Advanced Micro Devices (AMD) CEO Lisa Su stated on Friday that the company, in collaboration with its partners in Taiwan, China, is significantly ramping up production capacity to meet unexpectedly strong global demand for data center CPUs and GPUs. Leveraging the wave of AI inference and Agentic AI, which has driven CPU and GPU demand beyond expectations, AMD is strategically positioning itself not merely as a challenger to Nvidia in GPUs but as a comprehensive supplier of AI data center infrastructure. This includes CPUs, GPUs, high-performance networking, and rack-scale systems, with plans to deepen its integration with TSMC's most advanced manufacturing processes and packaging technologies.
Following a visit to China, Su noted she met with AMD's largest clients globally and in China to ensure that its long-term partner, TSMC, the world's leading contract chipmaker, can support a substantial increase in production of data center central processing units (CPUs) and AI GPU computing clusters. AMD also announced a major investment exceeding $10 billion in Taiwan's AI ecosystem to strengthen strategic partnerships and enhance capacity for advanced AI chip manufacturing and server cluster assembly.
Su emphasized that AMD is working closely with partners like TSMC to expedite production expansion in response to the surging demand for CPUs and GPUs fueled by AI inference and Agentic AI. From an engineering perspective, Agentic AI systems rely not only on GPUs for training large models but also on CPUs for tasks such as scheduling, data handling, networking, storage access, security, multi-agent orchestration, and inference service management. AMD's push to expand capacity in Taiwan is not just about producing more chips but ensuring synchronized scaling across its supply chain for EPYC CPUs, Instinct GPUs, advanced packaging, substrates, assembly, and rack-scale systems.
Amid the booming demand for data center CPUs, Wall Street sentiment is increasingly bullish on the major x86 CPU giants Intel and AMD, as well as Arm Holdings. Year-to-date, these three CPU leaders have seen significant stock price gains, with Intel up 222% so far in 2026. According to TipRanks, the highest analyst price target for AMD is $625, implying about 40% upside potential. AMD's stock has surged 110% this year, ranking it among the top AI-performing stocks globally.
From 2nm Venice to a $10 billion investment, AMD is aggressively securing capacity in the AI computing infrastructure race. Taiwan plays a pivotal role in the global AI supply chain for North American tech giants like Nvidia and Apple, underpinned by TSMC, which is the core supplier for AMD, Nvidia, Apple, Broadcom, Qualcomm, and others. Su stated, "Overall CPU demand is significantly higher than anyone predicted a year ago," noting that the CPU market is supply-constrained. She added that AMD is rapidly increasing capacity, with supply expected to grow each quarter this year and plans for substantially more output in 2027 and beyond.
This robust demand growth is driven by AI inference and Agentic AI. As enterprises increasingly adopt autonomous AI systems, CPUs have become a critical focus, even more so than GPUs, with computing needs extending beyond GPUs used for training large AI models. Su mentioned that China accounts for approximately 20% of AMD's revenue, remains a very important market, and the company will continue to work closely with Chinese clients while complying with U.S. export controls that restrict shipments of some high-end AI chips to China.
AMD's investment in Taiwan is not merely about placing more orders with TSMC but involves coordinating across advanced manufacturing, packaging, substrates, ODM/server, and rack-scale system ecosystems to expand strategic capacity. This reflects that the AI computing competition has evolved from individual GPUs/CPUs to a full-chain race encompassing advanced process nodes, Chiplet packaging, substrate technology, and rack-scale AI systems.
AMD also announced it has begun volume production ramp-up of its Venice CPU using TSMC's 2nm-class process technology. According to AMD, EPYC "Venice" is the sixth-generation EPYC data center CPU, the world's first 2nm-class data center CPU, and the first 2nm-class HPC/data center CPU to enter volume production ramp-up, marking the move toward large-scale commercial production of 2nm data center CPUs.
Su remarked, "We made two big bets. The first was on TSMC, which has proven to be a fantastic bet for us." The second bet, she noted, is that increasingly complex silicon manufacturing will require chips to be broken into smaller parts and integrated via advanced packaging, a trend widely adopted across the semiconductor industry as Chiplet technology.
Wall Street analysts view AMD not as simply replicating Nvidia's GPU dominance but as leveraging its entire AI infrastructure ecosystem—including EPYC CPUs, Instinct GPUs, Helios/rack-scale platforms, 2nm Venice, and advanced packaging—to redefine its valuation in the AI data center industry amid the rise of AI inference and Agentic AI. The narrative around AI computing infrastructure is expanding from "GPU dominance" to a full-stack reassessment involving AI GPUs/ASICs, CPUs, HBM/DRAM/NAND memory chips, and high-speed data center interconnect systems.
Against the backdrop of surging data center CPU demand, Citigroup this week significantly raised its 12-month price targets for Intel and AMD, also increasing its outlook for the data center CPU and overall CPU market size. At JPMorgan's 54th Annual Global Technology, Media, and Communications Conference, Intel CEO Pat Gelsinger stated that Intel 18A process technology (below 2nm) is supporting Panther Lake production, with yields improving about 7% monthly, exceeding internal expectations. Gelsinger also highlighted that as AI computing infrastructure shifts focus from training to inference, CPUs are becoming increasingly vital, with CPU-to-GPU ratios moving from 1:8 toward 1:1, potentially even reaching 4:1. Intel is also actively pursuing ASIC business, offering customized AI CPU or AI GPU solutions.
The transition from AI training to inference and Agentic AI significantly elevates the strategic importance of CPUs in AI data centers. While GPUs handle large-scale matrix computations, CPUs manage scheduling, I/O, memory, task orchestration, security, database access, networking, and multi-agent workflows. As AI applications evolve from single inferences to continuously running software labor, CPU demand shifts from traditional server refresh cycles to AI infrastructure expansion cycles.
Citigroup's latest model aligns with this view, projecting the total addressable market (TAM) for data center server CPUs to grow from $29.3 billion in 2025 to $131.5 billion in 2030, a compound annual growth rate of approximately 35%. According to TipRanks, the average analyst price target for AMD is around $460, with the highest target at $625 from Baird analyst Tristan Gerra, who recently raised his target from $300 to $625 while maintaining an Outperform rating. Gerra's bullish thesis is based on unprecedented CPU/GPU demand driven by Agentic AI and a revaluation of AMD's data center platform. He noted that AMD's opportunities in AI server CPUs, GPUs, and accelerated computing platforms are being repriced, with the market recognizing AMD as a key beneficiary in the AI computing产业链 as a provider of integrated CPU+GPU+platform solutions, driven by AI inference, Agentic AI, EPYC CPUs, Instinct GPUs, and next-generation AI infrastructure expansion.
Comments