For three years, the artificial intelligence trade has been an Nvidia story, built on graphics chips.
Bank of America now argues the next leg could run through the central processing unit (CPU), the general-purpose chip that has powered servers for decades.
In a note published Thursday, analyst Vivek Arya raised his price objective on Advanced Micro Devices Inc. (NASDAQ:AMD) from $620 to $720 and kept a Buy rating.
The question behind the call is whether AI agents turn CPUs from a supporting act into a growth engine of their own.
Why AI Agents Need More CPUs
Arya hosted Ben Bajarin, CEO and principal analyst of Creative Strategies, for a call on the server CPU market.
The argument works in steps.
First, a graphics processing unit, or GPU, does the “thinking” by generating tokens, the chunks of text an AI model produces.
Second, an AI agent has to act on that thinking. It calls tools, runs code, retrieves documents and queries databases.
Those tasks run on the CPU.
As a result, a slow CPU leaves an expensive GPU waiting, and the user waiting too.
“Agentic AI makes CPU/GPU demand symbiotic,” Arya said.
AMD And A Server CPU Market Seen Reaching $211 Billion
Bank of America models server CPU sales climbing from $61.4 billion in 2026 to $210.6 billion in 2030. AI-related chips would account for 86% of that total, up from 69% this year.
| Year | Server CPU sales | AI-related share |
|---|---|---|
| 2026 | $61.4 billion | 69% |
| 2027 | $104.6 billion | 80% |
| 2028 | $138.4 billion | 83% |
| 2029 | $170.9 billion | 84% |
| 2030 | $210.6 billion | 86% |
The bank sees AI CPU units rising from 16 million to 53 million over the same period. That would put them at roughly 1.5 times accelerator units by 2030, versus about 0.7 times today.
In addition, prices could rise with the mix.
Bank of America expects the average server CPU to sell for about $2,600 in 2030, up from roughly $1,600 in 2026.
Anthropic Just Put Real Money Behind the Thesis
The argument already has a real-world test.
Akamai Technologies Inc. (NASDAQ:AKAM) announced Thursday an $11.6 billion, seven-year agreement with Anthropic specifically aimed at supporting growing CPU workloads.
The relationship could eventually expand by another $9 billion, taking the potential commitment toward $20 billion.
That is exactly the workload shift BofA is describing.
The scarce resource is no longer simply how many GPU calculations a data center can produce.
It is how quickly the rest of the system can keep those expensive GPUs productive.
“We favor tasks/core, completion latency, concurrency and perf/watt over standalone benchmarks," Arya said.
In simpler terms, investors may need to stop counting processor cores and start measuring how quickly entire AI systems finish useful work.
Why AMD Is Bank of America’s Top CPU Pick
Chip designers are taking two different paths. Nvidia Corp. (NASDAQ:NVDA) favors fewer, faster cores that hand results back to the GPU quickly.
Meanwhile, Arm Holdings plc (NASDAQ:ARM) is planning a chip with more cores to run more tasks at once.
According to Arya, AMD offers the broadest portfolio across both approaches.
For this reason, he raised his 2027 and 2028 estimates by 2% to 3% and lifted his valuation multiple on AMD to 30 times 2028 earnings, from 27 times.
The new target implies about 14% upside from Thursday’s close of $629.26.
Nvidia, however, remains Arya’s top pick across the whole semiconductor sector. His $350 price objective sits about 56% above its $224.58 close.
He also reiterated a Buy on Intel Corp. (NASDAQ:INTC), with a $145 target, about 14% above $127.39, citing potential long-term foundry capacity for outside customers.
Photo: Shutterstock
Login to comment