Bernstein Raises 2030 Server CPU Market Forecast to $223 Billion
Alina Collins
Bernstein has raised its 2030 server CPU market forecast from $137 billion to $223 billion, driven by AI's shift from chatbots to Agentic AI — a change that is repricing the CPU's role inside data centers.
Why are CPUs suddenly worth more?
The key shift: the CPU-to-GPU ratio in AI data centers is rising from 1:4 or 1:8 to 1:1 or higher.
This means → GPUs used to carry the load almost alone; now CPU volume needs to match GPU volume.
The driver is Agentic AI — AI that plans, calls tools, reasons in steps, and acts, rather than just answering a single prompt. These workflows demand heavy CPU involvement for scheduling, data feeding, and memory management.
How does the $223 billion number break down?
Base-case assumptions: $3.5 trillion in AI data-center capex by 2030; a 1:1 CPU-to-GPU ratio in inference workloads.
Result: the server CPU market grows from $37.3 billion in 2025 to $223.4 billion in 2030 — roughly 6× today's level.
Bernstein maps three scenarios: bear case $137 billion (capex $3 trillion, ratio 1:2), base case $223 billion, bull case $330 billion (capex $4 trillion, ratio 1.5:1).
What exactly does the CPU do in Agentic AI?
An Agentic AI workflow is not "run one model, return one answer." It is retrieval → planning → tool calls → intermediate reasoning → model re-invocation → action execution — a multi-step chain.
GPUs still handle dense computation, but CPUs manage data feeding, task scheduling, tool-call coordination, and memory management — and keep accelerators from sitting idle.
In plain terms = the GPU does the heavy lifting; the CPU is the foreman on site. The more complex the job, the more you need the foreman.
What does the 2026 hardware roadmap look like?
AMD Venice: 1 CPU to 4 MI455X GPUs per compute tray.
Nvidia Vera: 1 CPU to 2 Rubin GPUs per superchip.
Google TPU7x: 1 CPU to 4 TPU chips per expansion unit.
This reflects a shared direction: all three vendors are moving toward a more balanced CPU-to-accelerator ratio. The CPU is no longer a small support part bolted onto a rack of accelerators.
Who benefits most?
Arm: target price raised to $500, rated outperform. 2030 revenue forecast lifted from $15 billion to $22 billion; EPS from $9.83 to $11.79. Bernstein calls Arm "the structural beneficiary of the CPU renaissance."
AMD: target price raised to $600, rated outperform.
Intel: target price raised to $100, rated market-perform.
Hygon (海光): expected to hold over 35% of China's x86 server CPU market by 2030; target price raised from ¥280 to ¥450. SoftBank's target rises to ¥11,200 on its Arm stake.
What could go wrong?
Bernstein acknowledges it is still assessing whether foundry and memory capacity can keep pace with CPU growth.
A second risk: if cloud vendors, facing rising costs for high-bandwidth memory, bypass CPU/accelerator packages and buy directly from memory suppliers, Bernstein's valuation based on CPU and accelerator pricing faces downside pressure.
This means → the forecast is sharply higher, but supply-chain bottlenecks and shifts in how customers buy could cut the actual number.
Content is for reference only, not financial advice.