Global Semiconductor: The Resurgence of CPUs? Beneficiaries of a $2.23 Trillion Market Size

Global Semis: The CPU Renaissance? Beneficiaries of a $223bn The shift from gen AI paradigm from 1.0 (chatbot) to 2.0 (agent) greatly increasesserver CPU demand.As discussed in SoftBank, Arm: From GenAI to Agentic AI; Initiatingwith Outperform Ratings, Agentic AI involves heavily autonomous task orchestration andexecution, which boosts the CPU workload vs. GPU. With the shift from chatbot to agentic David Dai, CFA+852 2918 5704david.dai@bernsteinsg.com Stacy A. Rasgon, Ph.D.+1 213 559 5917stacy.rasgon@bernsteinsg.com We raise server CPU TAM to $223bn ($137bn in 2030 in the base case, 6x of the2025 TAM of $37bn.This assumes $3.5tn of AI data center capex, and 1:1 CPU:GPUpairing ratio for inference. An alternative approach with 120mn CPU cores/GW yieldssimilar TAM. Our previous forecast of $137bn is now the bear case (assuming $3tn AIcapex, 1:2 CPU:GPU), while the upside now sits at $330bn ($4tn AI capex, 1.5:1 CPU:GPU). Qingyuan Lin, Ph.D.+852 2123 2654qingyuan.lin@bernsteinsg.com Mark Li+852 2123 2645mark.li@bernsteinsg.com Raising Arm PT to $500, as Arm is the structural beneficiary of the renaissanceof CPUs for agentic AI.Arm architecture is suitable for agentic AI workload given itsunparalleled power efficiency. In addition, Arm is shifting from just IP provider to CPUmaker, aiming to capture $15bn revenue by CY2030, but we now forecast $22bn as werevise CPU TAM to $223bn in 2030 (from $137bn). Arm’s 2030 EPS (FYE31) is now liftedto $11.79 ($9.83 prior). Based on 42x P/E (40x prior), welift Arm’s PT to $500 (21% Juho Hwang+852 2123 2632juho.hwang@bernsteinsg.com Jack Lin+852 2123 2683jack.lin@bernsteinsg.com Updating numbers, raising AMD and INTC PTs.Both companies should benefit fromstronger (and more sustained) server demand, though AMD’s products remain superiorfor now (and we believe they will continue their share gain trajectory). Our existing AMDmodel was already consistent with a stronger server CPU environment and estimates movemarginally, however we are now bringing our INTC model inline with those assumptionsand are raising estimates more materially; we also roll valuation horizon forward for both Carmine Milano, CFA+44 20 7762 1857carmine.milano@bernsteinsg.com Alrick Shaw+1 917 344 8454alrick.shaw@bernsteinsg.com Arpad von Nemes+1 917 344 8461arpad.vonnemes@bernsteinsg.com Hygon will benefit from strong x86 CPU demand and gain share in China.We expectChina to outpace global x86 growth from 2028 onward, with the easing of advanced-nodesupply constraints in China and accelerating AI investment unlocking CPU potential. Weexpect Hygon to steadily expand its share of China’s x86 server CPU market, exceeding35% by 2030, as it increasingly penetrates into CSPs, beyond its traditional customer baseof government and SOEs, supported by improved interoperability with domestic AI chips Francis Ma+852 2123 2626francis.ma@bernsteinsg.com What could go wrong?We’re still assessing if foundry/memory capacity will be sufficientto support the CPU growth. Additionally, the value of GPU/accelerator embeds the value ofHBM & the markup charged by NVDA, etc. now but the high cost of memory including HBMmay prompt hyperscalers to source directly from memory suppliers. Our projection is based BERNSTEIN TICKER TABLE PRICE TARGET CHANGE / ESTIMATE CHANGE IN BOLD Source: Bloomberg, Bernstein estimates and analysis. INVESTMENT IMPLICATIONS We rateSoftBank(PT=¥11,200) andArm(PT=$500.00)Outperform. AMD (Outperform, $600.00):Expectations remain high, but exposure to AI demand driving both a CPU and GPU story can INTC (Market-Perform, $100.00):Server strength is helping the company get back on their feet, and narrative/headlines mayfuel the vibe for now. NVDA (Outperform, $315.00):The datacenter opportunity is enormous, and still early. We rateHygon(PT=CNY 450.00)Outperform: we raise our PT from CNY 280 to CNY 450, based on 2028E EPS of CNY 6.30(+32% vs Bern. Old) and 71x P/E (Previously was Outperform, 80x P/E based on 2027E EPS CNY 3.48, PT CNY 280). DETAILS We raise our 2030 server CPU TAM to $223bn on the back of higher AI investments and strong CPU:GPU pairing ratio, and ourprior forecast of $137bn is now moved to the bear case (Exhibit 1). On the back of that, we raise PTs for Arm, SoftBank, AMD,Intel, Hygon. Our Server CPU Industry model can be downloaded here: Server CPU Industry Model. Updated Arm financial model RENAISSANCE OF CPU IN THE AGENTIC AI PARADIGM Since the rise of LLM, GPU/ASIC accelerators have been the core of AI computing. While training clusters once required a dense4:1 ratio to handle heavy data-loading, the focus shifted toward eliminating the 'CPU tax' that plagued high-scale inferencing. Incustom inference-optimized deployments like Google’s TPU v6e and Meta’s Grand Teton, the GPU-to-CPU socket ratio moved Agentic AI is pushing the CPU back to center stage (Exhibit 2) because AI systems are no longer just running a model once andreturning an answer. The GPU still perf