For much of the artificial intelligence boom, investors have heard one simple story.
Graphics processing units were at the center of it all.
That thesis made Nvidia (NVDA) a must-have in the stock market. It also drove investors to Advanced Micro Devices (AMD), Broadcom (AVGO), Marvell Technology (MRVL), and other companies involved in the data-center buildout.
But Bank of America (BAC) argues the next chapter of AI could give a forgotten corner of the semiconductor market a major boost.
Central processing units, or CPUs, could be significantly more crucial as AI evolves beyond basic chatbot-style responses toward agentic systems that plan, reason, retrieve information, summon software tools and accomplish multi-step tasks.
The move has prompted a significant call from Bank of America analyst Vivek Arya.
The firm raised its prediction for the server CPU market in 2030 to almost $170 billion from $125 billion in a note. It also upgradedIntel (INTC) to buy from underperform, raising its price target to $135 from $96.
That’s a rare piece of good news for Intel.
The chipmaker has spent years battling manufacturing delays, market-share erosion, and investor pessimism about its AI position.
Bank of America now argues that agentic AI might provide Intel a fresh path back into the discourse.
“We view the emergence of agentic AI as a powerful demand accelerant,” Bank of America analyst Vivek Arya wrote.
Bank of America says agentic AI changes the CPU story
There’s a reason the first stage of the AI trade was all about GPUs.
Training large models requires huge amounts of parallel computing resources. This thrust Nvidia’s accelerators into the most essential hardware in the market, with CPUs moved into a supporting role.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
CPUs still played a role, but mostly in the background. They took care of data prep, scheduling, tokenization, storage access and orchestration to keep the GPUs active.
Agentic AI shifts that equilibrium.
These systems don’t just take one cue and spit out one answer. They are capable of decomposing a task into subtasks, maintaining memory, calling application programming interfaces, accessing databases, retrieving documents, using software tools, and assessing intermediate outputs before arriving at a decision.
This makes the task more sequential, more sensitive to latency, and more reliant on input-output management.
Those are the places where CPUs matter.
Bank of America isn’t saying that GPUs will be replaced by CPUs. That would completely misunderstand the point. GPUs continue to be the workhorse for large-model inference, dense decoding, reasoning, and high-throughput compute.
The firm contends that agentic AI will grow the market.
Put another way, the next AI buildout could require more GPUs and more CPUs.
Related: Intel executive drops unexpected message on explosive chip demand
Bank of America now forecasts server CPU revenue growing from roughly $35 billion in 2025 to over $170 billion in 2030. That’s a 37% compound annual growth rate.
The business has divided that 2030 market into three buckets.
Traditional cloud and infrastructure CPUs are roughly $30 billion, and $70 billion is for AI computing and head node CPUs. Another $70 billion is for AI agentic standalone node CPUs.
That last category is the important one.
If agentic AI scales, data centers may need CPU-heavy systems to sit alongside GPU racks and manage orchestration, retrieval-augmented generation, tool execution, vector databases, memory-heavy workflows, and smaller model inference.
That would take the AI hardware story beyond the GPU explosion that propelled the first phase of the market.
Intel gets rare upgrade as AMD remains top CPU pick
The most startling thing in the note is the Intel upgrade.
Wall Street has not been kind to Intel as an AI winner. The dominant story in the AI accelerator space is Nvidia. AMD has taken market share from server CPUs. Taiwan Semiconductor Manufacturing (TSM) is today’s most valuable advanced manufacturing partner in the world.
Intel has been caught trying to prove it can solve its own execution, while establishing an external foundry business.
Now, Bank of America sees a superior setup.
The company thinks Intel can leverage a bigger server CPU opportunity and industry constraints in leading-edge wafers and advanced packaging.
Bank of America now predicts Intel will earn more than $6 a share by 2030, up from its previous range of $3 to $4. Intel server CPU sales will hit more than $40 billion by 2030, or a quarter of the firm’s forecast $170 billion market.
That doesn’t mean Intel’s recovery is a sure bet.
Execution is the major problem. Intel still has the challenge of delivering competitive server chips, improving manufacturing, ramping future nodes, and proving that outside customers would trust its foundry business.
But the overhaul changes the company’s image.
This is positioning Intel to be a potential beneficiary of the next generation of AI infrastructure, not a legacy chipmaker battling old conflicts.
And you could still make the case for AMD as the cleaner CPU winner.
Bank of America reaffirmed its buy rating on AMD and boosted its price goal to $560 from $500. AMD is the best-positioned CPU vendor, the business said, because of its incumbency, product pipeline, and performance prospects.
The note also highlights AMD’s upcoming EPYC Venice CPU, which is projected to top out at 256 cores. That is compared to 192 cores for Intel’s anticipated Diamond Rapids and 88 cores for Nvidia’s Vera CPU.
Key takeaways from Bank of America’s AI CPU call
- Bank of America raised its 2030 server CPU market forecast to more than $170 billion from $125 billion.
- The firm says agentic AI could create new CPU demand for orchestration, memory handling, tool use, and workflow management.
- Intel was double-upgraded to buy from underperform, with a $135 price objective.
- AMD remains Bank of America’s top CPU pick, with its price objective raised to $560 from $500.
- Nvidia remains the firm’s top overall AI sector pick because of its full-stack AI leadership.
- Arm is expected to gain CPU share, but Bank of America kept a neutral rating.
- Qualcomm retains an underperform rating because of tough competition in data center CPUs and accelerators.
That core-count advantage could be important in agentic AI, because more cores can assist in managing more simultaneous subtasks within the same power envelope.
CHENG YU-CHEN / Getty Images
Nvidia and Arm also gain from the CPU revival
The CPU comeback doesn’t take Nvidia out of first place.
Bank of America named Nvidia its top selection in the AI sector, maintaining its $350 price target. The explanation is simple: The next step of AI still needs tremendous accelerator power, networking, memory, and software.
Nvidia’s edge is because it has more of the system.
The company isn’t only selling GPUs. It’s designing whole rack-scale platforms that incorporate GPUs, CPUs, networking, and software into one AI infrastructure stack.
That matters because agentic AI isn’t just a CPU story, or a GPU one. It’s a systems tale.
As AI workloads get more complicated, bottlenecks can shift from basic computing to scheduling, memory management, networking, data migration, and tool latency. Companies that control more of the system than companies that sell one isolated component may be better off.
That’s why even in a report that argues CPUs are becoming more important, Nvidia still matters.
Arm might be a big winner, too. Bank of America anticipates Arm-based CPUs to take market share through 2030, aided by merchant chips from Nvidia, Arm, and Qualcomm and specialized cloud chips from Amazon, Google, and Microsoft.
Arm’s total server CPU value share might reach almost 50% by 2030, according to the company.
But valuation limits the call. Bank of America boosted its price target on Arm to $335 from $245 but maintained a neutral rating on the stock, saying the opportunity is fully priced in.
Qualcomm faces a tougher path ahead. The firm anticipates Qualcomm to discuss new AI CPU plans at its June 24 AI Day in New York, but maintains an underperform rating due to entrenched competition from Nvidia, AMD, Intel, and Arm.
For investors, the message is clear. The AI trade is maturing.
The first phase benefited enterprises most closely related to the shortages of GPUs and the need for AI training. The next step might reward corporations that let data centers operate more complicated, agentic AI systems at scale.
This provides Intel something it hasn’t had for a long time: a genuine AI comeback story.
That doesn’t make the turnaround simple, but it does make the stock harder to overlook.
Related: Bank of America makes major reset to Intel stock price target