Bank of America's Agentic AI Call Reprices Chip Architecture

TL;DR — Arm jumped 11% after Bank of America hiked its price target, forecasting a fivefold expansion in the server CPU market driven by a new wave of AI workflows. The upgrade coincided with filings showing Altimeter Capital dumped Alphabet to take a $260 million contrarian stake in the chip designer. The takeaway is the shifting infrastructure narrative: graphics processors train the models, but traditional processors are required to actually run them.

The move

Arm traded up 11.27% on Friday, moving from a previous close of $342.23 to settle at $380.81. While daily volume data remained temporarily unrecorded across major feeds, the price action marked a decisive continuation of a longer, aggressive arc. The stock has more than doubled since the spring, repeatedly shrugging off valuation concerns—currently trading at 156 times forward earnings—as institutional capital crowds into the purest plays on computing infrastructure.

What drove it

Two distinct signals hit the tape. First, Bank of America issued a profound upgrade, raising its price target to $335 from $245. The rationale hinged on "agentic AI"—artificial intelligence that can actively execute plans, coordinate software, and schedule tasks rather than just generating text. Wall Street expects this shift to drastically expand the market for server central processing units, forecasting a $170 billion total addressable market by 2030 (per Bank of America: driven by a forthcoming boom in agentic AI applications).

The second signal was a matter of smart money rotation. First-quarter 13F filings revealed that Brad Gerstner's Altimeter Capital completely exited a half-million share position in Alphabet, redeploying the capital to buy $260 million worth of Arm. The subtext is clear: some funds are choosing to sell the mega-cap tech companies bearing the massive capital expenditure costs of data centers, buying instead the companies selling the intellectual property those data centers rely on. Arm validated that demand by confirming it has secured over $2 billion in pipeline orders for its new AI-specific compute processors.

The bigger picture

To understand Arm, you have to understand the bottleneck of the modern data center. For the past two years, the market has cared entirely about graphics processing units. Those are the massive, power-hungry chips that do the heavy mathematical lifting required to train AI models. But as AI moves from training to inference—from learning how to think to actually doing daily work—the workflow changes. Complex AI agents require logic, task scheduling, and sequence execution. That work falls on traditional central processing units.

Arm does not build these chips. It writes the fundamental blueprints for them, then collects a licensing fee upfront and a royalty on every unit shipped. Because the industry is rapidly building custom silicon to escape the high margins of buying off-the-shelf accelerators—a dynamic outlined in A $7 Billion Equity Raise Exposes the True Cost of AI Servers—Arm's intellectual property becomes the default foundation. They operate an infrastructure tollbooth. When cloud providers design their own internal chips, they almost invariably use Arm architecture. The cycle is moving from a frantic scramble for pure graphics power into a more mature phase where energy-efficient, customized compute architecture dictates who wins the cloud war.

Macro overlay

A broader geopolitical relief rally provided the perfect backdrop for a high-multiple tech breakout. Reports of a potential near-term US-Iran peace agreement sent crude oil futures down 3% and stripped immediate fear from the broader market. The Volatility Index dropped 9% to 17.68, while the Nasdaq 100 drifted up 0.64%. When geopolitical risk premiums compress and energy costs look stable, institutional investors immediately reach further out on the risk curve, funneling capital right back into high-growth technology names.

What to watch

• Next quarter's royalty mix: Watch the specific breakdown between upfront licensing fees and per-chip royalties. A spike in licenses means companies are designing new custom chips; a steady climb in royalties means those custom chips are finally shipping in volume.
• Qualcomm's data center rollout: Qualcomm is leveraging architecture to build its own server chips. Any commentary on their enterprise adoption rates will signal how quickly Arm-based servers are actually taking market share from legacy x86 architecture.
• Hyperscaler custom silicon announcements: Keep an eye on the major cloud providers (Amazon, Google, Microsoft). Every time one announces a new internal data center chip for AI inference, it validates Arm's central position as the base layer of custom compute.