services and workloads, Broadcom aims to maximize power efficiency and reduce operational costs.

Market dynamics are also shifting visibly. As AI data-center construction costs surge, big tech companies are becoming increasingly sensitive to long-term operating expenses rather than simply securing GPU supply. In the expanding AI inference market, power efficiency and thermal management capabilities are emerging as variables just as critical as raw computing performance. Broadcom is targeting precisely this transition point. With strengths spanning high-speed networking semiconductors, interconnect technologies and advanced packaging capabilities, the company is viewed as being well positioned to offer supply strategies encompassing entire AI data-center architectures.

Broadcom’s growing influence on Wall Street has also become increasingly evident. In the first quarter of fiscal 2026, the company’s AI revenue reached $8.4 billion, up 106% year-over-year, while second-quarter AI semiconductor revenue is projected to climb to $10.7 billion. In a report released last month, Goldman Sachs identified both Broadcom and Nvidia as core beneficiaries of the AI computing market. Broadcom, in particular, is seen as possessing greater medium- to long-term earnings visibility thanks to rising demand from hyperscale customers for customized AI chips. Analysts interpret this as evidence that the center of gravity in AI infrastructure competition is shifting from large-scale purchases of general-purpose GPUs toward the construction of purpose-built AI systems.

Google Bets on TPU Commercialization

While AMD and Broadcom are accelerating their Nvidia challenge through packaging expansion and customized ASIC strategies respectively, Google is mounting a direct assault on Nvidia’s GPU ecosystem through vertical integration of proprietary AI semiconductors and cloud infrastructure. Last month, Google unveiled a new AI inference chip along with a next-generation Tensor Processing Unit (TPU) designed for AI training. The company has already signed an agreement providing Anthropic with access to approximately one million TPUs and has also established a partnership with Meta Platforms.

The market views the initiative as Google’s largest attempt yet to commercialize TPUs for external customers. Until now, Google had primarily utilized TPUs internally within its own data centers and for select strategic partners. Blackstone’s financial firepower also represents a key variable. The private equity giant has emerged as one of Wall Street’s most aggressive investors in AI infrastructure. Blackstone acquired data-center operator QTS Realty Trust in 2021 and signed a deal to acquire data-center company AirTrunk in 2024. According to The Wall Street Journal, the two companies plan to establish a U.S.-based joint venture in which Blackstone will hold the controlling stake, with Blackstone committing $5 billion in initial capital.

The venture’s primary competitor is CoreWeave, a so-called “neocloud” provider that has secured massive quantities of Nvidia AI chips and built AI cloud infrastructure for enterprise customers. As demand for AI model training and inference computing capacity explodes, many major AI companies have become heavily reliant on CoreWeave’s infrastructure. Google and Blackstone appear poised to challenge Nvidia’s dominance in AI computing by offering TPU-based cloud services in place of Nvidia GPUs while simultaneously attempting to capture CoreWeave’s customer base.