UBS: CPU-Demand Chips Poised for Further Gains

UBS raises AMD and ARM price targets as agentic AI supercharges server CPU demand, projecting the market to expand fivefold to $170 billion by 2030.

Lead

What Happened

UBS analyst Timothy Arcuri lifted his 12-month price target on AMD to $670 from $455 and on ARM to $470 from $260, maintaining Buy ratings on both. The upgrades arrived alongside materially higher server CPU revenue estimates: AMD's projections for 2027 and 2028 were revised upward to $23 billion and $29 billion, respectively, from prior estimates of $21 billion and $27 billion, with the 2026 baseline held at $16 billion. AMD's total company revenue target for 2030 was simultaneously raised to $50 billion from $41 billion.

Across the broader chips sector, UBS forecasts aggregate CPU revenues reaching $75 billion in 2026 and climbing to $96 billion in 2027, even as the personal computer market remains sluggish. The divergence underscores how demand originating from hyperscaler data centers — rather than consumer hardware — has become the dominant growth engine for the CPU segment.

Market Reaction

AMD shares advanced in Tuesday trading, reflecting the revised estimates and raised target. Arm Holdings shares initially pulled back despite the price-target increase, a move analysts attributed to broader profit-taking in the AI chip sector after a sustained run-up. The divergence illustrates how differently investors are positioning around the x86 and ARM architectures within the same constructive macro thesis.

Strategic Context

The central argument behind UBS's upgrade rests on the emergence of agentic AI — autonomous AI systems that reason, plan, and execute multi-step tasks — and its distinct computational demands. Unlike pure inference workloads, which lean heavily on graphics processing units, agentic architectures require orchestration layers, memory management, and persistent process handling. These functions map naturally onto CPUs.

For AMD, UBS argues the company is positioned to capture an outsized share because of its advantage in core count, multithreading throughput, and the deep software ecosystem around the x86 instruction set. Legacy enterprise software, middleware, and increasingly agentic AI orchestration frameworks are all calibrated for x86, giving AMD a structural incumbent advantage in standalone CPU racks.

For Arm Holdings, the thesis centers on latency and power efficiency. ARM-based architectures suit hyperscaler needs where energy-per-operation and per-rack heat budgets constrain deployment density. UBS assumes a 60/40 split favoring x86 over ARM in the standalone CPU server segment, while projecting that ARM-based designs will capture roughly 70% of the estimated 20-million-unit total addressable market for head-node servers by 2030.

The firm now models the server CPU total addressable market expanding from a $30 billion base in 2025 to approximately $170 billion by 2030 — roughly a fivefold increase — driven by both volume growth and rising average selling prices as workload complexity escalates.

AI and Technology Angle

The shift from GPU-centric AI infrastructure to hybrid GPU-CPU architectures reflects a maturation in how enterprises and hyperscalers are building AI systems. Early generative AI deployments were dominated by model training on GPU clusters. As attention moves toward inference, orchestration, and multi-agent pipelines, CPU workloads are growing in proportion to the overall AI compute mix.

Hyperscalers including the largest cloud platforms are ordering standalone CPU racks at a pace that is outrunning PC-cycle comparables by a wide margin. CPU revenues are climbing not because desktop demand has recovered but because data center demand has fundamentally repriced the segment. This structural shift makes CPU exposure less cyclical than it has historically appeared, lending credibility to multiyear revenue trajectory forecasts.

AMD's server CPU share gains are also coming directly at Intel's expense. Multiple independent server-market reports published through mid-2026 show AMD and ARM-architecture designs extending share gains, a trend UBS sees continuing as AMD's product cadence and ARM's design-win pipeline in hyperscaler custom silicon accelerate.

Semiconductor Sector Backdrop

UBS's chip-specific upgrades are embedded in a broader bullish semiconductor framework. The bank's sector team forecasts aggregate semiconductor industry sell-in of $1.62 trillion in 2026, up 118% year-over-year, before climbing a further 46% to $2.38 trillion in 2027. Global AI capital expenditure is projected to reach $500 billion in 2026 alone, reshaping procurement volumes across logic, memory, and connectivity components.

Memory chips are also seeing upward estimate revisions, with the firm projecting chip revenues in high-bandwidth memory, DDR5, LPDDR5, and NAND flash segments to nearly double between 2026 and 2027 as AI infrastructure build-outs place escalating demands on data throughput.

What Comes Next

The near-term catalyst calendar for AMD includes its next earnings report, where hyperscaler CPU order commentary and guidance on standalone rack deployments will be scrutinized closely against UBS's elevated estimates. For ARM, the critical variable is the pace of hyperscaler custom chip design wins — each new design-in at a major cloud provider extends the royalty revenue duration curve.

Intel's response to share losses remains a key swing factor. Any acceleration in Intel's product roadmap, or evidence that hyperscalers are diversifying back toward Intel x86 for cost or software reasons, would pressure both AMD and ARM share gains. Conversely, further delays in Intel's manufacturing transition could accelerate AMD's capture of the expanded TAM.

Outlook

The UBS thesis frames the server CPU market as entering a structural growth cycle driven by agentic AI's distinct compute requirements, not a short-term inventory restock. With the TAM projected to expand nearly fivefold by 2030, AMD and ARM are positioned as primary beneficiaries — AMD through x86 software compatibility and thread density, ARM through efficiency advantages at hyperscale. The central risk is execution: whether AMD can sustain product cadence and whether ARM design wins convert to volume at the pace the model assumes. As of June 24, 2026, the market appears to be pricing in growing confidence that both will.